Human Posture Detection

This is a Research based project. The purpose of this project is to develop a generic method for detecting global human body position in video sequences. A posture can be defined as a position of one?s limbs while carrying out an activity such as walking and sleeping etc. Before building the working

Project Title

Human Posture Detection

Project Area of Specialization

Software Engineering

Project Summary

This is a Research based project. The purpose of this project is to develop a generic method for detecting global human body position in video sequences. A posture can be defined as a position of one’s limbs while carrying out an activity such as walking and sleeping etc. Before building the working project, we will perform a thorough research on the previously made systems. Human Posture detection and recognition is essential to any automated video surveillance system. This could be used on a large scale or just to keep track of someone’s demeanor and movements in places like hospitals etc. For this system to work, initially, we will work in an obstacle free room, an empty hall, with the assumption that only one person is present at the time. In order to keep the track of movements, a video camera will be used. 

Project Objectives

The main objective of this Research is to develop a system that can detect Human posture with maximum accuracy. We will run different Datasets including KARD, UCF Crime, UP Fall, URFD and MCF on multiple algorithms such as Resnet-50, Resnet-101, Resnet-152, Inception V3 etc. and check for the best accuracy.

Project Implementation Method

The main goal of this research is to develop a system using deep learning methodologies to detect Human Postures. We read multiple research papers to find the different approaches used to build the system. The most efficient architectures we found includes: ResNet-50, ResNet-151, ResNet-152, Inception-v3 etc.

We collected datasets from research papers based on their availability online. We found a couple of datasets including MCF, KARD, and UCF. We had the complete information of the KARD dataset so we started with it. We manually captured the pictures from the given videos. Furthermore, we tested and trained the data mainly using Python language and using the Google Colab.

We ran KARD on the different architectures and found the accuracy of it on different architectures. The results are given the Table 3. Consequently, the best accuracy was achieved on the ResNet-50 architecture.

We will be using pixel by pixel approach using deep learning methodologies. A video camera and previously given datasets will be used. A picture has 3 axis i.e. x, y and z.

Benefits of the Project

Posture Detection has many applications in various fields. A few them are stated below:

• Supportive home environments
• Care centers for the disables
• criminal investigations
• Physiotherapy
• Surveillance Cameras
• Yoga
• Gym

Technical Details of Final Deliverable

The system will consist of a video camera with raspberry pi working in the background. 

Final Deliverable of the Project

HW/SW integrated system

Core Industry

IT

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Good Health and Well-Being for People, Quality Education

Required Resources

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