Enhanced Human Computer Interaction using hand gesture analysis on GPU

Mar 1, 20171 like209 views

This poster represent very active research topic in human computer interaction (HCI) as automatic hand gesture recognition using nvidia GPU. In this work neural network based video gesture are processed and recognize the finger counts. Due to real time requirement algorithm need to optimize and computationally efficient. We implemented the MATLAB code, it perform slow when neural network processing started. Implementing them in a parallel programming model such as GPU-CUDA would provide the necessary gain in processing speed. Algorithmic result validation is done using standard video data set and recognition rate is calculated. A performance improvement of 15x speed is achieved which is faster than Intel quad core processor.

Abstract
This poster represent very active research topic in human
computer interaction (HCI) as automatic hand gesture recognition
using nvidia GPU. In this work neural network based video gesture
are processed and recognize the finger counts. Due to real time
requirement algorithm need to optimize and computationally
efficient. We implemented the MATLAB code, it perform slow when
neural network processing started. Implementing them in a parallel
programming model such as GPU-CUDA would provide the
necessary gain in processing speed. Algorithmic result validation is
done using standard video data set and recognition rate is
calculated. A performance improvement of 15x speed is achieved
which is faster than Intel quad core processor.
Introduction
The goal of this project is to develop a program
implementing gesture recognition. At any time, a user can exhibit
his hand doing a specific gesture in front of a video camera linked
to a computer.
The program has to collect pictures of this gesture, analyze it
and to identify the sign. In order to lighten the project, it has been
decided that the identification would consist in counting the number
of fingers that are shown by the user in the input picture.
Database And MATLAB Results
CUDA Approach
1. Camera capture video – process frame by frame image
2. Extract features from the image using various segmentation techniques
implemented in MATLAB, CUDA. Output features then written to a text file.
3. Use the matrix obtained in step 2 to train the neural network(on CPU
using MATLAB).
4. Once the network is trained , we test it on GPU.
5. GPU implementation :-
a. Load the test image into GPU global memory
b. Process it as in step 2, output will be matrix of size (262144 x1).
c. Pad the matrix with 0's - output will be a matrix of size (262144 x16) -
Inputs matrix.
d. This is all done on the GPU. Now pass the neural network weights
matrix, to the GPU.
e. Multiply neuralnetwork_weights matrix by Inputs matrix - which
will give a 16 x 16 matrix. Extract the first five values from it. The
index of the value which has maximum value + 1 gives us the count.
NN Training Results
GPU (GeForce GT 525M)
References
[1] Malima, A.; Ozgur,E.; Cetin,M.; [2006] “A Fast Algorithm for Vision-Based
Hand Gesture Recognition for Robot Control”, Signal Processing and
Communications Applications, IEEE
[2] A Neural Network on GPU,
http://www.codeproject.com/KB/graphics/GPUNN.aspx
[3] NVIDIA Corporation, Compute Unified Device Architecture (CUDA),
http://developer.nvidia.com/object/cuda.html.
Enhanced Human Computer Interaction using hand gesture analysis on GPU
Mahesh Khadtare1, Pragati Dharmale2
1 – I2IT, Pune, IN; 2 – SNHU, NH, US;
Flow CPU GPU Speedup
(ms) (ms)
Preprocessing 45 2.12 21.22
Feature Extraction 52 3.54 14.68
Neural Network
Training 192 25.21 7.61
Neural Network
Testing 62 12.23 5.06
Total 351 43 8.16
Preprocessing Feature Extraction Pattern recognition
Pattern recognition
Image from
camera
Output
( No. of
count )
Training File
Preprocessing Feature Extraction
Image from
camera
(a) Training Phase
(b) Testing Phase
No of epochs MSE
Time in
seconds
2000 1.64272e-006 12.62
5000 1.58837e-006 19.875
8000 1.41844e-006 8.359
12000 1.92353e-006 9.484
No. of neurons in Hidden Layer: 7
MSE: 1.61835e-006
Elapsed time : 11.453000 seconds.
No. of neurons in Hidden Layer: 8
MSE : 1.39452e-006
Elapsed time: 8.281000 seconds.
contact Name
Pragati Dharmale: pragatiusstudies@gmail.com
Poster
P5235
Category: Computer Vision & Machine Vision - CV16

This chapter discusses a CUDA implementation of the structural similarity index (SSIM) for image quality assessment. SSIM is computationally intensive but can be parallelized for GPU implementation. The author implemented SSIM on CUDA and evaluated performance on Nvidia GPUs, achieving a speedup of 30x on GTX275 and 80x on C2050 compared to a single-core CPU implementation. Evaluation on test images showed the CUDA implementation produced SSIM values matching the CPU version while significantly reducing computation time.

Multi-core GPU – Fast parallel SAR image generationMahesh Khadatare

This document summarizes a poster presentation about using GPUs to parallelize synthetic aperture radar (SAR) image generation. It describes a proposed parallel algorithm that divides SAR data into blocks and processes the range and azimuth dimensions concurrently on GPU cores. Experimental results show that an Nvidia Fermi S2050 GPU achieved the highest performance and fastest speeds, processing a 1024x1024 SAR image over 74 times faster than a CPU.

Acceleration of the Longwave Rapid Radiative Transfer Module using GPGPUMahesh Khadatare

This poster presents Weather Research and Forecast (WRF) model is a next-generation mesoscale numerical weather prediction system designed to serve both operational forecasting and atmospheric research communities. WRF offers multiple physics options, one of which is the Long-Wave Rapid Radiative Transfer Model (RRTM). Even with the advent of large-scale parallelism in weather models, much of the performance increase has came from increasing processor speed rather than increased parallelism. We present an alternative method of scaling model performance by exploiting emerging architectures like GPGPU using the fine-grain parallelism. We claim to get much more than 23.71x, performance gain by using asynchronous data transfer, use of texture memory and the techniques like loop unrolling.

Slide tesiNicolò Savioli

This thesis implements the WHAM algorithm for estimating free energy profiles from molecular dynamics simulations using NVIDIA CUDA to run it on GPUs. WHAM iteratively calculates unbiased probability distributions and free energy values from multiple biased simulations. The author parallelized WHAM using CUDA, testing performance on different GPU architectures versus CPUs. GPU-WHAM achieved convergence comparable to CPU-WHAM but was up to twice as fast, with the speedup increasing with GPU computational capabilities. The GPU/CPU speed ratio remained constant with varying problem sizes, indicating good scaling.

