Study On Vehicle Identification And Tracking Based On Gabor Wavelet

In intelligent transportation systems, detection and tracking system based on video sequence has the features of simple installation, working stability, visual information-rich and easy to realize unmanned monitoring. It is currently a hot research both at home and abroad. In this paper, we have analysis and summarized the existing methods of identification and tracking in intelligent transportation systems. Based on this, we have a research on recognition and tracing of moving vehicles in traffic monitoring system.In this paper, Fourier transform, wavelet transform and Gabor wavelet transform are compared. Gabor transform has excellent properties in the analysis of the local area frequency and the direction information in digital image. In computer vision and texture segmentation it also has been widely used. In this paper, the performance of two-dimensional Gabor filter was analyzed, a multi-channel Gabor filter was designed, and the parameters of which is chosen. The texture features of the image are extracted by the multi-channel Gabor filter, which was descript by a feature vector.Based on the analysis of common identification methods, according to the texture model, this paper presents an identification technology which was combined by features extraction using Gabor wavelet and BP neural network algorithm for the classification of vehicle types. First of all, the video image sequence was collected. A background reconstruction algorithm is used in the case of moving targets in the sequences, which can acquire and update the dynamic background real-time. It can suppress the impact of outside environmental change. The background subtraction was employed to locate the target. For removing the noise, the images were filtered by median filtering. Finally, an 80×60 standard target image was intercepted. As the correlation degree of feature points'energy value which was extracted from different vehicle types is low, the database of four standard vehicle types including car, trucks, vans and bus was established. The extracted target features were matched with the standard database at last. For vehicles which were easy mistaken recognized, the vehicle types need to be further divided and the discrimination of Gabor wavelets needs to be improved. However the amount of Gabor feature points would be greatly increased. In order to ensure real-time systems, the discrimination of Gabor wavelet was increased. At the same time a BP neural network classifier was designed to train and identify the feature points. Experiments show that this method not only has a high-precision and a good real-time.The tracking of traditional template matching consumes a lot of time, for overcome this shortcoming, a tracking method based on Kalman filter was used. First, the possible vehicle location of the next frame was predicted by Kalman filter. The vehicle was located precisely by the method of matching the Gabor feature points in predicting region. Affine transformation model was introduced to correct targets in the case of target translation and scaling change. To further enhance the tracking speed, all feature points have been screened in the experiment. Some points of typical characteristics were selected to match with the standard vehicle templates in the database. The experimental results show that this method has good tracking results, and the vehicle which was blocked in a short time can be effectively tracked.