Study And Implement On The Key Techniques Of The Human Shape Recognition

Human shape recognition technology refers to a kind of technology that utilizes human body imaging feature through disposing images and pictures to finally discover and locate targets with human shape characteristics in image spaces. Human shape recognition technology is an important technology in smart security systems. It is an mixture of computer vision, model recognition, image processing technology and morphology technology and is an important researching issue in smart security system. It can be used in smart monitor, smart transportation, target location, target tracking areas etc.Human shape achievement can be divided into several parts such as targets detection, contour abstraction, human shape targets matching and human shape recognition. Relevant realizing technologies and algorithms were discussed sequentially from achievement process of human shape recognition in this thesis. In targets detection part, several common used target detection algorithms such as inter-frame difference method, background difference method and optical flow field method were briefly narrated. Combined with research contents in this paper, each method's merits and shortages were pointed out. In the end, it was showed that background difference method was appropriate for the paper. In the part for target locating, a target locating method based on gray level with head and shoulder model was put forward. Experimental results showed this method had certain accuracy. In targets matching and recognition part, common methods for targets matching and recognition was discussed and a method to match and recognize target utilizing shape features was given. In target matching, human upper body contour model was given and then seek the vary law of curvature through analysing human upper body contour's features. In the end human upper body contour area was obtained through matching key dots that was got from analysis of curvature vary law. Based on human body shape matching, human upper body contour was dived into several parts and each selection's curvature was computed for statistics. Finally target's recognition rate was obtained.New algorithms that were put forward was emulated in MATLAB7.0 aiming at human's forward body's erect images. Experiment showed that new algorithms were effective.