Question 2

Old age home activity recognition

In old age home, activity recognition of elderly people is a crucial task, the physiological monitoring of older people using
wearable sensors has shown great potential in improving their quality of life and preventing undesired events related to
their health status. Under that premise, an activity recognition scheme is proposed for older people exploiting feature
extraction and machine learning, along with heuristic computational solutions to address the challenges due to inconsistent
measurements in non-standardized environments.

The proposed approach is devised as follows-

1. Computer Vision-based activity recognition
● A live video-processing software sends live video with HTTP Live Streaming (HLS) through the UDP
protocol​(Transport Layer)​ for faster uploads and generates fragments of equal length (10 seconds) and sends it
to the server cloud storage.
● An upload of each video fragment into the storage service triggers a lambda function.
● The lambda function simply invokes a machine learning endpoint for activity detection with the cloud storage
URL of the video fragment as the REST API payload input.
● The machine learning inference container reads the video from the cloud storage, pre-processes it, detects the
person in the frame as well as its activity, and saves the prediction results back to the cloud storage where it can
be further processed using streaming analytics software for later usage.
● If the activity detected falls under the criterion of “At-risk”, an SNS(messaging service) lambda is triggered which
alerts the caretakers of the elderly to quickly look into the matter.
2. Sensors based activity recognition
● Pulse sensors and accelerometer-gyroscope sensors can be placed on the wrist of the users to collect the heartbeats
and body temperature.
● Smartphones which are in itself an IoT device capable of sensing the environment, collecting, processing, and
transmitting data can be used to collect data and transmit to the server for processing.
● Anomaly detection methods can be applied for such data which can detect sudden movements of the elderly
which may be triggered due to a fall, cardiac dysfunction, or any other health issue.
 ● Mobile phones equipped with triaxial-accelerometer and triaxial-gyroscope can be used as sensors delivering data
for events like fall detection.
● Data can be transmitted periodically and form a data stream of JSON objects.(​Application and Presentation
Layer​)
The Stream Analytics module processes data streams from connected smartphones. The module has one input and several
outputs. The input accepts sensor readings as well as CCTV footage chunks that are sent from smartphones and cameras
through the IoT Hub. Messages reaching the IoT Hub service are automatically transferred to the Stream Analytics module,
which extracts appropriate features that will be used by the Machine Learning module and are then directed to appropriate
ML inference containers.

To reduce the amount of data transmitted to the Cloud, falls and other activities can also be detected on the smartphone of
the monitored person. This creates an alternative path in the data flow in the architecture The local ML model can be built
with the use of the CoreML libraries for Apple devices as well as TensorFlow Lite for Android Users. The approach to this
problem primarily relies on tuning hyperparameters of already trained models(Transfer Learning).

Precision Farming
Precision Agriculture (PA) or Site-Specific Crop Management (SSCM) is technically described as ‘a technology-enabled
approach to farming management that observes, measures, and analyzes the needs of individual fields and
crops’.Conventional farming practices are area-centric. There is a general set of crops cultivated throughout an area. All
the farmers in that area follow the same procedures with respect to sowing, nourishing, irrigation, and harvesting period.
What these practices result in is unpredictability, overuse of resources, and uncontrolled waste production. The rising
world population and unpredictable climate changes are the two major problems for the current agricultural scenario.
Hence farmers have started to transition their operations to precision agriculture. Most farms are in remote locations where
data network connectivity can be sparse. Frequent weather fluctuations can cause damage to or loss of sensors and devices.
These types of challenges require robust, simple, and easy to manage ubiquitous solutions. Measuring and collecting field,
crop, and ambient data are key capabilities enabled by mature IoT solutions.
Here are a few of the challenges related to implementing a cost-efficient precision farming solution with image
recognition:

● The amount of data generated by the cameras.

● The bandwidth and cost required to transmit the data for processing off-site.
● Access to image recognition intelligence on the cameras.