CUDA and Caffe for deep learningAmgad Muhammad

This document discusses GPU computing and CUDA programming. It begins with an introduction to GPU computing and CUDA. CUDA (Compute Unified Device Architecture) allows programming of Nvidia GPUs for parallel computing. The document then provides examples of optimizing matrix multiplication and closest pair problems using CUDA. It also discusses implementing and optimizing convolutional neural networks (CNNs) and autoencoders for GPUs using CUDA. Performance results show speedups for these deep learning algorithms when using GPUs versus CPU-only implementations.

Image Processing TARUS: a FARMBOTShivang Rana

Block diagram of Robot functioning. Hardware required to make robot. Programming of image processing in MATLAB. (simulation) Implementation of control circuit in ARDUINO. (simulation) Development of “TAURUS: v1.0”: a farmbot. Hardware for dual axis Camera focus system. (video) Programming in the software: Processing. Feature Extraction for a Object: Pixel Detection & Motion Detection Conclusion

PFN Summer Internship 2019 / Kenshin Abe: Extension of Chainer-Chemistry for ...Preferred Networks

Parallel Implementation of K Means Clustering on CUDAprithan

IIBMP2019 講演資料「オープンソースで始める深層学習」Preferred Networks

2019年9月に開催された日本バイオインフォマティクス学会2019年年会第8回生命医薬情報学連合大会（IIBMP2019）のスポンサーセッション講演資料です。・鈴木脩司「Preferred Networksについて」・大野健太「深層学習フレームワーク Chainer」・菅原洋平「深層学習初学者向け無料オンライン教材の紹介」・中郷孝祐「化学・生物学分野のための深層学習ライブラリ ChainerChemistry」・太田健「ハイパーパラメータ自動最適化フレームワーク Optuna」

IMQA PosterVignesh Kannan

On the Computational Performance of Single-GPU and Multi-GPU Implementations of the MAD IQA Algorithm The document describes implementations of the MAD image quality assessment algorithm on single and multiple GPUs. The single GPU version achieved speedups over a CPU implementation by parallelizing kernels. The multi-GPU version further improved performance by distributing tasks across GPUs, hiding PCIe transfer latency through asynchronous memory copies. Experimental results showed the multi-GPU version was 30% faster than the single GPU implementation due to improved GPU utilization and overlap of computation and data transfers.

An Open Source solution for Three-Dimensional documentation: archaeological a...Giulio Bigliardi

The modern techniques of Structure from Motion (SfM) and Image-Based Modelling (IBM) open new perspectives in the field of archaeological documentation, providing a simple and accurate way to record three dimensional data. The software Python Photogrammetry Toolbox (PPT) is an Open Source solution that implements a pipeline to perform 3D reconstruction from a set of pictures. It takes pictures as input and performs automatically the 3D reconstruction for the images for which 3D registration is possible. It is composed of python scripts that automate the different steps of the workflow. The entire process is reduced in two commands, calibration and dense reconstruction. The user can run it from a graphical interface or from terminal command. Calibration is performed with Bundler while dense reconstruction is done through CMVS/PMVS. Despite the automation, the user can control the final result choosing two initial parameters: the image size and the feature detector. Acting on the first parameter determines a reduction of the computation time and a decreasing density of the point cloud. Acting on the feature detector influences the final result: PPT can work both with SIFT (patent of the University of British Columbia - freely usable only for research purpose) and with VLFEAT (released under GPL v.2 license). The use of VLFEAT ensures a more accurate result, though it increases the time of calculation. Python Photogrammetry Toolbox, released under GPL v.3 license, is a classical example of FLOSS project in which instruments and knowledge are shared. The community works for the development of the software, sharing code modification, feed-backs and bug-checking.

Find nuclei in images with U-netDing Li

U-Net is a convolutional neural network used for biomedical image segmentation. It takes in an input image and outputs a segmentation map identifying nuclei pixels. The U-Net architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. The model was trained on microscopy images annotated with nuclei masks to achieve an intersection over union score of 85% after data augmentation was applied.

An35225228IJERA Editor

International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.

NVIDIA 深度學習教育機構 (DLI): Image segmentation with tensorflowNVIDIA Taiwan

深度學習圖像分割課程目標與介紹許多重要的應用程式在圖像中偵測一個以上的對象，這時候就必須將圖像分割成小空間區域並且標明不同種類。圖像分割常被應用在醫學圖像分析或是自主駕駛車等等領域。本課程將帶領學員透過 Tensorflow 架構，實際操作整理好的醫學影像與自主駕駛車資料庫做學習，您將有機會熟悉如何在複雜的醫療影像中分解不同類型的人體組織，血管或異常細胞，並進一步由原始影像分割出特定器官。

IRJET-ASIC Implementation for SOBEL AcceleratorIRJET Journal

This document summarizes an ASIC implementation of a Sobel edge detection accelerator. It begins with an introduction to edge detection and accelerators. It then describes the proposed pipeline architecture of the Sobel accelerator, which takes in pixels from an image and processes them through multiplication, addition and other operations to produce output derivative pixels. The document discusses the ASIC design flow, including frontend steps like simulation, synthesis and DFT insertion, as well as backend steps such as floorplanning, placement, routing and timing analysis. It provides diagrams of the accelerator architecture and screenshots of synthesis reports.

HC-4012, Complex Network Clustering Using GPU-based Parallel Non-negative Mat...AMD Developer Central

딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)Hansol Kang

The document summarizes the basics of Deep Convolutional Neural Networks (DCNNs) including AlexNet and VGGNet. It discusses how AlexNet introduced improvements like ReLU activation and dropout to address overfitting issues. It then focuses on the VGGNet, noting that it achieved good performance through increasing depth using small 3x3 filters and adding convolutional layers. The document shares details of VGGNet configurations ranging from 11 to 19 weight layers and their performance on image classification tasks.

