Method

HRI-SFMOT[la][gp] [HRI-SFMOT]


Submitted on 14 Jan. 2023 13:42 by
Zhicheng Huang (HIKVISION-HRI)

Running time:0.1 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
Two cues for handling data association:1) location
of the closest keypoint on 3D bbox;2)velocity vector
calculated by scene flow
A new multistage matching algorithm for balancing
accuracy and efficiency
Parameters:
None
Latex Bibtex:

Detailed Results

From all 29 test sequences, our benchmark computes the commonly used tracking metrics CLEARMOT, MT/PT/ML, identity switches, and fragmentations [1,2]. The tables below show all of these metrics.


Benchmark MOTA MOTP MODA MODP
CAR 92.63 % 86.79 % 92.66 % 89.21 %
PEDESTRIAN 49.13 % 64.84 % 50.21 % 88.68 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 96.47 % 96.96 % 96.71 % 37153 1166 1358 10.48 % 44527 748
PEDESTRIAN 64.31 % 82.53 % 72.28 % 15032 3183 8344 28.61 % 20090 237

Benchmark MT PT ML IDS FRAG
CAR 87.85 % 8.46 % 3.69 % 12 41
PEDESTRIAN 30.93 % 34.71 % 34.36 % 249 1085

This table as LaTeX

[1] K. Bernardin, R. Stiefelhagen: Evaluating Multiple Object Tracking Performance: The CLEAR MOT Metrics. JIVP 2008.
[2] Y. Li, C. Huang, R. Nevatia: Learning to associate: HybridBoosted multi-target tracker for crowded scene. CVPR 2009.


eXTReMe Tracker