Research On Segmentation Of Overlapped Vehicles In The Traffic Video

Intelligent Transportation Systems has become a hot topic for researchers home and broad due to its intelligent management to traffic.How to quickly get traffic parameters is crucial to the whole ITS,and vehicle detection is the essential and critical link.However,in traffic monitoring system,the traffic video image which is captured by the road camera may appear the overlapped phenomenon between target vehicles due to camera angels etc.,which can affect vehicle extraction,identification,tracking and other following works.Hence,this paper focuses on vehicle extraction,detection and segmentation of overlapped vehicles.First of all,this paper adopts fixed window method based on histogram statistics to obtain background,which establishes a fixed window for each pixel to take the most frequency gray value in continual frames as its background gray value.For the shadow and overlapped vehicles caused by shadow,the paper adopts a method based on OSTU double threshold grayscale for solving the problem,which uses OSTU double threshold method to segment target vehicles into different areas and enhance them by different gray levels,then shadow can be eliminated by using background difference algorithm to obtain target vehicles.Next,for detection and segmentation of overlapped vehicle,this paper proposes an algorithm which based on contour feature point.(1)The detection of overlapped vehicles.Firstly,detects the feature points of target vehicle by Freeman chain code,and gets its minimum enclosing rectangle;secondly,according to the number of feature points and the occupancy ratio of target area,it can judge whether the vehicles being overlapped.(2)The segmentation of overlapped vehicles.Generally speaking,the segmentation points are in the recessed areas of overlapped vehicles,so this paper firstly carries out convex hull analysis for the target area to find convex sequence;Secondly,according to the convex sequence,calculates recessed span and depth to obtain recessed areas of target area;Finally,the optimal segmentation points can be found bycalculating the distance between feature points and convex vertex,and then overlapped vehicles can be segmented.Compared with related methods based on concavity analysis and ellipse fitting,this method can segment the overlapped vehicles more precisely and have better adaptability without any prior knowledge besides the vehicle shape.