基于树形尺度估计的车辆跟踪

Vehicle tracking based on tree scale estimation

为解决智能视频监控系统车辆跟踪过程中尺度伸缩变化造成的跟踪器模型漂移的问题,在SiamFC(基于全连接层的孪生网络目标跟踪算法)的基础上,提出一种基于树形尺度池的车辆跟踪算法。通过判断当前图片与模板图片中目标的大小,为其分配尺度因子,再通过尺度因子之间的响应大小确定当前目标的最佳尺度。此外,为保证模板图片能够适应车辆外观的不断变化,在确定尺度的条件下对模型进行自适应更新,提高跟踪算法整体的精确度和成功率。实验表明,该算法可以有效解决车辆跟踪中尺度变化导致的车辆漂移情况,且相对于其他孪生算法有更好的跟踪性能。

In order to solve the problem of tracker model drift caused by scaling changes in a vehicle tracking process in a intelligent video surveillance system,a vehicle tracking algorithm based on a tree scale pool is proposed on the basis of SiamFC(a twin network target tracking algorithm based on a fully connected layer)algorithm.By judging the size of the target in the current picture and the template picture,a scale factor is assigned to it,and then the optimal scale of the current target is determined by the size of the response between the scale factors.In addition,in order to ensure that the template picture can adapt to the continuous changes in the appearance of the vehicle,the model is adaptively updated under the condition of a certain scale to improve the overall accuracy and success rate of the tracking algorithm.The experiments show that the algorithm can effectively solve the vehicle drift caused by the change of the scale in the vehicle tracking,and it has better tracking performance than other twin algorithms.