Real Time Image based Human Activities Recognition

Project Title

Real Time Image based Human Activities Recognition

Project Area of Specialization Artificial IntelligenceProject Summary

This project aims to recognize human activities from series of images considering different human actions performed in an indoor environment. This Image based  Human Activity recognition(HAR) system can provide assistance for elderly and disabled peoples in emergency, law-enforcement using video surveillance ,human computer interaction and efficient resource utilization. It is more pertinent in today busiest world that most of the elderly people stay longer and longer alone in the home so this system can be utilized for human surveillance and health caring. Efficient resource utilization is one of the major concerns in this modern world so depending upon the human activity recognition in an indoor environment like rooms and offices, energy appliances can be efficiently controlled or monitored thus avoiding the wastage of energy resources caused by unattended appliances. Activities recognized through this system can be used for law enforcement in places like offices and banks to prevent criminal activities. We propose, an RGB and joints based human activities recognition using convolutional neural network (CNN). The proposed system will be implemented in a distributed computing paradigm. First subsystem includes acquisitioning of RGB image with joints information, detection and localization of human in the image, and network infrastructure to send the image for recognizing the activities. The second subsystem is activities recognition server using the CNN trained with 40K images. This projects aims to implement the system with minimum recognition (computing) time and accuracy more than 90 percent.

Project Objectives

• Design and development of a human activity recognition system using image and joint information.
• Person localization in an image using Feature based approach.
• Real time training model for human activities recognition.
• Acquisitioning of at least 40k data set consisted of images with basic human activities.
• Raspberry PI based hardware system for person localization and Region of interest extraction.

Project Implementation Method

In this project we developed an algorithm for real time human activities recognition using images obtained from sensor. First the acquired image is pre-processed to enhance the quality of information. After this, person localization is performed using feature based approach and region of interest (ROI) is extracted. This ROI will be transmitted to recognition server using Wi-Fi where a real time training model will recognize it using the acquired dataset. The proposed  architecture of the proposed recognition system is as follows:

The Flow Chart of the proposed system is as below:

Benefits of the Project

• This human activities recognition system can be used for providing assisted living technologies for older and disabled peoples to help them in their daily life.
• Moreover activities recognition of criminals achieved using this system can help law enforcement agencies.
• Efficient resources consumption such as lights and fan control can be performed depending upon the results of human activities recognized using this system.
• This system can also be used as Human-Computer-Interactive technology that can be utilized in gaming, and teaching to special children.

Technical Details of Final Deliverable

• Efficient image acquisition within the range of up to 6m with angular field of view of 57 degree horizontally and 43 degree vertically.
• Overcoming challenges like contrast, resolution, background clutter etc. using pre-processing and feature extraction.
• Software for human activity recognition with accuracy greater than 90percent.
• Raspberry PI based platform for transmission and recognition of region of interest (ROI).
• At least publication of one paper.

Final Deliverable of the Project HW/SW integrated systemType of Industry IT , Energy , Health , Security Technologies Artificial Intelligence(AI), Internet of Things (IoT)Sustainable Development Goals Good Health and Well-Being for People, Sustainable Cities and Communities, Responsible Consumption and ProductionRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	62000
Raspberry PI Kit	Equipment	1	15000	15000
Kinect Sensor	Equipment	1	23000	23000
Arduino board	Equipment	2	1000	2000
Smart screen	Equipment	1	12000	12000
Thesis printing and poster	Miscellaneous	1	10000	10000