Design And Implementation Of Pedestrian Detection Module Under Dynamic Background

With the development of information technology,the amount of video files has rapidly increased.It is of great significance and widely applicable value to research on how to retrieve the key events and targets in a large number of videos,especially in traffic monitoring and security surveillance videos with much redundant information.However,how to detect targets and pedestrians quickly and accurately in the video with dynamic background is still an urgent problem in the field of moving object detection.Optical flow method is an effective method to detect moving objects and pedestrians.It does not need any priori knowledge and does not depend on the background model.So,it has been widely applied in the field of moving object detection.However,the original optical flow method has defects such as be sensitive to noise and high algorithm complexity.Aiming at these problems,an improved optical flow method is proposed and applied to detect moving targets in this dissertation.Firstly,aiming at the global dynamic background caused by the camera motion,an improved gray projection algorithm is used to compensate the dynamic background.Aiming at the defect of accumulative error existing in gray projection algorithm,the dissertation proposes replacing the reference frame every other three frames to reduce the accumulative error which is caused by selecting the fixed frame as reference frame;Aiming at the defect that calculation error caused by moving objects existing in the projection area is relative major,the dissertation proposes selecting the edges of video frames as the projection area to calculate the motion vector,this selection method of projection area greatly reduced the influence of moving targets on motion compensation.Secondly,the dissertation proposes an improved optical flow method based on gradient threshold and feature suppression and applies it to moving object—?—detection.In this dissertation,the LK optical flow method is combined with HS optical flow method to improve the optical flow constraint equation.The brightness constancy constraint is used for larger gradient points,and the global smoothness constraint is used for smaller gradient points to ensure the applicability of the optical flow constraint equation.As a method to judge the effective optical flow points,feature suppression is used to suppress the noise and the local dynamic background.Finally,on the basis of moving target detection algorithm,the improved shape complexity characteristic is adopted to categorize the moving targets in the videos.Through comparing the value range of shape complexity of different types of object,the threshold which could distinguish pedestrian from other types of moving targets is determined.According to the threshold,the pedestrian can be detected from videos effectively.The test results show that the proposed pedestrian detection method under dynamic background can obviously improve the accuracy and stability compared with other methods.In addition,the method proposed in this dissertation is robust to camera motion and random dynamic background,which provides a good foundation for the next research work.