DESIGN AND DEVELOPMENT OF

MOBILE OBJECT DETECTION USING

TENSOR FLOW LITE AND DEEP
LEARNING APPROACHES

Guided by, By,

S.Yogadinesh.,M.E D.Nivetha(920122421006)
 Abstract
• Mobile object detection has become an important research area
due to its various practical applications
• we propose a deep learning-based approach for mobile object
detection using TensorFlow Lite.
• We use a deep neural network with multiple convolutional layers
to extract features from the input image.
• . The extracted features are then fed to a set of detection heads to
predict the bounding boxes and corresponding class labels of the
objects in the image
• We use the MobileNetV2 architecture as our backbone network,
and the SSD (Single Shot Detector) algorithm as our detection
framework.
• Our approach can be used in a wide range of applications, such
as autonomous driving, surveillance, and augmented reality.
 INTRODUCTION
• Deep Learning approach it can automate the feature extraction
process and is effective for image recognition
• The TensorFlow Object Detection API is used to detect
multiple objects in real-time video streams
• Mobile object detection algorithms use a combination of
machine learning and computer vision techniques to analyze
images and identify objects within them
• These algorithms can be trained on large datasets of labeled
images to learn to recognize specific objects or categories of
objects.
• Once trained, the algorithms can be deployed on mobile
devices to analyze real-world images or video in real-time.
• object detection can be accomplished by using two approaches:
• machine learning based
• deep learning based.
• In the machine learning-based approach, the features of
computer vision are used to identify a group of pixels. The
features used are primary properties such as edges, shapes,
texture and colour. The features are imported to a regression-
based algorithm which returns the label and location of the
target object.
• In the deep learning-based approaches, Convolutional Neural
Networks (CNN) are employed. The usage of CNN result in an
unsupervised end-to-end object detection without the need for
feature extraction.
 EXISTING SYSTEM
• Image classification system based on a structure of a
Convolutional Neural Network (CNN).
• The image classification system employs the bi-scale CNN
with 120 trained data and the auto-stage training achieves
81.6% detection rate with only six false positives on Face
Detection Data Set and Benchmark (FDDB).where the current
state of the art achieves about 80% detection rate with 50 false
positives.
• It proposed fast image classification by boosting the Fuzzy
Classifiers.
DISADVANTAGES
• Low Accuracy
• Required High Memory Space
 PROPOSED SYSTEM
• We propose image recognition in Deep Learning using
convolutional neural network algorithm.
• The fundamental task of image classification is to make sure
all the images are categorized according to its specific sectors
or groups.
• The various applications such as vehicle navigation, robot
navigation and remote sensing by using image classification
technology.
ADVANTAGES
• High accuracy
• Real-time performance
• Efficient memory and power consumption
• Customizable
• Portability
 Algorithm
• CIFAR-100 Dataset as it suggests has 100 different categories of
images in it. There is a total of 60000 images of 10 different
classesnaming
Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse, Ship,
Truck. All the images are of size 32×32. There are in total 50000
train images and 10000 test images.
• Single Shot Detectors (SSDs) are a popular and efficient method
for object detection. They use a single convolutional neural network
(CNN) to predict bounding boxes and class labels for objects in an
image, making them faster and more efficient than other methods.
• TensorFlow Lite lets you run TensorFlow machine learning (ML)
models in your Android apps. The TensorFlow Lite system provides
prebuilt and customizable execution environments for running
models on Android quickly and efficiently, including options for
hardware acceleration
 System specification
HARDWARE SPECIFICATION

• Hard Disk: 500GB and Above

• RAM: 4GB and Above
• Processor: I 3 and Above

SOFTWARE SPECIFICATION

• Operating System: Windows 7, 8, 10 (64 bit)

• Software: Python 3.7
• IDE: Pycharm
 Module description

Examine and understand dataset

• “A dataset (or data set) is a collection of data, usually
presented in tabular form.
• Each column represents a particular variable.
• Each row corresponds to a given member of the dataset in
question.
• It lists values for each of the variables, such as height and
weight of an object. Each value is known as a datum.
Build an input dataset
 Training sample images

