基于序列蒙特卡罗滤波的车辆跟踪算法研究

Vehicle Tracking Algorithm based on Sequential Monte Carlo Filter

针对非线性目标跟踪中模型或函数近似等最优估计缺陷问题,提出了基于帧间预测和特征匹配的序列蒙特卡罗滤波跟踪算法。算法中采用在HSV色彩下的空间加权直方图描述跟踪车辆的状态特征,通过简单的随机漂移模型实现估测样本的帧间传递,利用估测样本与期望目标间的相似度量完成样本权重赋值运算,最终利用加权样本值估计实现待测目标的后验状态。实验结果表明,基于序列蒙特卡罗滤波的车辆跟踪算法计算简单有效,能够在复杂环境下实时、准确跟踪道路上无规律、非线性运动的车辆,并能够有效适应车辆部分遮挡和短时丢失等情况。

Against the defect of the optimal estimation achieved by models or functions approximation in the nonlinear target tracking, this paper proposes a sequential Monte Carlo filter tracking algorithm based on inter-flame prediction and feature matching. In the algo- rithm, the state of vehicle tracked is described by the weighted spatiograms in HSV color space, the samples relay is completed by a normal random drift model, the computation of the sample weights is also achieved by the similarity measure between sample and target, and the posterior probability of state vector is computed by the weighted samples. The experiment research shows that this approach is simple and computationally efficient and can track the irregular and nonlinear moving vehicles real time and accurately in a complex environment. In addition, the means can also deal with the situations of partial occlusion and short-term loss as well.