| As the foundation of the intelligent transportation system, Vehicle tracking basedon computer vision has been one of popular research topic for years. Visual vehicletracking is an interlaced subject, which is built on the basis of computer vision, imageprocessing technology, artificial intelligence and pattern recognition. In addition, theaccuracy of vehicle tracking is still a difficult point due to the variability of vehicles inmotion, the interference of background, occlusion of the vehicles and the rapidmovement of the vehicle. Therefore, a kind of high accuracy and robust vehicletracking algorithm is one of the problem which need setting urgently at present.The paper introduces the status of vehicle tracking for the first, secondly analyzedthe traditional Mean Shift algorithm in the application of vehicle tracking. However itsusage could be inefficiency, or even failure for locating while the scaling and occlusionof the vehicles are becoming significant. Moreover, the background interference andfast moving vehicle are also affecting the accuracy, As the comparability coefficientbetween the target model and candidate model will be decreased due to the variabilityof vehicles in motion and occlusion of the vehicles which lead to trap in local optimum.And then an improved Mean-Shift algorithm which combined with object informationand motion estimation of vehicle will overcome these problems. Based on the Mean-Shift algorithm, the new method has combined with the object information of thetarget, which could optimize the bandwidth of the kernel function and the target model.In addition, by using a Kalman filter to estimate the motion of the object, it solves theocclusion problem; otherwise, tacking failure caused by fast moving vehicle could besolved as Kalman filter avoid the defect which Mean-Shift algorithm using Taylor’sseries to estimate the initial search center. In the end, Experiments show that theimproved Mean-Shift algorithm can improve the tracking ability. |
PDF Full Text Request