الفهرس | Only 14 pages are availabe for public view |
Abstract The proposed work is a human action recognition system that relies on the amount and shape of change of different body parts to recognize a given action in a recorded video. Newly proposed features are extracted and encoded to describe the visual way of change of human body parts. The first step in the technique is skeleton extraction for the subject person. Then, novel features are extracted from this skeleton and encoded to obtain a limited short length code that represents the whole video. Training and testing step where performed using benchmark datasets, namely: KTH, Weizmann, Berkeley, MSR-action3D |