SEGMENTATION OF MULTIPLE HUMAN OBJECTS IN VIDEO SEQUENCES

This article presents a method for segmentation of multiple human objects in a video stream. Given the base frame of a video stream, a fuzzy self-clustering technique is used to group similar pixels of the frame into a set of segments. Combinations of segments are checked if candidate face regions can be formed using chrominance values and the face shape feature. The existence of a face within each candidate face region is investigated by searching for possible constellations of eyes-mouth triangles and verifying each eyes-mouth combination based on the predefined template and a set of criteria. By referring to the position and orientation of a face, the corresponding body is found. Then rough foreground and background are formed. Finally, human objects in the base frame and the remaining frames of the video stream are located by a fuzzy neural network, which is trained by a SVD-based hybrid learning algorithm. From experimental results, we have found that our system can locate human objects more accurately and efficiently than other systems.