IEEEFYP2014poster Track 5 Vincent Yeong Chun KiatVincent Yeong

This document presents a multi-core processor architecture on an FPGA for image feature extraction and classification. It shows that an FPGA-based multi-core processor can process images faster than a single-core processor, with correlation results that are comparable to MATLAB. The FPGA implementation was able to achieve image processing times that were 3.419 times faster than a single-core processor. This FPGA-based multi-core architecture allows for flexible and high-performance image processing.

Image Processing using Matlab ( using a built in Highboost filtering,averagin...Majd Khaleel

1) The document describes a student's implementation of highboost filtering in MATLAB. The student used an averaging filter to blur an image, then applied a Laplacian filter with a factor of k=3 to the blurred image, making it sharper than the original. 2) Figures 2.1 and 2.2 show the original image, blurred image, and output of applying the code to the blurred image. 3) The student tested different values of the factor k, including k=5, k=1, k=9, and k=18, and found that the best value was k=1.5.

Implementation of PD controller in attitude of quadcopterTack-geun You

Ultra Fast SOM using CUDAQuEST Global (erstwhile NeST Software)

SOM (Self-Organizing Map) is one of the most popular artificial neural network algorithms in the unsupervised learning category. For efficient construction of large maps searching the best-matching unit is usually the computationally heaviest operation in the SOM. The parallel nature of the algorithm and the huge computations involved makes it a good target for GPU based parallel implementation. This paper presents an overall idea of the optimization strategies used for the parallel implementation of Basic-SOM on GPU using CUDA programming paradigm.

Operation cost reduction in unit commitment problem using improved quantum bi...IJECEIAES

Unit Commitment (UC) is a nonlinear mixed integer-programming problem. UC used to minimize the operational cost of the generation units in a power system by scheduling some of generators in ON state and the other generators in OFF state according to the total power outputs of generation units, load demand and the constraints of power system. This paper proposes an Improved Quantum Binary Particle Swarm Optimization (IQBPSO) algorithm. The tests have been made on a 10-units simulation system and the results show the improvement in an operation cost reduction after using the proposed algorithm compared with the ordinary Quantum Binary Particle Swarm Optimization (QBPSO) algorithm

Improving initial generations in pso algorithm for transportation network des...ijcsit

Transportation Network Design Problem (TNDP) aims to select the best project sets among a number of new projects. Recently, metaheuristic methods are applied to solve TNDP in the sense of finding better solutions sooner. PSO as a metaheuristic method is based on stochastic optimization and is a parallel revolutionary computation technique. The PSO system initializes with a number of random solutions and seeks for optimal solution by improving generations. This paper studies the behavior of PSO on account of improving initial generation and fitness value domain to find better solutions in comparison with previous attempts.

IRJET- Image Compression Technique for Quantized Encrypted Images using SVDIRJET Journal

This document proposes an image compression technique for encrypted images using discrete wavelet transform, singular value decomposition, and Huffman coding. At the sender side, the image is first encrypted using quantized wavelet coefficients and a secret key. Significant and less significant wavelet coefficients are then selected. The encrypted low-low sub-band is quantized and compressed with Huffman coding, while other sub-bands are compressed by discarding irrelevant information using SVD and Huffman coding. At the receiver side, decompression and decryption are performed using the compressed bit streams and inverse processes to reconstruct the original image. The technique achieves good compression performance while maintaining image quality, as shown through compression ratio and PSNR evaluation on test images.

Mapping Parallel Programs into Hierarchical Distributed Computer SystemsMikhail Kurnosov

This document presents research on mapping parallel programs to hierarchical distributed computer systems. It introduces a heuristic algorithm called TMMGP that partitions task graphs and maps them to optimize execution time. Experiments show TMMGP reduces execution time of MPI benchmarks on two computer clusters by up to 40% on average compared to standard MPI mapping tools. The research aims to develop models and algorithms that better account for the hierarchical organization of modern distributed systems and structure of parallel programs. Future work includes mapping to arbitrary subsystems and integrating the algorithm with resource management systems.

B Eng Final Year Project Presentationjesujoseph

Ppt finalravi teja

This document presents a face recognition methodology based on information theory. It discusses preprocessing steps like normalization, filtering and feature extraction. A pattern classifier is then used to compare extracted features from an input face image to those stored in a database, classifying the image as known or unknown. Potential applications include person identification, human-computer interaction and security systems. While the current system depends on consistent backgrounds, future work could integrate background removal to improve performance. Eigenfaces are used to represent feature vectors input to a neural network for classification and recognition.

Hand Gesture Recognition System for Human-Computer Interaction with Web-Camijsrd.com

This paper represents a comparative study of exiting hand gesture recognition systems and gives the new approach for the gesture recognition which is easy cheaper and alternative of input devices like mouse with static and dynamic hand gestures, for interactive computer applications. Despite the increase in the attention of such systems there are still certain limitations in literature. Most applications require different constraints like having distinct lightning conditions, usage of a specific camera, making the user wear a multi-coloured glove or need lots of training data. The use of hand gestures provides an attractive alternative to cumbersome interface devices for human-computer interaction (HCI). This interface is simple enough to be run using an ordinary webcam and requires little training.

ADITYA UMARE_CVAditya Umare

Aditya Umare has over 5 years of experience in project engineering and management. He currently works as a Project Engineer at BBM Acoustics India Pvt. Ltd., where he is responsible for project planning, scheduling, coordination, and ensuring timely completion. Previously, he worked at Tecpro Systems Ltd. as a Trainee Engineer and Project Engineer, managing projects in the power industry. He holds a B.E. in Mechanical Engineering and has expertise in project planning, management, scheduling, and coordination.

