Image Edge Detetion Algorithm And Its Application In Traffic Video Segmentation

Edge is one of the most important features of image, it is one of the basicpre-processing methods in digital image processing. Until now, more edge detectionmethods are proposed. Cellular neural networks have unique advantages in terms ofapplications and hardware implementation, so that it becomes a research hotspot inedge detection in recent years. Motion information is the basic of video objectsegmentation, edges can provide good information of the image motion information.Making good use of edge information can get better video segmentation resultsIn general, the main contributions of this paper can be described as follows:①There are such problems as gray image edge detection could not detect theedge detail, it proposes an edge detection method based on CNN and PSO. The mostimportant essential point of CNN application is to find a set of accurate templates, inorder to slove this problem, it employs linear matrix inequality and particle swarmoptimization (PSO) to carry out the optimization parameters of cellular neuralnetworks (CNN). Simulation results indicate that the proposed method introduced inthis paper offers superior performance in both binary and gray images edge detection.②There are such problems as gray image edge detection could not detect theedge if edge exists in the region that has same brightness but different color andtraditional color image edge detection could not detect fine edge. In order to solvethese problems, the paper proposes a multi-channel edge detection method based onLab color space. This paper employs linear matrix inequality and particle swarmoptimization based on cellular neural networks to do edge detect to L channel; andmeanwhile employs Lab color difference function and improved sobel algorithm to doedge detection to a and b channel; finally blending the images to get accurate imageedge. Experiment results showed that the method proposed by the paper could fullydetect the color image edge information, enhance edge detection accuracy andcontinuity, and it could obtain better effects on the region that has rich details andsmall changes.③There are such problems as it’s difficult to set a Threshold when use Framedifference method to do video object segmentation, a spatial-temporal segmentationalgorithm for video object segmentation is proposed. Firstly, a method setting theadaptive threshold is utilized for the problem that the binary threshold is difficult toset in the differential image by means of frame difference. Secondly, the differential templates are filtered through noise filtering method. Thirdly, the more precisely edgeof the moving image is obtained by combining the detection results and thedifferential templates. Finally, the target template is obtained by area filling, and thenthe moving target image is detected. The experiment results show that the movingobject can be detected from the video sequence real-time effectively by the method.The algorithm can be simulated successfully by using matlab developmentplatform based on Windows environment in this paper.