复杂交通场景中运动车辆的检测与轨迹跟踪

针对复杂交通场景提出一种基于高斯分布假设的背景图像自回归估计算法,该算法能同时适应白天和夜间光环境,实时性好.对于二值图像的分割问题,提出并论证一种新的连通标记算法,该算法只需遍历像素1次,因此时间复杂度达到了理论上的最小.根据运动车辆的随机过程特性,提出基于Kalman滤波的轨迹跟踪算法,给出状态转移矩阵和观测矩阵,并讨论初始状态矢量的获取方法.为了解决半遮挡混合图块的分割问题,提出了图像相似度的计算方法以及局部图块与全图块相匹配的思想.在实际道路上的实验表明,所提出的方法实用有效,其中车辆跟踪准确率达到95.63％.

Detection and Tracking of Moving Vehicles in Complicated Traffic Scene

A new autoregression algorithm, based on Gaussian Distribution hypotheses and feasible in daytime and night lamp environment, was proposed for the background estimation in digital image sequences. A new pixel labeled algorithm, which only traversed the pixels one time hence was the time optimal algorithm, was put forward and demonstrated. Furthermore, the transition matrices and observation matrix of Kalman filter adopted to track vehicles were presented, while the method to get the first state vector of Kalman filter was also studied and proposed. Additionally, a calculation approach of image correlation was presented for matching the original image with the part-occlusion combined image or with the image segmentation which was segmented from the combined one. The experimental results indicated that the algorithms and methods presented in this paper were effective and performed well in detecting and tracking vehicles, and the success ratio of tracking vehicles was 95.63%.