Moving Human Body Detection And Face Recognition In Video Images

This dissertation describes our investigation on some algorithm of computer vision. Research fruits mainly include segmentation algorithm and precise outline segmentation algorithm of moving objects from sequential images, detection algorithm of moving human body from un-crowding scene, faces detection algorithm from video sequences, face recognition under different expressions and multi-views etc. The motivation of this dissertation is how to find simple and effective algorithms, so as to combine the moving body detection, face detection and recognition as a series problem. The contributions of this dissertation are listed below.Aiming at existing defects of traditional segmentation by temporal differencing method, such as sensitive to noise, sensitive to the dynamic background etc, we suggest a new moving object segmentation algorithm based on eigenobject. Eigenobject is computed by improved KL transform. It not only contains the location information and the shape information of the moving objects, but also corresponds to the middle frame of the original image sequence. The moving object can be extracted by using eigenobject. In this step, three frames are enough for rigid and nonrigid object. Furthermore, a moving object outline precise extracting algorithm is designed based on a KLT/Snake model. The experimental results show that, the algorithm proposed in the paper has a high speed and a definite meaning; it can detect and segment the moving objects accurately from complex background.On the basis of studying prior work, a human body detection algorithm based on the combination of temporal information and shape information is designed. Firstly, moving objects are detected using the proposed segmentation algorithm. Secondly, shape information is used to distinguish human body and other moving object. Furthermore, occlusion during a short time is handled by detecting the shape of moving object in continues frames. The algorithm has the advantages of accurately moving object detection and the detection result don't effect by the body pose. Experiment results show that the algorithm is robust to noise and can detect the human bodies under complex circumstance.A face detection algorithm combined moving information with edge projection function is proposed. Aiming at existing defaults of traditional Sobel operator such as sensitive to noise and the detected edges are to thick, a two threshold Sobel operator is...