Presentación matrimonios 2015oscarpley2012

More Related Content

What's hot (18)

IIBMP2019 講演資料「オープンソースで始める深層学習」Preferred Networks

IMQA PosterVignesh Kannan

An Open Source solution for Three-Dimensional documentation: archaeological a...Giulio Bigliardi

Find nuclei in images with U-netDing Li

An35225228IJERA Editor

NVIDIA 深度學習教育機構 (DLI): Image segmentation with tensorflowNVIDIA Taiwan

IRJET-ASIC Implementation for SOBEL AcceleratorIRJET Journal

HC-4012, Complex Network Clustering Using GPU-based Parallel Non-negative Mat...AMD Developer Central

딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)Hansol Kang

IEEEFYP2014poster Track 5 Vincent Yeong Chun KiatVincent Yeong

Image Processing using Matlab ( using a built in Highboost filtering,averagin...Majd Khaleel

Implementation of PD controller in attitude of quadcopterTack-geun You

Ultra Fast SOM using CUDAQuEST Global (erstwhile NeST Software)

Operation cost reduction in unit commitment problem using improved quantum bi...IJECEIAES

Improving initial generations in pso algorithm for transportation network des...ijcsit

IRJET- Image Compression Technique for Quantized Encrypted Images using SVDIRJET Journal

Mapping Parallel Programs into Hierarchical Distributed Computer SystemsMikhail Kurnosov

B Eng Final Year Project Presentationjesujoseph

IIBMP2019 講演資料「オープンソースで始める深層学習」Preferred Networks

IMQA PosterVignesh Kannan

An Open Source solution for Three-Dimensional documentation: archaeological a...Giulio Bigliardi

Find nuclei in images with U-netDing Li

An35225228IJERA Editor

NVIDIA 深度學習教育機構 (DLI): Image segmentation with tensorflowNVIDIA Taiwan

IRJET-ASIC Implementation for SOBEL AcceleratorIRJET Journal

HC-4012, Complex Network Clustering Using GPU-based Parallel Non-negative Mat...AMD Developer Central

딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)Hansol Kang

IEEEFYP2014poster Track 5 Vincent Yeong Chun KiatVincent Yeong

Image Processing using Matlab ( using a built in Highboost filtering,averagin...Majd Khaleel

Implementation of PD controller in attitude of quadcopterTack-geun You

Ultra Fast SOM using CUDAQuEST Global (erstwhile NeST Software)

Operation cost reduction in unit commitment problem using improved quantum bi...IJECEIAES

Improving initial generations in pso algorithm for transportation network des...ijcsit

IRJET- Image Compression Technique for Quantized Encrypted Images using SVDIRJET Journal

Mapping Parallel Programs into Hierarchical Distributed Computer SystemsMikhail Kurnosov

B Eng Final Year Project Presentationjesujoseph

Viewers also liked (20)

Ppt finalravi teja

Hand Gesture Recognition System for Human-Computer Interaction with Web-Camijsrd.com

ADITYA UMARE_CVAditya Umare

Presentación matrimonios 2015oscarpley2012

22 diferente do outro, avaliação clínica do tratamento restaurador atraumáti...Luanna Melo

Este documento descreve um estudo clínico que avaliou o tratamento restaurador atraumático (ART) em crianças do Movimento Sem-Terra na região de Promissão, São Paulo. O ART envolveu a remoção do tecido cariado manualmente e restauração com ionômero de vidro. Após 3 meses, 63% das restaurações foram consideradas aceitáveis e 37% não aceitáveis. Os resultados demonstraram que o ART é útil para atendimentos sociais e pacientes carentes.

Presentació p 4 2015-2016 (edit)Gaudicastelldefels

graphing experimental resultHelal Uddin Mullah

The document discusses different types of graphical representations used for experimental data. It describes that William Playfair invented the first statistical graphs in the late 18th century, including line graphs, bar graphs, and pie charts. The document then provides details on different graph types, such as defining line graphs as showing changes over time, bar graphs as comparing categories, and pie charts as dividing a whole into parts. It also discusses other graphs like histograms, polygraphs, and compound bar graphs.

Día de los muertosap10349

Anwendungsbeispiele Industrie Terminals 2Nicole Eisele

Coursera DUVSFFTADYC2Ferdi Ramdhon

Estudio de casosCarlos Marín

El cuento del gusanoAlicia Martínez

Job Profiles In Gaming IndustryAnimation Kolkata

Real time gesture recognition of human handVishnu Kudumula

This document discusses real-time gesture recognition of the human hand. It begins by defining what gestures are and how they can be used for human-computer interfaces and sign language. It then discusses challenges like finding the hand in an image, recognizing its shape and motion, and determining its position in 3D space. Potential methods explored include datagloves, colored gloves, and vision-based systems using features like color segmentation and tracking hand shape trajectories over time. The document considers techniques such as hidden Markov models, hierarchical searching, and predicting and tracking hand positions.

gesture recognition!mehran kordavani

This document provides an overview of gesture recognition technology, including what gestures are, the history and basic workings of gesture recognition, different types of gesture recognition and sensing technologies, algorithms used, applications, and challenges. It discusses hand, facial, and sign language recognition and technologies like wired gloves, cameras, and controllers. Benefits include interacting without mouse/keyboard and with 3D environments without physical contact. Applications include rehabilitation, sign language, gaming, and assisting those with disabilities.

Que es el mantenimiento de un artefacto tecnologicoMarleny Gutierrez Pedroza

El documento habla sobre el mantenimiento de artefactos tecnológicos. Explica que todo objeto tecnológico requiere mantenimiento periódico para detectar problemas en su funcionamiento. Luego describe dos tipos de mantenimiento preventivo: el mantenimiento preventivo que se realiza en equipos en buen estado de funcionamiento para evitar fallas, y el mantenimiento correctivo que se aplica una vez detectada una falla para repararla. Finalmente, se enfoca en el mantenimiento de las computadoras personales, explicando que consiste en rutinas perió

SPEECH RECOGNITION USING NEURAL NETWORK Kamonasish Hore

The project was started with a sole aim in mind that the design should be able to recognize the voice of a person by analyzing the speech signal. The simulation is done in MATLAB. The design of the project is based on using the Linear prediction filter coefficient (LPC) and Principal component analysis (PCA) on data (princomp) for the speech signal analysis. The Sample Collection process is accomplished by using the microphone to record the speech of male/female. After executing the program the speech is analyzed by the analysis part of our MATLAB program code and our design should be able to identify and give the judgment that the recorded speech signal is same as that of our desired output.

Artificial Neural Network Based Object Recognizing RobotJaison Sabu

Neural tool boxMohan Raj

The Matlab neural network toolbox provides tools for designing, implementing, visualizing and simulating neural networks. It supports common network architectures and training functions. The GUI allows users to create and train networks, view network performance, and export results to the workspace. Sample code shows how to create a network, design a parity problem network, train it, and view the network weights and performance.

