基于状态依赖检测的视频运动目标轨迹跟踪算法

龚畅　王华君　李荣　徐平平

摘 要： 目前大多数的目标跟踪算法都假设目标检测的概率与轨迹的状态无关，但是随着目标跟踪性能的降低，这个假设通常不正确，针对此问题，提出一种基于状态检测的目标运动轨迹跟踪算法。不同于大多数现有算法，该算法在目标运动轨迹与检测状态相关的基础上展开。首先预测从先前时刻到当前时刻的每个跟踪的概率密度函数；然后从测量值集合中选取一个子集用于更新，并计算相似度测量；最后迭代计算跟踪质量和目标轨迹评估，将目标检测的状态依赖概率融入到多目标跟踪器中。利用二维多目标跟踪进行仿真实验，交替固定传感器每秒都对监控区域进行扫描，结果显示，跟踪确认率可达97%，显著提高了跟踪性能。此外，该算法在其他交通视频上也有较好的跟踪效果，具有较好的实用性。

关键词： 多目标跟踪； 视频跟踪； 运动轨迹仿真； 状态检测

中图分类号： TN919?34； TP391 文献标识码： A 文章编号： 1004?373X（2016）07?0051?06

Abstract： The majority of target tracking algorithms assume that the probability of the target detection is independent of track state， but it is usually incorrect with the performance degradation of the target tracking. For this issue， a target moving trail tracking algorithm based on state detection is proposed. Unlike most current algorithms， the proposed algorithm is unfolded based on the correlation of target moving trail and detection state. Each tracked probability density function from previous moment to current moment is forecasted， and then a subset for update is selected from the measured value set and the similarity measurement is calculated. The tracking quality is iteratively calculated， the target trail is finally evaluated， and the state dependence probability of target detection is fused into multi?target tracker. The simulation test was performed by means of the two?dimensional multi?target tracking. The alternate fixed sensors are used to scan the monitored area per second. The scanning results show that the tracking identification rate can reach up to 97%， and the tracking performance is significantly improved. In addition， the algorithm has a good tracking effect in other traffic videos， and has good practicability.

Keywords： multi?target tracking； video tracking； moving trail simulation； state detection

0 引 言

目标跟踪[1]一般是指监控系统对目标的数量和轨迹进行评估，在军事、交通、安防等重大领域应用非常广泛[2?3]，是一个非常热门的研究课题。但对轨迹的评估过程容易受到虚假信号（噪声、杂波和多路径）的影响。如何处理目标轨迹位于发生目标信号衰减的检测区域的目标跟踪非常重要[4]。

目标跟踪按照跟踪的数量可以分为单目标跟踪和多目标跟踪。已经有很多学者对其进行了研究。文献[5]提出改进的背景加权直方图算法，是一种单目标跟踪算法。……