Research On Moving Object Detection In Surveillance Videos

Three shortcomings in the classic VIBE algorithm may hinder its performance:The first frame containing the moving object will lead to a ’ghost ’ foreground object for subsequent detection;the fixed searching radius in the model will low its discrimination in complex scenes;and there will be holes in the detected foreground in complex scenes.This paper proposes three corresponding methods to overcome these shortcomings.(1)A ghost removal method based on the comparison of contour similarities for removing the ’ghost’ foreground This method compares the similarities of the contour extracted by Canny operator between static region and the corresponding gray region.It can distinguish whether the stationary region is a ’ghost’ region or a stationary object,and then take a different strategy on each.(2)An adaptive method for the searching radius based on LBP-T descriptor A new descriptor,LBP-T descriptor,is introduced to describe the degree of difference between the current pixel and the sample in its background model.LBP-T directly reflects the degree of change in the scene in the video and thus an adaptive searching radius can be obtained so that more foreground points can be detected when the degree of change is slightly in video.This method can also prevent the pixel with less fluctuation from being detected as the foreground point when its intensity varies significantly and consequently reduce the noise.(3)A novel hole-filling method based on super-pixel segmentation and saliency detection This method smartly uses morphological processing to greatly reduce the size of the detection region,and then computes the regions to fill by the saliency based similarity for the super-pixels.Finally,the exact region is obtained by fusing these regions with those from VIBE.Experimental results show that:The improved method based on contour similarity comparison can remove ’ghost’ with fewer frames;the adaptive searching radius based on the LBP-T descriptor makes the results more accurate;and the hole-filling algorithm can fill the holes well when there is a large hole.