REALTIME MULTI HUMAN TRACKING AND ACTION DETECTION

Project Title

REALTIME MULTI HUMAN TRACKING AND ACTION DETECTION

Project Area of Specialization Artificial IntelligenceProject Summary

We focus on the development of intelligent surveillance systems that aim at interpreting human activity and identify or detect their actions, instead of using a passive monitoring system, which is the most commonly employed technology. However, in general, action means a simple pattern of human movement – such as walking or taking steps, running, sitting and/or standing – and can afterwards be used to infer a complex task that involves the identification of several actions, and interactions with objects in the scene.

Project Objectives

Our objective is to develop a system where a camera is used to detect and track multiple people and identify the actions performed. We aim to detect and track one or two people initially which can be extended further. Project objectibve is:

• a system that is initially capable of detecting one or two people.
• a system that will be keeping continous track of those people after detection.
• a system that will be capable of identifying different actions performed by them such as walking, sitting, picking, running etc.

Project Implementation Method

An object detection algorithm such as posenet is applied on the input video sequence, then from joints coordinates positions, different body parameters such as slopes and distances for ankles, legs, wrists and arms are computed, and then these are sent for training CSV file, a model is obtained and then tested.

Benefits of the Project

Today’s increase in threats to the security in cities and towns around the world makes the use of video cameras to monitor people necessary. The attacks on humans, thefts, fights, vandalism, and traffic issues are just some cases where action recognition systems are needed. An algorithm for human motion detection digests a video into a compact description of the human presence in that scene. This description can then be put to use in enormous other applications. Video based human action recognition has many applications such as human-computer interaction, smart home health-care systems, old-age/infant monitoring, surveillance systems etc

Technical Details of Final Deliverable

last deliverables referred to the completion of:

• human joints position identification
• legs,wrists,arms, hips, ankles slope computations and  various body parts distance computation
• angle between arms and wrists, angle between legs and thighs computation
• CSV training and testing
• trials on mode, priority, voting schemes for different actions

Final Deliverable of the Project HW/SW integrated systemType of Industry Education , Health , Security Technologies OthersSustainable Development Goals Peace and Justice Strong InstitutionsRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	7500
Camera	Equipment	1	5000	5000
Router	Equipment	1	2500	2500