The Research On Superpixel Based Moving Object And Shadow Detection

Moving object detection is an important research direction in the field of pattern recognition,and it is also one of the core technologies of modern intelligent monitoring systems.Current moving object detection algorithms have many shortcomings in eliminating false alarms generated by dynamic background,filling holes inside the moving objects,removing the ghost,and diminishing the false positives caused by shadows.These challenges limit the application of the algorithms.This paper analyzes the multiple challenges encountered by the moving object detection methods in complex scenes.Represented by ViBe algorithm,this paper presents solutions for the above problems.The main contributions of this paper are given as following.(1)An improved ViBe algorithm based on superpixel information feedback is proposed.Firstly,it divides the original image into several superpixel regions,in which the object edges are preserved.Based on the information of superpixel edges and the difference between different superpixels,a mapping is established between real objects in ViBe mask and their superpixels.In this way,true object regions are recognized and ghost regions are isolated.Moreover,by classifying the superpixels,the holes in object superpixels are accurately filled and the false alarms caused by high-frequency disturbance in background superpixels are eliminated.Finally,the experimental results show that the proposed method performs well on both the speed of ghost removal and the object detection accuracy.(2)A regional radiation consistency based moving shadow PCC detection method is proposed.Firstly,the current image is segmented into several regions and points collection in contour(PCC)of each region is established.It assumes that the intensity of each pixel obeys Gaussian distribution,based on which the Regional Radiation Consistency property in Shadow(RRCS)is proposed.That is to say,in terms of the points in shadow PCCs,their intensity mean in current image is reduced compared with that in background image,whereas the standard deviation changes slightly.Combined with the intensity reduction caused by shadow and terminal weight of PCC,all PCCs are classified into the object PCCs and the shadow PCCs.(3)A PCC mask seeds based region growing method and reverse foreground object mask growing technique are proposed.Based on certain growing conditions,shadow PCCs grow into complete shadow areas.Moreover,by fusing grown object PCCs with object parts in ViBe mask,a relatively complete object mask image can be obtained.Finally,the experimental result shows that the precision and recognition rate of shadow detection are remarkably high,while also the grown part of object mask achieves high accuracy.