• Image dataset of CIFAR- 100 which has numerous super-

classes of general object images and a number of subclass
categories of each superclass. CIFAR-100 has 100 classes of
images with each class having 600 images each These 600
images are divided into 500 training images and 100 testing
images for each class, therefore, making a total of 60,000
different images.
 TRAIN THE MODEL

• Teachable Machine is a GUI tool that allows you to create

training dataset and train several types of machine learning
models, including image classification, pose classification and
sound classification. Teachable Machine uses TensorFlow.js
under the hood to train your machine learning model. You can
export the trained models in TensorFlow.js format to use in
web browsers, or export in TensorFlow Lite format to use in
mobile applications or IoT devices.
• Here are the steps to train your models:
• Go to Teachable Machine website
• Create an Image project
• Record some sound clips for each category that you want to
recognize. You need only 8 seconds of sound for each
category.
• Start training. Once it has finished, you can test your model
on live audio feed.
• Export the model in TFLite format.
 TEST MODEL

• 1. The dataset was downloaded from caltech’s website. There

were 101 classes which contained multiple images. We then
split the dataset into 80% training and 20% testing. The model
was designed using tensorflow in keras with a combination of
convolutional and max pooling layers. This was then flattened
and provided to the fully connected layer. The validation
accuracy that we got was around 30%. With some tweaking of
the hyper parameters we were able to get accuracy of up to
62%.
• 2. This was happening due to class imbalance. So we decided
to restrict our training to 10 classes which had maximum
number of images.
• 3. As our model was ready, we had a lot of space to
improveour model’s performance on 10 classes. We decided to
go ahead with 2 activation functions and then compare the
results. The two activation functions were, ‘tanh’ and ‘relu’. As
an initiative to learn the working of the neural network, we
decided to visualize the output of every convolutional and max
pooling layer. Using tanh activation we were able to get 32%
accuracy.
• 4. We then used relu activation function for the same model.
With relu activation function, we were able to get 93%
accuracy for our model.
• 5. We even plotted multiple graphs throughout the progress of
the project. We compared the training accuracy vs validation
accuracy for all the models, training loss vs validation loss for
all the models. We even plotted the comparison plots for tanh
and relu activation functions.
Model Development
 CONCLUSION
• In conclusion, a mobile object detection system is a powerful
tool that enables real-time and accurate detection of objects in
images or videos. It has various applications in the fields of
surveillance, autonomous vehicles, robotics, and augmented
reality. Mobile object detection systems typically use deep
learning algorithms such as Convolutional Neural Networks
(CNNs) to analyze and classify images.
• These algorithms are trained on large datasets of labeled images
and are capable of detecting objects with high accuracy. Mobile
object detection systems are continuously evolving, and
advancements in machine learning and computer vision
technologies are making them more efficient and effective.
However, there are still some challenges that need to be
addressed, such as improving the accuracy of object detection in
complex environments and reducing the computational resources
required to run these systems.
 REFERENCES
• A. B. Nassif, M. A. Talib, Q. Nasir, Y. Afadar, and O. Elgendy,
‘‘Breast cancer detection using artificial intelligence
techniques: A systematic literature review,’’ Artif. Intell. Med.,
vol. 127, May 2022, Art. no. 102276, doi:
10.1016/j.artmed.2022.102276.
• A. B. Amjoud and M. Amrouch, ‘‘Transfer learning for auto-
matic image orientation detection using deep learning and
logistic regression,’’ IEEE Access, vol. 10, pp. 128543–
128553, 2022, doi: 10.1109/ACCESS.2022.3225455.
• Z. Sun, Q. Ke, H. Rahmani, M. Bennamoun, G. Wang, and J.
Liu, ‘‘Human action recognition from various data modalities:
A review,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 45,
no. 3, pp. 3200–3225, Mar. 2022, doi:
10.1109/TPAMI.2022.3183112.