Fuzzy and Neural Approaches in Engineering MATLABESCOM

This document provides an introduction to a MATLAB supplement for the book "Fuzzy and Neural Approaches in Engineering". It describes MATLAB as an educational software package for technical computing. The supplement contains MATLAB code examples that demonstrate concepts from the book, such as neural networks, fuzzy logic, and hybrid systems. It is intended to help readers gain a practical understanding of implementing soft computing techniques in MATLAB.

Ppt finalravi teja

Hand Gesture Recognition System for Human-Computer Interaction with Web-Camijsrd.com

ADITYA UMARE_CVAditya Umare

Presentación matrimonios 2015oscarpley2012

22 diferente do outro, avaliação clínica do tratamento restaurador atraumáti...Luanna Melo

Presentació p 4 2015-2016 (edit)Gaudicastelldefels

graphing experimental resultHelal Uddin Mullah

Día de los muertosap10349

Anwendungsbeispiele Industrie Terminals 2Nicole Eisele

Coursera DUVSFFTADYC2Ferdi Ramdhon

Estudio de casosCarlos Marín

El cuento del gusanoAlicia Martínez

Job Profiles In Gaming IndustryAnimation Kolkata

Real time gesture recognition of human handVishnu Kudumula

gesture recognition!mehran kordavani

Que es el mantenimiento de un artefacto tecnologicoMarleny Gutierrez Pedroza

SPEECH RECOGNITION USING NEURAL NETWORK Kamonasish Hore

Artificial Neural Network Based Object Recognizing RobotJaison Sabu

Neural tool boxMohan Raj

Fuzzy and Neural Approaches in Engineering MATLABESCOM

Similar to Enhanced Human Computer Interaction using hand gesture analysis on GPU (20)

Hand Finger Counting using Deep Convolutional Neural Network (CNN) on GPUMahesh Khadatare

This poster represents active research topic in human computer interaction (HCI) as automatic hand finger counting using deep Convolutional Neural Network (CNN). To accelerate projected algorithmic program, leverage CUDA 8.0 platform from the NVIDIA GPU. Hand finger Counting and recognition deals with real time application, that leads us optimize algorithm with maximum number of images for CNN training. Projected methodology implemented in C, CUDA. Algorithmic program is complicated a part of feature extraction boosted up using multi-threaded CUDA calls. Application of this algorithm is proposed for autonomous fire-fighting robot which has on-board camera and embedded GPU processor. Testing accuracy is measured with known and unknown image dataset, typical testing accuracy achieved 98% for unknown finger counting. A CUDA GPU (GT820M) performance improvement of 40x over the single core Intel processor.

Computer-Vision_Integrating-Technology_MOB_17.06.16Schuyler Kennedy

This document discusses computer vision and Mobica's work in the field. It provides an overview of computer vision, including common uses and relevant technologies like OpenCV. Mobica has experience implementing and optimizing computer vision algorithms using technologies like OpenCL. They have worked on projects involving object recognition, image processing, augmented reality, and using computer vision in applications like automotive systems and gesture recognition. Mobica sees opportunities to improve computer vision performance and make it more accessible to developers.

Benchmark of common AI accelerators: NVIDIA GPU vs. Intel MovidiusbyteLAKE

Implementation of embedded arm9 platform using qt and open cv for human upper...Krunal Patel

: In this Paper, A novel architecture for automotive vision using an embedded device will be implemented on ARM9 Board with highly computing capabilities and low processing power. Currently, achieving real-time image processing routines such as convolution, thresholding, edge detection and some of the complex media applications is a challenging task in embedded Device, because of limited memory. An open software framework, Linux OS is used in embedded devices to provide a good starting point for developing the multitasking kernel, integrated with communication protocols, data management and graphical user interface for reducing the total development time. To resolve the problems faced by the image processing applications in embedded Device a new application environment was developed. This environment provides the resources available in the operating system which runs on the hardware with complex image processing libraries. This paper presents the capture of an image from the USB camera, applied to image processing algorithms to Detect Human Upper Body. The application (GUI) Graphical User Interface was designed using Qt and ARM Linux gcc Integrated Development Environment (IDE) for implementing image processing algorithm using Open Source Computer Vision Library (OpenCV). This developed software integrated in mobiles by the cross compilation of Qt and the OpenCV software for Linux Operating system. The result utilized by Viola and Jones Algorithm with Haar Features of the image using OpenCV.

Gpudigital lab for english partnersOleg Gubanov

This document describes a scientific simulation platform called GPUDigitalLab created by Oleg Gubanov using Microsoft DirectCompute. The platform uses a computational kernel to parallelize simulations across GPUs. It splits complex tasks into independent computational agents and uses a framework to control agent behavior and update results. Simulations can program 3D animations. The platform is intended for use in scientific modeling, simulations, and data analysis applications across various domains.

VisionizeBeforeVisulaize_IEVC_FinalMasatsugu HASHIMOTO

The document discusses visualization systems and proposes concepts for their future development. It summarizes: 1) The "Visual Realityware" visualization software development environment, which uses an abstraction layer to allow developers to freely select mainstream graphics technologies and expand applications across multiple platforms with minimal bugs. 2) An application called "Virtual Anatomia" developed using Visual Realityware to visualize 3D biological data in real-time. 3) The concept of "Visionize" which is defined as a risk management methodology using visual communication to allow sharing of goals and visions in order to identify and prevent risks before issues arise.

Machine-Learning-based Performance Heuristics for Runtime CPU/GPU SelectionAkihiro Hayashi

Sample of project synopsis (2)TH8B

This document provides a synopsis for a final year project on a Driver Sleep Detection System. It includes an abstract describing the problem of drivers feeling sleepy while driving and the goal to develop an efficient and low-cost detection system. It also includes a literature review of different detection techniques, a proposal for the algorithm and hardware requirements, and a basic project plan and timeline. The goal is to use computer vision and a microcontroller to accurately detect when a driver's eyes are closed for too long, and trigger an alarm.

CNC plotter controlled using Android applicationYuval Yoskovits

MAJOR PROJECT sandeep amaravadi

This document is a project report on a face recognition and tracking system. It includes an acknowledgements section thanking those who helped with the project. It also includes an abstract describing the project as building a system for face recognition and tracking using image processing and computer vision toolboxes in MATLAB. The document outlines the various chapters that will be included, such as introductions to image processing and the hardware and software used, including Arduino and MATLAB. It provides block diagrams of the overall system design and hardware.

