Forward Vehicle Detection And Tracking Based On Video Images

With the development of social economy and the increasing number of car ownership, the problems of traffic congestion, traffic jams and traffic pollution become more and more serious. And Intelligent Transportation System (ITS) is a traffic control system, which comprehensively applies the advanced computer processing technology, the information technology, the data communication transmission technology and the electronic automatic control technology to traffic management system. It is also a running orderly system combined with the human and the vehicle. It plays a very significant role in improving transportation efficiency and effectiveness, and ensuring the security. Video vehicle which acts as a part of the Intelligent Transportation System is widely used in assisting safe driving, autonomous navigation, traffic monitoring and many other fields. So it has vital functions in improving the safety of the vehicle driving.In this thesis, our research objective is video image which get from camera installed in the rear view mirror of the vehicle. We extract vehicle candidate area from the video image, then identify and track the vehicle candidate region. Its main contents include:1. Extract vehicle candidate area. In this method, the collected video image is converted into HSV space. The red area position is extracted by H component of the video image and horizontal edge of vehicle bottom position is extracted by V component. Combining the above two position, we determined the vehicle candidate region of the image.2. Optimize Gabor filter group. Gabor filter group is improved by Quantum Evolutionary Algorithm (QEA). Optimized Gabor filter group is used to extract the characteristics of the video image candidate region, and then using support vector machine to train and recognize characteristics of the selected candidate area. The filter group is improved through Quantum Evolutionary Algorithm (QEA) which introduced the niche co-evolutionary algorithms. And the improved filter group is clustered to reduce or lower redundancy. Compared with the recognition rate of fixed filter group and genetic algorithm optimized filter group, our optimized algorithm have higher recognition rate and the recognition speed is quicker.3. Track vehicles which have been confirmed on the video image. Using improved Kalman filter to forecast and track. Kalman gain is improved through variable forgetting factor which will avoid the accumulation of errors. The vehicle position in the next frame image is predicted by the improved Kalman filter. And then compared with the position of the extracted candidate region, if the coordinate positions are identical, the location is confirmed as the position of the tracking vehicles. Otherwise, we should use the posterior predicted value of the previous frame to replace the real value and forecast the next frame. Through this method, we can avoid the vehicle identification process and raise the tracking speed.