Moving Human Detection Based On The GMM Of Scene Partition And HOG Of Background Replacement

Aiming at the problem that the traditional moving object detection method can not effectively identify pedestrian targets,a moving pedestrian detection method based on Gaussian mixture model moving target detection with scene partition and HOG pedestrian detection with background substitution is proposed.A randomized updating Gaussian mixture model method based on scene partition is proposed.The computation of the algorithm in the stable region of the background is reduced by scene area division and the detection accuracy of the algorithm under complex background disturbance and slow moving target is improved through the random update mechanism.Background replacement HOG pedestrian detection method based on Kalman filtering is proposed.The algorithm provides a more complete window for detection through Kalman filtering,and the pedestrian detection accuracy is improved by detecting the same object under different background by background replacement.Aiming at the non-use of scene information in Gaussian mixture model,the scene is divided into the background stable region and the background perturbation region by the scene division method according to the weight distribution characteristics of the pixel model.In the background stable region,the computation of Gaussian mixture model is reduced;maintain or increase the computation of Gaussian mixture model in the background disturbance region,and then the algorithm redundancy is reduced with the detection accuracy guaranteed.The Kalman filtering is used to avoid incomplete detection range of subsequent HOG pedestrian detection caused by erroneous moving object detection.According to the pedestrian regions of detected sequence frames,the target area of the current frame is predicted by Kalman filtering.This prediction,together with the current frame moving target detection results,provides a window to be detected for HOG pedestrian detection.In order to improve the detection accuracy of HOG pedestrian detection in the background disturbance area,the background is replaced by synthetic image.The Gauss distribution peak background disturbance pixel background threshold region within the random combination,generating a plurality of local background,as the first part of the synthesis of the image pixels in the current frame to the foreground detection value as the second part of image synthesis,eventually forming a plurality of image to be detected,thereby reducing the false alarm rate of HOG.Experimental results show that the improved hybrid Gauss motion detection method and the improved HOG pedestrian detection method have better performance compared with the original algorithm.The proposed moving pedestrian detection method performs well in all experimental scenarios,and has theoretical and practical value.