Human Identification Using Motion Vector Analysis

The field of human recognition has existed for past few decades, but gained popularity in recent years for its use in avariety of applications. In security industry, suspicious persons/activities could be detected in high-profile areas. In medical industry, systems could be trained to detect patterns of motion indicating painOR to detect a lack of motion if a person has fallen and unable to move. However, algorithms with reliable accuracy are difficult to implement in a real-time environment due to computational complexity and ambiguous decisions.

This work will develop an efficient way of extracting and using data from a walking style of human in a video frame to identify the human. Following background subtraction, a thinning algorithm offered a more robust feature such as motion vector extraction method. Training and testing approach which is a basic Neural network approach/classifiers used to identify a number of activities, such as walking, running, waving and jumping. This entire human activity recognition system will be tested with a MATLAB implementation. The algorithm will be developed to achieve maximum classification accuracy in video feeds.