Research And Realization Of Motion Vehicle Detection And Tracking

As the technology for detecting and tracking moving objects is being used in public places, intelligent traffic control, and military aspects more and more widely, that has been a hot study field of computer vision today, which makes many researchers focus on this technology. In this technology field, the detection and tracking of moving vehicles belong to the core part of the modern intelligent traffic, but the effects of illumination changes, object occlusion, similar objects and many other factors which reduce the real-time and accuracy of existing algorithms in the practical environment. Therefore, the study for the movement of a high-efficiency vehicle detection and tracking algorithm has been a great challenge for us. Based on the summary and analysis of existing vehicle detection and tracking technology, the corresponding vehicle detection and tracking algorithms are proposed depending on the reality of confounding factors presented in the environment.In part of moving vehicle detection, this thesis analyzes the four commonly used experimental detection algorithms:frame difference method, the average background subtraction, optical flow, and mixed Gaussian. Then for their shortcomings, this thesis proposes a vehicle detection algorithm based on codebook. By the analysis of the original models and color space, this thesis reduce the three color channels to the one color channel, eliminating three variables about time and frequency, replacing the maximum and minimum luminance values with the luminance difference, and introducing the weights of the codeword, deleting or changing some unnecessary parameters, so that the processing speed of detection algorithm is improved on the basis of the higher accuracy. Finally, this thesis sets the segmentation threshold by the brightness between the shadow and the car body to eliminate the shadow, which provides a more accurate target area for the vehicle tracking.In the part of moving vehicle tracking, firstly, for the problem of the traditional Cam Shift algorithm based on the color feature can not effectively track dynamic target by the affects of vulnerable illumination changes, similar color background interference and other environmental factors. To solve this problem, vehicle tracking based on multiple features fusion is presented. The proposed method combines the local three values model (local ternary pattern, LTP) with the HSV (hue, saturation, value) color feature to perform a features fusion that can effectively improve the accuracy of the Cam Shift tracking algorithm. Then, for the effect of the object occlusion, similar objects, we combine Kalman filter which containing information about target motion with the improved Cam Shift algorithm, forming a improved Cam Shift algorithm based on Kalman filter, which using the measured values generated by Kalman filter and the predictive value generated by Cam Shift algorithm.The combined algorithm can effectively overcome the interference of these two factors and track the moving vehicles under the complex conditions.The experimental results show that, this tracking method has a high tracking accuracy, strong anti-interference ability, high robustness characteristics, and better practical value.