Proposed End-to-End Approach

● Rigorous data collection is done on soil testing, plot measurement, weather pattern analysis, and crop analysis
through sensor-equipped devices placed along the plantation fields. The data will be used to devise conclusions
and based upon those results a very detailed and precise set of practices can be adopted
● Sensory data from various IoT sensors is transmitted to a single edge device(Raspberry Pi 4, Nvidia Jetson)
through peer-to-peer sockets(​Internet Layer​) to track various factors affecting agricultural yield.
● The edge device sends the data to the cloud server through the API Gateway via a 3G/4G network service​(Data
Link Layer) ​where a lambda function gets triggered.
● The lambda function then processes the sensory data to provide valuable information and predictions using the
ML model containers and reverts the updated model parameters to the local ML core model, as well as arranges
the necessary action to be taken manually in a notification mail to the farmer via SNS.
● Water Management can be monitored and adjusted according to the data ingressed and processed, and instructions
can be sent back to the edge device which further controls the sprinklers through relays.
● Drones allow farmers to monitor how far along crops are in their respective growth periods. Additionally, farmers
can spray ailing crops via drones with substances to bring them back to life. Drones can spray fertilizer 40 to 60
times faster than doing so by hand. The Drone Camera data can be processed in Computer Vision models at the
local edge to provide surveillance of their lands(obtain detailed maps of both the topography and resources in the
area) and generate useful crop data.
 ● Performing inference locally on devices reduces the latency and cost of sending device data to the cloud via
3G/4G services to make a prediction. Rather than sending all data to the cloud for performing machine learning
inference, you run inference directly on the device. Performing edge computing we can capture the results, detect
outliers, and send data back to the cloud to ML model containers where it can be reclassified, tagged, and used for
model retraining to improve the accuracy of the model.
● Seed providers also use sensory data to design new seed varieties. They can improve seed characteristics by
evaluating the yields that their customers achieve in different soil types and weather conditions. A farmer could
also decide to share data from one field or only share data from one planting season with a partner increasing
revenue sources.
● Farmers can use their smartphones to remotely monitor their equipment, crops, and livestock, as well as obtain
stats on their livestock feeding and produce. They can even use this technology to run statistical predictions for
their crops and livestock.
● The factors that can be affected and improved using this method are -Fertilizers and their application timing,
irrigation and precipitation patterns, air temperature and moisture levels(in greenhouse environments)Soil
composition and treatment, crop care processes, etc.
● Pictures from the smart drone cameras will be used to regularly monitor, recognize, and count plant leaves,
flowers, plant height, and to calculate predicted greenhouse yields. Optimizing greenhouse operations by
effectively using the appropriate inputs of nutrients, water, and other elements at the most appropriate plant stage
growth makes greenhouse farming more economically and environmentally sustainable.

Conclusion
Before the use of tech in agriculture, a farmer’s probability of yielding good produce was as good as tossing a coin and
wishing for heads. Since farmers had no information on their farms, there was no way of learning the causes of crop loss.
This practice pushed the farmers towards losses and debt. Advancements in big data analytics, IoT, and accessible satellite
imagery created optimism for the agriculture sector, thereby combating the issue of unpredictability.
Drinking-Water Management System

Water Management especially potable water requires automated distribution, quality control as well as wastage prevention
in the current world scenario. Potable water is an unevenly distributed natural resource, hence requires an efficient way of
reaching the masses with standard pressure as well as parameters like pH, turbidity, salt content needs to always be kept in
check during the process of distribution. This water management system can be efficiently monitored and catered to with
the help of IoT sensors integrated with a cloud server that detects anomalies and reports the point of the outlier to the
management staff who can diagnose the problem quickly and efficiently.