IRJET- Self Driving RC Car using Behavioral CloningIRJET Journal

This document describes a project to convert a remote-controlled car into a self-driving car using behavioral cloning and deep learning. The researchers recorded video of themselves driving the RC car manually and used this data to train a convolutional neural network to map images from the car's camera to steering commands. Their model achieved 84.5% accuracy in testing and was able to autonomously drive the car forward, left, and right based on the training. While successful, the approach has limitations such as requiring an expert driver for training and not handling dynamically changing scenarios well. The researchers suggest ways to improve the system such as using more powerful hardware and adding navigation and object detection capabilities.

Using GPUs to Handle Big Data with JavaTim Ellison

A copy of the slides presented at JavaOne conference 2014. Learn how Java can exploit the power of graphics processing units (GPUs) to optimize high-performance enterprise and technical computing applications such as big data and analytics workloads. This presentation covers principles and considerations for GPU programming from Java and looks at the software stack and developer tools available. It also presents a demo showing GPU acceleration and discusses what is coming in the future.

Image Processing Application on Graphics processorsCSCJournals

In this work, we introduce real time image processing techniques using modern programmable Graphic Processing Units GPU. GPU are SIMD (Single Instruction, Multiple Data) device that is inherently data-parallel. By utilizing NVIDIA new GPU programming framework, “Compute Unified Device Architecture” CUDA as a computational resource, we realize significant acceleration in image processing algorithm computations. We show that a range of computer vision algorithms map readily to CUDA with significant performance gains. Specifically, we demonstrate the efficiency of our approach by a parallelization and optimization of image processing, Morphology applications and image integral.

Computer Graphics PracticalNeha Sharma

The document discusses computer graphics and provides examples of its applications. It discusses graphics inbuilt functions such as arc(), initgraph(), closegraph(), and line(). It provides code snippets and explanations for these functions. It also lists algorithms for direct and Bresenham lines, circles, ellipses, and their code programs. Finally, it mentions static and dynamic applications of computer graphics.

Marek Suplata Projectsguest14f12f

The document contains summaries of several projects completed by Marek Šuplata including a moving object tracker, simulator of coordinating productions, face biometric recognition system, medical CT volume data visualization, power network blackouts monitor, and motion control projects in Matlab/Simulink including a positional servosystem and direct vector control loops for an asynchronous motor. Details provided for each project include description, source code size, tasks, technologies used, and duration.

Monte Carlo on GPUsfcassier

Monte Carlo simulation is well-suited for GPU acceleration due to its highly parallel nature. GPUs provide lower cost and higher performance than CPUs for Monte Carlo applications. Numerical libraries for GPUs allow developers to focus on their models rather than reimplementing basic components. NAG has developed GPU libraries including random number generators and is working with financial institutions to apply Monte Carlo simulations to problems in finance.

Nvidia (History, GPU Architecture and New Pascal Architecture)Saksham Tanwar

A GPU is an electronic circuit that rapidly manipulates memory to accelerate image processing and display. Modern GPUs use parallel processing, rendering each pixel and storing color, location and lighting data. GPUs have dedicated video memory and more cores than CPUs, making them better for processing large blocks of data. The Pascal GPU uses 16nm technology, HBM2 memory, NVLink interconnect and unified memory to improve performance for graphics and deep learning applications like physics simulations and image processing.

Smart Face Recognition System AnalysisVishal Aditya

The document discusses optimizing a face recognition model for processing images from multiple IP cameras with low memory usage and fast response times. It proposes using the LBPH face recognition algorithm with a database structure to match faces from the camera stream to trained images. Tests were able to recognize faces from a wireless camera with 95% accuracy using this approach on Google Cloud servers. Future work could involve object recognition, surveillance applications, and using deep learning models.

201001162_reportShirish Jadav

This document summarizes Shirish Jadav's B-Tech project involving internships at two startups - Aspirations and Transpose. At Aspirations, Shirish tested APIs for cloud storage, online gaming rooms and basic camera motions in a bike racing game. At Transpose, Shirish built hardware with a Raspberry Pi to capture traffic video data using a camera, process it to count vehicles, and send the data to a server. The projects helped Shirish gain experience with game development, hardware prototyping, and communicating across disciplines.

Interfacing of MATLAB with Arduino for Object Detection Algorithm Implementat...Panth Shah

This document describes a system that uses MATLAB and Arduino to detect and track objects in real-time video from a camera. The object detection algorithm is developed in MATLAB using digital image processing techniques. When an object is detected, its position is sent over serial communication to an Arduino board. This controls LEDs connected to the Arduino, indicating the object's detected position. The goal is to visually detect and track an object, sending the data to an Arduino board to control LEDs based on the object's motion. MATLAB is used for image processing and object detection, while Arduino receives serial data and controls the LED outputs.

Hand Finger Counting using Deep Convolutional Neural Network (CNN) on GPUMahesh Khadatare

Computer-Vision_Integrating-Technology_MOB_17.06.16Schuyler Kennedy

Benchmark of common AI accelerators: NVIDIA GPU vs. Intel MovidiusbyteLAKE

Implementation of embedded arm9 platform using qt and open cv for human upper...Krunal Patel

Gpudigital lab for english partnersOleg Gubanov

VisionizeBeforeVisulaize_IEVC_FinalMasatsugu HASHIMOTO

Machine-Learning-based Performance Heuristics for Runtime CPU/GPU SelectionAkihiro Hayashi

Sample of project synopsis (2)TH8B

CNC plotter controlled using Android applicationYuval Yoskovits

MAJOR PROJECT sandeep amaravadi

IRJET- Self Driving RC Car using Behavioral CloningIRJET Journal

Using GPUs to Handle Big Data with JavaTim Ellison

Image Processing Application on Graphics processorsCSCJournals

Computer Graphics PracticalNeha Sharma

Marek Suplata Projectsguest14f12f

Monte Carlo on GPUsfcassier

Nvidia (History, GPU Architecture and New Pascal Architecture)Saksham Tanwar

Smart Face Recognition System AnalysisVishal Aditya

201001162_reportShirish Jadav

Interfacing of MATLAB with Arduino for Object Detection Algorithm Implementat...Panth Shah

Recently uploaded (20)

MODULE 4 BUILDING PLANNING AND DESIGN SY BTECH HVAC SYSTEM IN BUILDINGDr. BASWESHWAR JIRWANKAR

ENERGY STORING DEVICES-Primary Battery.pdfTAMILISAI R

[HIFLUX] Lok Fitting&Valve Catalog 2025 (Eng)하이플럭스 / HIFLUX Co., Ltd.

