Multi-Object Trackers

Introduction to Multiple Target Tracking

简单介绍了多目标物跟踪的流程、方法。

1、single target tracking (STT) 单目标跟踪，不需要association和assignment。

2、 multiple target tracking (MTT) 多目标跟踪，很难，原因：

①、Target or detection distribution，目标分布很紧，不同目标、轨迹紧挨着。

②、Probability of detection (Pd) of the sensor，识别的概率比较低，容易丢帧，或者错误匹配。

③、Sensor resolution，传感器分辨率低。

④、Clutter or false alarm rate of the sensor，误识别。

⑤、The number of targets and detections，轨迹或者目标多，计算指数增长。

3、工作流程，略

4、Detection，识别

传感器对于一个target（真实目标），只有一个detection（探测）。当然也可以有0个探测，或者虚假目标（Farse alarm）。

某些传感器对于一个target可能有多个detection，但是对于这些算法，必须先聚类。(such as trackerGNN, trackerJPDA, and trackerTOMHT)。

5、Gating and Assignment

简单介绍了GNN、JPDA、MHT方法。

6、Track Maintenance

Track maintenance refers to the function of track initiation, confirmation, and deletion.包括三块。

7、Filtering

①、用track预测

②、匹配

③、更新track

8、Tracking Metrics ？？？

评价目标跟踪算法，有三类：

① 、trackAssignmentMetrics ：评价目标匹配。 provides indexes such as number of the track swaps, number of divergence steps, and number of redundant assignments.

②、 trackErrorMetrics ：算track误差的。

③、 trackOSPAMetric：OSPA，provides three scalar error components — localization error, labelling error, and cardinality（基数？） error to evaluate tracking performance.

9、Non-Assignment-Based Trackers？？？

这部分也看不懂。

trackerGNN, trackerJPDA, and trackerTOMHT are assignment-based trackers, meaning that the track-to-detection assignment is required. The toolbox also offers a random finite set (RFS) based tracker, trackerPHD. You can use its supporting features ggiwphd to track extended objects and gmphd to track both extended objects and point targets.