Research Of Rapid Image Segmentation For Relative Movement Objects On Grade Road

The work in this paper is an important part of the project research of recognitionmethod for scenery of grade road in fog that is composed of image filtering, objectsegmentation, camera calibration and object tracking. This project involves patternrecognition, artificial intelligence and computer vision techniques.This paper is mainly about the research of rapid image segmentation for trackingof moving objects. Objects to be segmentated are cars and lines of driveway. Thecharacteristics of this problem are that both objects and background are moving,background is very complicated and real-time processing is highly demanded.First, effectiveness of conventional segmentation methods, such as thresholding,region growing, edge detection, are discussed. Then corresponding algorithms areimproved and the obtained results are listed below:1. Adaptive threshold is used to process image and empirical relations are got.This algorithm can adapt to the change of light intensity outside and extract the linesof driveway. Not only the process is rapid, but also the veracity of segmentation ishigh.2. The images in this paper are segmented with improved region growing method.Projection-histogram is used to generate seeds. Moreover, the rule of growth is alsoimproved. The result of segmentation is satisfactory. Objects, cars and lines of thedriveway, are segmented from background and the speed of this algorithm is veryrapid.3. The region of interest is established. In this region, vehicles are segmented bythresholding. Thus, the process time is saved. At the same time, accuracy ofsegmentation is ensured. This result is fused with the image of driveway linessegmented by adaptive threshold, and the final output can be obtained, which reachesthe demand of this segmentation task.Second, the image is segmented with curve evolution. The improved algorithmof C-V partial differential function is carried out with level set and applied to thissegmentation task. The curve has evolved, but the speed of convergence is too slow.This algorithm of curve evolution is tested with simple background image later. Andthe result shows that the curve evolves towards the edges of objects and stops at theedges of objects at last. It shows segmentation of objects is successful with thisalgorithm.Anyway, what this paper does is just elementary research. Both traditionalsegmentation and curve evolution need to be further developed.