Application Security and Secure Software Development LifecycleDrKavithaP1

[HIFLUX] High Pressure Tube Support Catalog 2025하이플럭스 / HIFLUX Co., Ltd.

HVAC Air Filter Equipment-Catalouge-Final.pdf FILTRATION ENGINEERING & CUNSULTANT

Optimize Indoor Air Quality with Our Latest HVAC Air Filter Equipment Catalogue Discover our complete range of high-performance HVAC air filtration solutions in this comprehensive catalogue. Designed for industrial, commercial, and residential applications, our equipment ensures superior air quality, energy efficiency, and compliance with international standards. 📘 What You'll Find Inside: Detailed product specifications High-efficiency particulate and gas phase filters Custom filtration solutions Application-specific recommendations Maintenance and installation guidelines Whether you're an HVAC engineer, facilities manager, or procurement specialist, this catalogue provides everything you need to select the right air filtration system for your needs. 🛠️ Cleaner Air Starts Here — Explore Our Finalized Catalogue Now!

ISO 4548-9 Oil Filter Anti Drain Catalogue.pdf FILTRATION ENGINEERING & CUNSULTANT

world subdivision.pdf...................bmmederos10

Tesia Dobrydnia - A Leader In Her IndustryTesia Dobrydnia

Tesia Dobrydnia brings her many talents to her career as a chemical engineer in the oil and gas industry. With the same enthusiasm she puts into her work, she engages in hobbies and activities including watching movies and television shows, reading, backpacking, and snowboarding. She is a Relief Senior Engineer for Chevron and has been employed by the company since 2007. Tesia is considered a leader in her industry and is known to for her grasp of relief design standards.

DIY Gesture Control ESP32 LiteWing Drone using PythonCircuitDigest

Build a gesture-controlled LiteWing drone using ESP32 and MPU6050. This presentation explains components, circuit diagram, assembly steps, and working process. Read more : https://circuitdigest.com/microcontroller-projects/diy-gesture-controlled-drone-using-esp32-and-python-with-litewing Ideal for DIY drone projects, robotics enthusiasts, and embedded systems learners. Explore how to create a low-cost, ESP32 drone with real-time wireless gesture control.

2. CT M35 Grade Concrete Mix design ppt.pdfsmghumare

Video Games and Artificial-Realities.pptxHadiBadri1

Highway Engineering - Pavement materialsAmrutaBhosale9

Kevin Corke Spouse Revealed A Deep Dive Into His Private Life.pdfMedicoz Clinic

Kevin Corke, a respected American journalist known for his work with Fox News, has always kept his personal life away from the spotlight. Despite his public presence, details about his spouse remain mostly private. Fans have long speculated about his marital status, but Corke chooses to maintain a clear boundary between his professional and personal life. While he occasionally shares glimpses of his family on social media, he has not publicly disclosed his wife’s identity. This deep dive into his private life reveals a man who values discretion, keeping his loved ones shielded from media attention.

Introduction to Machine Vision by CognexRicardoCunha203173

Forensic Science – Digital Forensics – Digital Evidence – The Digital Forensi...ManiMaran230751

Air Filter Flat Sheet Media-Catalouge-Final.pdf FILTRATION ENGINEERING & CUNSULTANT

This presentation provides a detailed overview of air filter testing equipment, including its types, working principles, and industrial applications. Learn about key performance indicators such as filtration efficiency, pressure drop, and particulate holding capacity. The slides highlight standard testing methods (e.g., ISO 16890, EN 1822, ASHRAE 52.2), equipment configurations (such as aerosol generators, particle counters, and test ducts), and the role of automation and data logging in modern systems. Ideal for engineers, quality assurance professionals, and researchers involved in HVAC, automotive, cleanroom, or industrial filtration systems.

MODULE 5 BUILDING PLANNING AND DESIGN SY BTECH ACOUSTICS SYSTEM IN BUILDINGDr. BASWESHWAR JIRWANKAR

Prediction of Unconfined Compressive Strength of Expansive Soil Amended with ...Journal of Soft Computing in Civil Engineering

Expansive soils (ES) have a long history of being difficult to work with in geotechnical engineering. Numerous studies have examined how bagasse ash (BA) and lime affect the unconfined compressive strength (UCS) of ES. Due to the complexities of this composite material, determining the UCS of stabilized ES using traditional methods such as empirical approaches and experimental methods is challenging. The use of artificial neural networks (ANN) for forecasting the UCS of stabilized soil has, however, been the subject of a few studies. This paper presents the results of using rigorous modelling techniques like ANN and multi-variable regression model (MVR) to examine the UCS of BA and a blend of BA-lime (BA + lime) stabilized ES. Laboratory tests were conducted for all dosages of BA and BA-lime admixed ES. 79 samples of data were gathered with various combinations of the experimental variables prepared and used in the construction of ANN and MVR models. The input variables for two models are seven parameters: BA percentage, lime percentage, liquid limit (LL), plastic limit (PL), shrinkage limit (SL), maximum dry density (MDD), and optimum moisture content (OMC), with the output variable being 28-day UCS. The ANN model prediction performance was compared to that of the MVR model. The models were evaluated and contrasted on the training dataset (70% data) and the testing dataset (30% residual data) using the coefficient of determination (R2), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) criteria. The findings indicate that the ANN model can predict the UCS of stabilized ES with high accuracy. The relevance of various input factors was estimated via sensitivity analysis utilizing various methodologies. For both the training and testing data sets, the proposed model has an elevated R2 of 0.9999. It has a minimal MAE and RMSE value of 0.0042 and 0.0217 for training data and 0.0038 and 0.0104 for testing data. As a result, the generated model excels the MVR model in terms of UCS prediction.

