|
With the development of modern intelligent, networked, and intuitive requires, intelligent vehicle research has become a hot research topic in the field of computer vision. At the same time, intelligent vehicle in many fields have a wide range of applications, such as military, traffic. Moving target detection and tracking is the basis and the key problems of the intelligent system.This thesis focuses on the intelligent vehicle driving navigation technology in vehicle detection and tracking technology research field of computer vision, around some of the algorithm to achieve the technique research. The thesis includes four parts: the research background, the classical algorithm, vehicle detection module design and tracking module design. Through the comparison of the advantages and disadvantages of various algorithms, this paper finally selected as the main optical flow detection algorithm, combined with the feature points, the Pyramid model and the region of interest on vehicle detection, after the prospect of vehicle segmentation, vehicle contour, contour tracking of vehicles.First of all, this paper analyses the traffic scene, on different traffic scene using different assistive technology. Under the simple fixed scene pretreatment, feature point extraction is relatively simple, analysis can obtain good the optical flow motion vector by clustering and denoising; Dynamic background processing is much more complex, it is difficult to distinguish which is the effective motion between background and foreground, the extraction of characteristic points from image become difficult, and interference is complex, too. Based on the method of region of interest and template, we can get much better feature points by means of my papers’algorithm, then using the optical flow to detect the moving vehicles.Secondly, this thesis put to use the background modeling by Gaussian model to segment image for better detection results of pretreatment. After obtaining the Gauss prospect, using contour convex function extract vehicle contour, then the contour can take along many useful information. Contour convex hull added a section foreground pixels for differential image and Gauss modeling foreground image lost, making the vehicle outline complete. At the same time, setting the perimeter and area of contours can removed a portion of the background pixels which are mistaken as foreground pixels. In this way, we can optimize the results of segmentation.Finally, basised on the result of detection by optical flow field, background Gaussian modeling and mathematical morphology processing, we use the profile of the vehicle to track interested vehicle contour by using CamShift algorithm. The results are very well through practice, the algorithm that this paper studies can well track the target vehicle.At present, the scholars of this research filed increasingly grow in quantity, the technology is used more and more widely.
|