结合粒子滤波的判别尺度空间目标跟踪算法

为了解决目前跟踪算法在运动目标被遮挡和尺度变换时跟踪效果不佳的问题，提出了一种结合粒子滤波的判别尺度空间跟踪算法。提取相邻两帧的目标区域，计算目标区域的结构相似性并与更新阈值进行比较，从而判断目标是否发生遮挡；其次，若发生遮挡，启用基于颜色分布的粒子滤波算法跟踪目标，反之，用判别尺度空间跟踪算法(DSST)中的位置滤波器确定目标位置；最后，利用尺度滤波器确定目标尺度并根据目标尺度更新粒子滤波的目标模型。经过在OTB2015测试集上进行实验，与判别尺度空间跟踪算法(DSST)、核相关滤波算法(KCF)等主流算法相比该算法的精确度和成功率均有所提高，尤其在发生遮挡后的跟踪效果表现最优。

Discriminative scale space target tracking algorithm based on particle filter

he tracking algorithm shows poor tracking effect when the moving target is occluded and the scale is changed. A discriminant scale space tracking algorithm combined with particle filter is proposed. Firstly, the target area of two adjacent frames is extracted, the structural similarity of the target area is calculated and compared with the updated threshold to determine whether the target is blocked. Secondly, if occlusion occurs, the particle filter algorithm based on color distribution is enabled to track the target, otherwise, the position filter in Discriminative Scale Space Tracking(DSST) is adopted to determine the target position. Finally, the scale filter is utilized to determine the target scale and the particle filter target model is updated according to the target scale. After the experiment on OTB2015 test set, the accuracy and success rate of the algorithm in this paper are improved compared with that of the mainstream algorithms such as DSST and Kernel Correlation Filter(KCF), especially the best tracking performance after occlusion occurs.