Multi-vehicle Tracking In Fuzzy Video And CUDA Acceleration

In the field of computer vision,target tracking plays a very important role.Excellent tracking algorithm can be used to monitor the number of people in the traffic video,and even can be used for missile tracking,to intercept illegal intrusion.In order to improve the performance of tracking the vehicle target in harsh environment,this paper studies the characteristics of rain and snow,haze and the characteristics of the images taken in these scenes.Based on the Stuck tracking algorithm,we proposed a fast adaptive tracking algorithm based on Stuck in bad scene,and applied the algorithm to multi-target tracking process.The main work of this paper is as follows:Images taken in the snow,fog and haze will appear blurred,low contrast,fuzzy phenomena.In the paper,we compared the image restoration effect using multi-scale Retinex,dark channel prior algorithm and the optimization of the contrast enhancement algorithm through the experiments,and calculated the running time of three algorithms in different scenarios and resolutions.In order to solve the scale change of the vehicle target in the tracking process,and combined with the method of generating candidate samples in the original Struck algorithm,this paper proposes an adaptive tracking algorithm based on multi-scale factor array.By adjusting the number of scale factors in the array and changing it's range,we can generate different numbers and scales of candidate samples,so that the tracking window can adapt to the vehicle size,to further improve the tracking accuracy.Although the improved adaptive tracking algorithm can improve the accuracy of tracking,but the computation of the large number of candidate samples,classifier evaluation and update operations can reduce the real-time performance of the algorithm.In order to apply this algorithm to multi-target tracking of the scene,so we use CUDA architecture to optimize the algorithm in parallel,and put time-consuming process on the GPU.Through the parallel optimization and improvement of the modules of the original algorithm,the algorithm runs three times faster than before.In order to solve the problem that the original algorithm is only applicable to the single target and the robustness to the occlusion process,this paper realizes the tracking of multiple targets by initializing multiple trackers and adding multiple threads to make the Struck algorithm more widely.The experimental results show that the proposed algorithm can deal with the scale change of the vehicle target and the shortcoming of the target loss in the tracking process.Compared with the original target tracking framework,the improved Struck algorithm has improved greatly in tracking accuracy and speed.