Method

3D mot based on simultaneous optimization of object detection and scene flow [la] [on] [DetFlowTrack]


Submitted on 23 Feb. 2022 08:20 by
Yueling Shen (Shanghai Jiao Tong University)

Running time:0.2 s
Environment:8 cores @ >3.5 Ghz (Python)

Method Description:
We proposes a 3d multi-target tracking algorithm
based on multi-task learning, and simultaneously
optimizes object detection and scene flow
estimation. The scene flow is detection induced
for accurate inter-frame association. The post-
processing module realize inter-frame box
association based on detection and scene flow
results, and finally realizes 3D multi-target
tracking.
Parameters:
-
Latex Bibtex:
-

Detailed Results

From all 29 test sequences, our benchmark computes the HOTA tracking metrics (HOTA, DetA, AssA, DetRe, DetPr, AssRe, AssPr, LocA) [1] as well as the CLEARMOT, MT/PT/ML, identity switches, and fragmentation [2,3] metrics. The tables below show all of these metrics.


Benchmark HOTA DetA AssA DetRe DetPr AssRe AssPr LocA
CAR 71.52 % 72.87 % 70.89 % 79.56 % 82.98 % 73.47 % 90.64 % 87.79 %
PEDESTRIAN 39.64 % 40.90 % 38.72 % 47.54 % 56.35 % 41.48 % 66.03 % 72.04 %

Benchmark TP FP FN
CAR 30922 3470 2051
PEDESTRIAN 15391 7759 4140

Benchmark MOTA MOTP MODA IDSW sMOTA
CAR 83.34 % 86.42 % 83.95 % 210 71.12 %
PEDESTRIAN 46.75 % 65.53 % 48.60 % 429 23.83 %

Benchmark MT rate PT rate ML rate FRAG
CAR 72.15 % 22.92 % 4.92 % 360
PEDESTRIAN 31.27 % 51.55 % 17.18 % 1460

Benchmark # Dets # Tracks
CAR 32973 1119
PEDESTRIAN 19531 1128

This table as LaTeX