Vision Based Human Motion Analysis for Aquatic Environment

Description

Analysis of human motion video captured in aquatic environments is a relatively new research topic. The detection of humans, tracking and recognition of human motion are challenging problems. The water surface and splashes exhibit spatially and temporally varying specular reflections in the video frames. Background modeling/subtraction, which is a common first step to detect humans, is very difficult. Motion tracking is subject to the problems of occlusion and limited camera field of view.The researchers propose to develop a computer vision system for analyzing swimmer motion. Firstly, to make a robust swimmer detection method, they propose a novel specular reflection removal algorithm for pre-processing the video frames. The accuracy and range of motion tracking are enhanced by the use of multi-view videos and Kalman filter. The motion cues are derived via the skeletonization of the target regions. Finally, probabilistic inference schemes are designed for the classification/ recognition of specific swimming styles.