A Research On The Algorithms Of Vehicle Information Extraction In Its And The Related Hardware Implement

In the recent years, ITS (Intelligent Traffic System) has become the focueproblem along with the increasing demand for the traffic. ITS which consists of thetechnologies of digital image processing, pattern recognition, intelligent control andso on can solve many problems in the traffic system, but it has some defects in theaccuracy and robust characters.The research of the paper is mainly on two aspects: One is using thetechnologies of digital image processing, pattern recognition, computer vision, andinformation analysis to detect and track the vehicles on roads. The final purpose is toextract the information of vehicles for traffic management, including the total numberof vehicles, the velocity and type of each vehicle and so on. This paper mainlyfocuses on improving the performance and algorithms in real-time and robustness toadapt to various traffic environments. The other aspect is transplanting and realizingthe system to the platform based on TMS320DM642.In the aspect of algorithm research, the extraction of vehicle information can notbe implemented by only one algorithm. It should be realized by a system. The paperdoes not only consider the algorithms of vehicle detection and tracking but also otheralgorithms which could do assistance to extract the vehicle information in a real-timeand robust manner. Therefore, the system can be divided into several pans: electronicimage stabilization, background updating, image calibration, vehicle detection andvehicle tracking. This paper mainly puts emphasis on the following three parts:Electronic Image Stabilization: Compared with the human vision, the computervision is more sensitive to the image's stabilization. Even little oscillation will affectthe judgment of computer vision. The algorithm of electronic image stabilization isused to minimize the influence caused by the camera shaking.Vehicle Detection: After the background updating and image segmentation, themoving objects which exist as blobs in binary image can be extracted. The vehicledetection includes blob analysis, blob clustering and new vehicle detection. The totalnumber of vehicles and the type of each vehicle can be obtained in this step. Vehicle Tracking: With the position and color information in image and realcoordinates, we use the algorithm of Kalman Filter to predict the position of eachvehicle among the video frame sequences, and then track the vehicles in several wayswhatever the number of vehicles is. In this step, the velocity of each vehicle can beextracted.As shown by the experiment result, the system can adapt to different trafficenvironments, such as the highway and city roads. The extraction of vehicleinformation can be implemented in a real-time manner and achieve high accuracy.In the aspect of hardware implementation, the algorithms are transplanted to theplatform based on TMS320DM642 and the system is realized in the embeddedhardware system stably and quickly by using various technologies of codeoptimization.