Method

EagerMOT: 3D Multi-Object Tracking via Sensor Fusion [EagerMOT]
https://github.com/aleksandrkim61/EagerMOT

Submitted on 26 May. 2020 13:32 by
Aleksandr Kim (Technical University of Munich)

Running time:0.011 s
Environment:4 cores @ 3.0 Ghz (Python)

Method Description:
A simple real-time 3D tracking pipeline built using standard components:

During each frame, independent detections from 3D and 2D are fused together into individual object instances. These instances are matched to existing tracks during two consecutive stages: first using 3D information (3D bounding box IoU) and then using 2D information (2D bounding box IoU).

The method is suitable for 3D Multi-Object Tracking and requires only bounding box level detections. To adjust the method for MOTS, mask information is used for more precise fusion.
Parameters:
Using segmentation masks from MOTSFusion
Latex Bibtex:
@inproceedings{Kim21ICRA,
title = {EagerMOT: 3D Multi-Object Tracking via Sensor Fusion},
author = {Kim, Aleksandr and Osep, Aljo\v{s}a and Leal-Taix{'e}, Laura},
booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
year = {2021}
}

Detailed Results

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


Benchmark sMOTSA MOTSA MOTSP MODSA MODSP
CAR 74.50 % 83.50 % 89.60 % 84.80 % 92.10 %
PEDESTRIAN 58.10 % 72.00 % 81.50 % 73.30 % 94.10 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 86.50 % 98.10 % 91.90 % 31792 629 4968 5.70 % 37505 1229
PEDESTRIAN 75.60 % 97.10 % 85.00 % 15640 459 5057 4.10 % 18753 704

Benchmark MT PT ML IDS FRAG
CAR 67.10 % 29.40 % 3.50 % 457 811
PEDESTRIAN 43.30 % 43.00 % 13.70 % 270 633

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