检测前跟踪算法性能评估指标体系

Performance evaluation metrics for Track-Before-Detect algorithms

摘要:

针对传统性能评估指标难以有效评估检测前跟踪(TBD)算法性能的问题，构建了TBD算法性能评估通用流程，并从单帧检测和虚警、航迹检测和虚警及计算资源3个方面，建立了一套基于单目标模型的性能评估指标体系。在此基础上，比较了Hough变换和递推贝叶斯滤波2种典型TBD算法的性能，仿真结果验证了该指标体系的有效性。

Abstract:

A performance evaluation process is proposed to deal with the problem that the traditional performance metrics being hard to evaluate the Track-Before-Detect(TBD) algorithms effectively. Some performance evaluation metrics based on single target are proposed, from three aspects including single frame detection and false alarm, track detection and false alarm, and computing resources. On that basis, the performances of Hough transform and recursive Bayesian filtering are compared in simulation, which verifies the validity of the performance evaluation metrics.