关联深度自适应的多假设跟踪研究

多假设跟踪(multiple hypothesis tracking, MHT)方法是一种在多个扫描上评价关联假设并由此做出决策的贝叶斯型关联跟踪方法,此方法能够在信噪比低10~100倍的状况下获得与单扫描方法相当的性能,但同时会带来相当大的计算量。本文研究了面向航迹MHT中的关键算法,包括航迹得分计算与航迹树的生成、将航迹聚类和假设生成建模为图论问题并求解、N-扫描回溯剪枝等,特别关注了这些算法过程的实现;提出了一种关联深度自适应(adaptive association depth, AAD)方法,使关联深度随关联场景的复杂程度自适应变化;仿真研究了本文提出的AAD-MHT跟踪密集目标的性能,结果和分析表明,与深度值固定为6的MHT相比,最大深度为6的AAD-MHT既能保证性能又有效降低了计算量。

Multiple hypothesis tracking with adaptive association depth

Multiple hypothesis tracking(MHT) is a Bayesian association method that evaluates association hypotheses among multiple scans and makes evaluation based decisions. Comparing with the single hypothesis method, MHT can work reasonably under 10~100 times lower signal-noise ratio (SNR) but it needs much more computational load. The implementation of track-oriented MHT (TOMHT) is studied and some key points are investigated, include calculating the track score, generating the track tree, modeling track clustering, hypotheses generating as problems in graph theory and N-scan pruning, etc. In the TOMHT framework, an adaptive association depth (AAD) method is proposed. This method makes the association depth change adaptively with the complexity of scenarios. Its performance is investigated by several simulation experiments on tracking closely targets. The results and analysis show that the performance of AAD-MHT is nearly the same as MHT but the computational load is much lower.