Research On Video Processing Technique In Intelligent Transportation Systems

In recent years, with the increase of vehicle possession, Intelligent Transportation Systems (ITS) have been of major importance for enforcing traffic management policies and investigated by many researchers. Aiming at the advanced traffic management systems for online surveillance and detailed information gathering on traffic conditions, techniques from image analysis and computer vision can be applied to traffic video analysis, such as vehicle counting, vehicle classification and queue detection. It is well recognized that vision-based surveillance system are more versatile for traffic parameter estimation than the others. However, these techniques often suffer from the failing segmentation or detecting error. In the last several years, an extensive research work has been done and many vehicles monitoring systems have been exploited. Based on the past research productions, this paper presents a robust and real-time method for detecting vehicles from a sequence of traffic images taken by a single roadside mounted camera, tracking vehicles over the duration, in which vehicles appear within the field of view, and classifying the vehicle. The proposed algorithm includes three stages: object region extraction, vehicle tracking and vehicle classification.The proposed algorithm of object region extraction includes three steps: first, extract moving object region from the current input image by background subtraction method, second, eliminate moving cast shadow which is often caused by moving vehicle and, at last, detect vehicle so that there can be a unique object associated with each vehicle.Based on the segmented vehicle shape, which can be represented by a simple square model, this paper propose a three-step method for vehicle tracking: first, extract vehicle state features from each square model and predict the future state of each vehicle in the next frame through Kalman filter, second, match features between the predicted data and the new vehicle state so that we can obtain tracking result, at last, update the innovation and calculate the new Kalman gain according to the prediction error power in order to implement the next prediction recursively. At last, we classify the type of the vehicles throw Support Vector Machine.In the above processing, we advanced a arithmetic to found original background by performing a difference on three consecutive inter-frames and eliminating moving object region, presented a real-time background renewal method based on statistic strategy, and developed a effective shadow suppression way according to the common feature of moving cast shadow, especially the comparability between the moving cast shadow and the background.Then, after studied a variety of vehicle features, we presented a projection segmenting method based on the sub-features point to verify vehicle. This method adopt the position information and colour information of vehicle sub-features point for vehicle tracking based on the Kalman filter. In this paper, we selected vehicle corner as the sub-features point. As the corner is often random and legion, these points need to be pretreated. A part of these corners would be united through dilation arithmetic operators, and then we can use the thinning processing to extracts the sub-features points from these blocks. Comparing with other arithmetics, our method is accurater for vehicle tracking.Aiming at vehicle classifying, we select the structure moment as the classifying feature and solve the problem with a high degree of accuracy throw Support Vector Machine.Our method has been tested on a number of monocular traffic-image sequences and the experimental results on the real-world videos show that the algorithm is effective and real-time.