Human Motion Analysis using Machine Learning

In order to assist and improvise the living conditions of the physically-challenged or age-affected individuals based at remote or isolated location, a home-based human-monitoring system is designed. This system comprises of a network of wearable sensing devices attached to an individual?s body. The

Project Title

Human Motion Analysis using Machine Learning

Project Area of Specialization

Wearables and Implantable

Project Summary

In order to assist and improvise the living conditions of the physically-challenged or age-affected individuals based at remote or isolated location, a home-based human-monitoring system is designed. This system comprises of a network of wearable sensing devices attached to an individual’s body. The sensors are interconnected to each other and a central microprocessor through either wires or a wireless medium. These sensors monitor the subject’s physical movements and communicate this real-time information to the microprocessor, which then reads and transmits it to a storage device. Afterwards, the data is sent to a local server where the implementation of the different machine learning algorithms occurs, translating it from a machine-readable form into a human-interpretable one. For further analysis, this information is relayed to the concerned healthcare personnel who, in case of any emergency, can be alarmed to take the immediate and necessary follow-ups.

This system focuses more on managing wellness rather than illness unlike the existing health care systems. The existing systems either use the less-power demanding wireless communication channels or the more-power demanding and complex, high-level wireless system like Bluetooth solely to transfer raw data from the sensors to the concerned personnel.  These features restrict their use for monitoring over long-time durations. Furthermore, it is very important that these sensors must be low-power consuming, as the research proceeds on finding out ways to save energy by making the self-powered implantable sensors in the future.  It is expected that the systems such as this, can become a gateway to a wide variety of revolutionizing possibilities in the healthcare sector that could significantly improve our lives.

Project Objectives

• To construct a working prototype of a body-worn system of multiple sensors, interlinked with a central receiver module.
• To implement different supervised machine learning algorithms on the acquired dataset to perform real-time human-motion analysis

Project Implementation Method

The hardware design of the system is an assembly of low-power consuming, small-sized, non-invasive and light-weight sensor units interconnected to form a network with wireless transmission links, operable across a specified range.

Multiple tiny heterogenous sensors are strategically placed on different parts of human body, creating a star network around it. Moreover, since the data is communicated through radio waves between the sensors, the complexity of design of the system is very minimal, thereby not creating any inconvenience and difficulty for the user to carry it around and allowing them to perform their daily activities without any physical hindrance.

A single sensor unit is composed of three sub-sensors which are accelerometer, magnetometer and gyroscope, interfaced with a Radio Frequency (RF) module via a microprocessor chip powered by a battery. The receiver module consists of another RF module, and another microprocessor board.

Data processing begins in the Arduino Nano board embedded in each sensor unit, as soon as the inertial measurements are received from the Inertial Measurement Unit (IMU). This incoming inertial data from each of the three sub-sensors (accelerometer, magnetometer and gyroscope), is read in Nine Degrees of Freedom, because the measurements are done in three separate axes (x, y and z). With the implementation of a fusion algorithm, these nine separate readings are fused into a tri-dimensional dataset.

The three dimensions, namely Roll, Pitch and Yaw, indicate the movement or rotation of each sensor unit, thus giving the orientation of the body as integer numbers. Once this information from all of the five sensors reaches the central microprocessor board through Radio-Frequency waves, it is then stored on the local server with an available Wi-Fi access which processes this collected information by applying the supervised machine learning algorithms on it. After the data is converted into a human-readable form, a medical professional based at a different location, analyses it and provides the required assistance to the patient. In case of any emergency, the system can efficiently issue timely warnings to the patient, as well as the responsible personnel.

Benefits of the Project

Majority of the existing fall detection systems are based on wearable sensor modalities, making it difficult for such individuals to carry out the activities of their daily lives unassisted. This significantly cuts down their contribution to the society and crushes their will to live. Our system prevents this from happening. Not only does it allow a safe, convenient and effective way for monitoring of patients in real time, it also gives them confidence and makes them feel looked after.

It can be further made use of in other application areas of similar interest, such as for the monitoring of physically or mentally challenged individuals, children who have been left alone and individuals of all sorts (criminals, athletes etc.) under observations. Improvisations can even allow location mapping, pose estimation, long term human motion prediction, behavioural analysis and usage even in medical robotics.

Technical Details of Final Deliverable

The hardware of wireless body area network consists of following

1. Sensor node
2. PDA
3. Multipoint access

Sensor node

 A sensor node is an electronic unit/circuitry that picks raw data from the human body. The sensor nodes undertake three tasks:

• Signal detection
• Signal processing
• Wireless transmission

Signal detection/picking is done by using MPU-9250 which is continuously collecting accelerometer, gyroscope, magnetometer readings, and simultaneously converting these into yaw, pitch and roll with the help of Arduino Nano.

Raw data signal processing is related to digitizing, coding, sampling controlling of raw signals to make sure that data is in understandable and readable form for multi access communication and finally wireless transmission using radio frequency module NRF24L01+.

There are multiple wireless technologies available in the market through which we can easily make our IMU star network. During the selection of the wireless technology, one should have to be careful about these factors:

• Power consumption
• Effectiveness
• Range
• Frequency

Software:

In the sensor node there is proper programming involved that helps in sampling, digitizing, controlling, and finally transmitting signal to a receiver station; a microprocessor (Arduino UNO) with NRF24L01 that receives information from our star network of IMUs.

After successfully collecting data, we have defined proper algorithms of machine learning that would help our system to recognize different bodily conditions and would able to differentiate between normal and abnormal activities and events.

We have to properly define states and conditions that is, in what conditions the personalized data about specific unwanted conditions has to be sent to the remote health care professionals.

Final Deliverable of the Project

HW/SW integrated system

Type of Industry

Medical , Health

Technologies

Artificial Intelligence(AI), Internet of Things (IoT), Wearables and Implantables

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com