| The research for traffic image detection and process methods, with important theoretical meaning and wide practical value for traffic management and traffic control methods, has become a focus of attention. With the help of video detection and recognition methods, we can sample real-timely traffic messages, such as traffic flow, road situation and so on, and get the relative information so as to avoid traffic jam before it happens, On the other hand, to real-time traffic breach detection, video detection and recognition methods is also very effective, especially, it can provide strong evidence for traffic policemen. In this paper, considerable research on image detection and process techniques have been done and so an effective, simple and fast algorithm have been proposed. The most distinctive parts would be described in the following aspects:1. Because the contact traffic detection device only can measure little information and be easy to fail, we propose a new traffic flow detection algorithm using virtual trapline. The method is better than others. Firstly, the system sets up several models to assist detection work, such as road model, illumination model, shadow model and relationship model among them; Secondly, it introduces hue . saturation degree information to distinguish vehicles from shadow; Thirdly, it makes use of binary decision tree to analyze pressing line of vehicles to improve the reliability of the system; Fourthly, it puts forward a way of one dimension video tracing to resolve the problem of vehicle velocity detection. The results indicate that our method is indeed better than existing methods in accuracy rate and real-time performance.2. A new multi-object vehicle tracking algorithm suited for traffic surveillance is presented based on two kinds of object tracking algorithms(region tracking and corner tracking). Firstly, forming the whole feather of vehicles by means of region tracking algorithm, then delivering the information to corners so as to build group relationship among corners, and then accomplishing the creation of new objects, Secondly, achieving the old object tracking with multiple-dimensioned corner matching, then feeding back all sorts of information of moving corners to region tracking algorithm to promote the separation of "barrier" among objects in order to gain the tracking of the whole motive objects.3. An algorithm based on mathematic morphology and fuzzy evaluation functions is put forward used for license plate location on line. The method is different fromothers, not a single feature or a group separated features but mathematic morphology and fuzzy evaluation functions are used as the carrier of multi-features combination. This would improve the ability of anti-interference and accuracy of location. On the other hand, mathematic morphology operation method based on point is explicated to form a new method based on line. Combining with recognition based on object, it would increase accuracy of location and real-time nature,4. At the stage of preprocessing, a layer-by-layer overburdening algorithm is proposed to locate rightly the bound of area where the characters exist for avoiding license plate frame, rivet and car body 's effect on character segmentation. The other side, a good result can be achieved by applying fuzzy theory and genetic algorithms to character slitting. Besides, taking into account the asymmetry sunshine, deferent gray and other factors' effect on license plate, binarization method is modified, a special local binarization method is presented for car plate image.5. At the stage of recognition of license plate character, in order to improve the accuracy of recognition, a new way is invented for it. Firstly, the system is divided into four sub systems: Chinese characters recognition system, character recognition system, recognition system including characters and digits, and digital recognition system; Secondly, the statistic information about left (right) scan line free length of every character is used as a group feature...
|
PDF Full Text Request