View invariant real-time gesture recognition

The invention of several machine assisted systems has increased the scope of security in the society. Human activity recognition is one such very useful human computer interactive system that has gained the attention of the researchers for decades. By identifying and recognizing different actions performed by people, proper surveillance of different criminal activities or activities of elder people staying in the home alone can be performed. This paper as a whole contributes to this cause by developing a robust technique to identify the activities performed by person in a room. This technique provides a solution to most of the previous camera conscious approaches, by generating a system prototype that is view-invariant. For this purpose multiple Kinect sensors have been used. The proposed system is fast and accurate under different illumination conditions and the activity is recognized instead of the fact whether the person is looking at any camera or not.