Research On The Vehicle Image Segmentation Algorithm Based On Concave Points Matching And Watershed Transformation

In the vedio monitoring system, the vehicles need to be counted at the intersection accurately. And the number of vehicles is applied to the intelligent system to timely control the traffic light transformation, so that the traffic rate at the intersection can be raised. In addition, in order to manage parking spaces efficiently, it is also necessary to count the vehicles precisely in those large parking lots. However, in the traffic scene, the vehicle overlapping happen frequently since the cameras are installed inappropriately or the distance between vehicles is too short. Since the adhesion question between vehicles has a serious bad effect on the counting of subsequent vehicles and feature extraction, so the research of segmentation algorithms of the overlapped vehicle images has very important practical value.Generally, the vehicle images obtained from the nature scene not only exist those adherent questions but also accompanied with noises, low dynamic ranges and interference foregrounds(the foreground pixels closer to gray value of the vehicle). Therefore, it is quite necessary that preprocessing procedures was done before the segmentation of the overlapped vehicle images, such as enhancing the contrast of images and removing noise interferences.For adhesive vehicle problem, two different adhesion segmentation algorithms are representated in this paper. Firstly, a watershed algorithm based on distance map reconstruction is studied in this paper. The innovative point of this algorithm is that some seed points the watershed segmentation algorithm need are optimized. And on the base of the optimized seed points the distance image is reconstructed. Then apply the watershed algorithm dealing with the reconstructed distance map to get the segmented images. The proposed algorithm can not only separate the adhesion vehicle well, but also effectively suppress the over-segmentation. Secondly, this paper also studied another segmentation algorithm based on concave points matching and the core of this algorithm is a transform of operation. To begin with, use the segmentation algorithm to obtain concave area of the overlapped vehicle image, and then get the concave point setted by shortest distance between the point of concave area outline and the given line. In the last step, select the best dividing line for adherent vehicle division. The segmentation algorithm can effectively and accurately separate the adherent vehicles, and the calculation procession is relatively simple. In addition, in order to automatically get the counting of vehicles, a label counting algorithm is used in this paper.