Filters for Electromagnetic Compatibility ApplicationsMathias Magdowski

In this lecture, I explain the fundamentals of electromagnetic compatibility (EMC), the basic coupling model and coupling paths via cables, electric fields, magnetic fields and wave fields. We also look at electric vehicles as an example of systems with many conducted EMC problems due to power electronic devices such as rectifiers and inverters with non-linear components such as diodes and fast switching components such as MOSFETs or IGBTs. After a brief review of circuit analysis fundamentals and an experimental investigation of the frequency-dependent impedance of resistors, capacitors and inductors, we look at a simple low-pass filter. The input impedance from both sides as well as the transfer function are measured.

MODULE 4 BUILDING PLANNING AND DESIGN SY BTECH HVAC SYSTEM IN BUILDINGDr. BASWESHWAR JIRWANKAR

ENERGY STORING DEVICES-Primary Battery.pdfTAMILISAI R

[HIFLUX] Lok Fitting&Valve Catalog 2025 (Eng)하이플럭스 / HIFLUX Co., Ltd.

Application Security and Secure Software Development LifecycleDrKavithaP1

[HIFLUX] High Pressure Tube Support Catalog 2025하이플럭스 / HIFLUX Co., Ltd.

HVAC Air Filter Equipment-Catalouge-Final.pdf FILTRATION ENGINEERING & CUNSULTANT

ISO 4548-9 Oil Filter Anti Drain Catalogue.pdf FILTRATION ENGINEERING & CUNSULTANT

world subdivision.pdf...................bmmederos10

Tesia Dobrydnia - A Leader In Her IndustryTesia Dobrydnia

DIY Gesture Control ESP32 LiteWing Drone using PythonCircuitDigest

2. CT M35 Grade Concrete Mix design ppt.pdfsmghumare

Video Games and Artificial-Realities.pptxHadiBadri1

Highway Engineering - Pavement materialsAmrutaBhosale9

Kevin Corke Spouse Revealed A Deep Dive Into His Private Life.pdfMedicoz Clinic

Introduction to Machine Vision by CognexRicardoCunha203173

Forensic Science – Digital Forensics – Digital Evidence – The Digital Forensi...ManiMaran230751

Air Filter Flat Sheet Media-Catalouge-Final.pdf FILTRATION ENGINEERING & CUNSULTANT

MODULE 5 BUILDING PLANNING AND DESIGN SY BTECH ACOUSTICS SYSTEM IN BUILDINGDr. BASWESHWAR JIRWANKAR

Prediction of Unconfined Compressive Strength of Expansive Soil Amended with ...Journal of Soft Computing in Civil Engineering

Filters for Electromagnetic Compatibility ApplicationsMathias Magdowski

Enhanced Human Computer Interaction using hand gesture analysis on GPU

1. Abstract This poster represent very active research topic in human computer interaction (HCI) as automatic hand gesture recognition using nvidia GPU. In this work neural network based video gesture are processed and recognize the finger counts. Due to real time requirement algorithm need to optimize and computationally efficient. We implemented the MATLAB code, it perform slow when neural network processing started. Implementing them in a parallel programming model such as GPU-CUDA would provide the necessary gain in processing speed. Algorithmic result validation is done using standard video data set and recognition rate is calculated. A performance improvement of 15x speed is achieved which is faster than Intel quad core processor. Introduction The goal of this project is to develop a program implementing gesture recognition. At any time, a user can exhibit his hand doing a specific gesture in front of a video camera linked to a computer. The program has to collect pictures of this gesture, analyze it and to identify the sign. In order to lighten the project, it has been decided that the identification would consist in counting the number of fingers that are shown by the user in the input picture. Database And MATLAB Results CUDA Approach 1. Camera capture video – process frame by frame image 2. Extract features from the image using various segmentation techniques implemented in MATLAB, CUDA. Output features then written to a text file. 3. Use the matrix obtained in step 2 to train the neural network(on CPU using MATLAB). 4. Once the network is trained , we test it on GPU. 5. GPU implementation :- a. Load the test image into GPU global memory b. Process it as in step 2, output will be matrix of size (262144 x1). c. Pad the matrix with 0's - output will be a matrix of size (262144 x16) - Inputs matrix. d. This is all done on the GPU. Now pass the neural network weights matrix, to the GPU. e. Multiply neuralnetwork_weights matrix by Inputs matrix - which will give a 16 x 16 matrix. Extract the first five values from it. The index of the value which has maximum value + 1 gives us the count. NN Training Results GPU (GeForce GT 525M) References [1] Malima, A.; Ozgur,E.; Cetin,M.; [2006] “A Fast Algorithm for Vision-Based Hand Gesture Recognition for Robot Control”, Signal Processing and Communications Applications, IEEE [2] A Neural Network on GPU, http://www.codeproject.com/KB/graphics/GPUNN.aspx [3] NVIDIA Corporation, Compute Unified Device Architecture (CUDA), http://developer.nvidia.com/object/cuda.html. Enhanced Human Computer Interaction using hand gesture analysis on GPU Mahesh Khadtare1, Pragati Dharmale2 1 – I2IT, Pune, IN; 2 – SNHU, NH, US; Flow CPU GPU Speedup (ms) (ms) Preprocessing 45 2.12 21.22 Feature Extraction 52 3.54 14.68 Neural Network Training 192 25.21 7.61 Neural Network Testing 62 12.23 5.06 Total 351 43 8.16 Preprocessing Feature Extraction Pattern recognition Pattern recognition Image from camera Output ( No. of count ) Training File Preprocessing Feature Extraction Image from camera (a) Training Phase (b) Testing Phase No of epochs MSE Time in seconds 2000 1.64272e-006 12.62 5000 1.58837e-006 19.875 8000 1.41844e-006 8.359 12000 1.92353e-006 9.484 No. of neurons in Hidden Layer: 7 MSE: 1.61835e-006 Elapsed time : 11.453000 seconds. No. of neurons in Hidden Layer: 8 MSE : 1.39452e-006 Elapsed time: 8.281000 seconds. contact Name Pragati Dharmale: [email protected] Poster P5235 Category: Computer Vision & Machine Vision - CV16

Enhanced Human Computer Interaction using hand gesture analysis on GPU

Recommended

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Enhanced Human Computer Interaction using hand gesture analysis on GPU (20)

Recently uploaded (20)

Enhanced Human Computer Interaction using hand gesture analysis on GPU