Proposed End-to-End Solution

● IoT sensors can be deployed at the regional water tank centers to measure various water properties like turbidity,
pH level, and salt impurity levels. This can ensure the potability of water being deployed in the supply chain.
● Water pressure levels perform a key role in the water supply chain, hence pressure sensors, as well as sensory leak
detectors, can be deployed at a particular distance(0.5 km each) to ensure proper pressure level.
● At the household level, water wastage prevention becomes a major issue that can be kept in check through
ultrasonic sensors with Physical layer direct-wired connection to alert sensors/alarms to ensure water motors to be
turned off/on at the right time. Methods like rainwater harvesting can also be promoted through IoT sensors at the
household level.
 ● All IoT module devices consist of a TCP/ IP self-supported programming module that can include header and
trailer frame format.
● The data collected through sensors can be transmitted wirelessly (Wifi) to the Cloud Infrastructure where
anomalies can be detected through ML models and specific areas in the city can be easily detected where the issue
is found, hence can be catered with efficiently by sending appropriate alerts to the worker staff.
● The transmitter and receiver on the IoT module of the water pipes can be sent with a pipe address as a 16-bit
number for easy detection and location of the anomaly or leakage.
● The IoT enabled automatic meter reading solution consists of IoT smart water meters that send daily readings to
the cloud, forming a two-way data sharing between meters and CRM/billing systems and eliminating the need for
a personal meter reading. With the meter data being available in real-time, more flexible billing cycles would be
available to the water consumers instead of following the standard utility read cycles.

Hence the three major pain points of the drinking-water management system i.e automated meter reading, leakage
detection, and water quality monitoring are targeted and can be effectively achieved with our end-to-end approach.

Consumer Electronics Diagnosis System

● The proposed method contributes to recovering old fashion domestic appliances through the adoption of IoT &
Cloud (Ubiquitous Computing). The technical layman user will be able to use the proposed system to identify
faults that today only specialized technicians can observe. With the resources available, the detected faults can be
sent directly to the user or to a maintenance company that will proceed with the new set-up, fine-tune, or even
repair of the monitored appliance. The system can be designed to monitor a set of consumer electronic devices at
the same time.
● The approach can be integrated into a smart home environment. The idea is to keep the system running in order to
detect failures and develop digital services able to decide if the observed malfunctioning can be fixed by the
household owner or if it is necessary for professional help. In this second case, automatic calls will be sent to
registered suppliers.
● The manufacturer normally publishes the technical specifications of the monitored appliance. This data is part of
the off-line data collection of our system and will be one of the subsets of data that we will consider as part of the
analysis/filtering process.
● Data Collection in Real-Time Device – A low-cost micro-controlled device will be responsible for the collection
of data related to the operation of the domestic device in real-time. It will be able to connect to the Internet
permanently and will have a unique identification number. The device´s electrical magnitudes are obtained
through appropriate non-invasive sensors connected to them. The obtained data will compose the data set
necessary to identify the current state of the domestic device and will be sent to the cloud for further analysis.
● Preprocessing is responsible for collecting data in order to eliminate or minimize problems such as nonsense
numeric or symbolic data, clean data or noise, and data imperfections like incorrect, inconsistent, duplicate, or
missing values.
● Issues like unique identification of the connected device, as well as, data security and privacy are handled
by the cloud. It is also responsible for acting over domestic devices whenever necessary and possible.
For instance, in case of behavior that demands the shutting down of the monitored appliance, it will send
the corresponding signal to the microcontroller device of the respective electronic. In some cases, it may
be impossible, because there is no on/off device available. In that case, the shutdown command is
redirected to the User Module through SMS alerts that report the user about this situation.
● The FDD process involves the monitoring of the current state and the comparison to a reference of an
operation pattern. The operation parameters are distinct between each manufacturer and are normally
available in corporate databases. Over time, it is required to observe the behavior of the electrical
magnitudes and compare them with the expected value for the permanent regimen. When a fault occurs
and it is diagnosed as a risk to the device’s operation, the system will try automatically to shut it down.
For every type of fault, the cloud will store its residual values in a database. The fault is diagnosed
through the classifier and the FDDs decision-maker, which will identify the degree of relevance of the
fault by comparing it to the manufacturers´ information stored in the database and the Fault Data History
database.
● Finally, the cloud communicates the actual status of the appliances and report tendencies observed by the
Machine Learning Module(Fault Detection) to the service technician in case an in-depth diagnosis is
required as well to the customer to their smartphones to schedule their maintenance check accordingly.