Video Image Target Detection And Recognition Algorithm

As an important part of the field of computer vision, video image objects detection and recognition has been the general attention of the people in recent years as well as widely applied. In this thesis, moving objects detection and pedestrian recognition algorithm is deeply studied. A more accurate moving objects detection method is proposed. And a more optimizational method for HOG descriptor, which is most commonly used in pedestrian recognition field, is given.Firstly the thesis summarizes the status of the development of video image objects detection and recognition fields at home and abroad.Secondly, this thesis proposes an algorithm which can effectively resolve the problems of disturbances caused by noise and illumination mutation. Edge Gaussian mixture model (EGMM) method is combined with an improved Neighborhood-based difference method in this paper to solve these problems and improve the moving objects detection.Thirdly, HOG descriptor is deeply studied in this thesis. By doing dimensionality reduction process, time of classifier training and judgment can be effectively reduced. And experiments proved that the speed advantages of the improved method.Finally, algorithm of moving objects detection part is optimized and transplanted to the DSP.