Top-View based Human Action Recognition using Depth and Color Information

Sittisuk Seawpakorn and Nikom Suvonvorn


Action recognition, Color UV-disparity, RGB-D image


Human action recognition is an important step for human behavior analysis, which can apply to many applications, such as, surveillance system, and medical analysis tools. In this paper, we propose an alternative technique for human action recognition using color and depth image from top-view camera. The specific representation, called the Color UV-disparity, is proposed to solve the universal unsolved problem. The four basic actions are concerned: standing or walking, sitting, bending and laying. The neural network is applied as classification method. Our method can recognize actions accurately in real-time.

Important Links:

Go Back