Data Association Methods via Video Signal Processing in Imperfect Tracking Scenarios: A Review and Evaluation

Table 3

Classification of video MOT datasets.

Name	Number of goals	Number of boxes	Camera mode	Scale	Tracking category	Features	Scenes
TownCenter [66]	16	—	Stable	4500	Pedestrian	Simple; annotation is complete; clear; blocked frequently	1
PETS09-S2L1 [67]	8	—	Stable	795	Pedestrian	Sparse crowd; high-speed nonlinear mode; blocked frequently	1
TUD-Stadtmitte [68]	—	—	Stable	179	Pedestrian	Low angle of view; severe mutual occlusion; complete occlusion	1
Parking Lot [69]	14	—	Stable	1000	Pedestrian	Parking lot; mutual blocking is more serious than TUD	2
PETS09-S2L2 [39]	—	—	Stable	168	Pedestrian	Medium-density crowds; high speed; blocked severely	1
MOT16 [56]	517	110407	Mobile/stable	5316	Pedestrian	Many scenes; comprehensive data; large amount of data	7
MOT17 [70]	1638	564228	Mobile/stable	15948	Pedestrian	Many scenes; more comprehensive than MOT16, magnitude of the data is larger	7
MOT20 [70]	2332	1336920	Stable	8931	Pedestrian	Wide scene at night; high crowd density	3
UA-DETRAC [71]	8250	1210000	Stable	140000	Vehicle	Canon EOS 550D camera records at 25 fps; rich scenes; large data volume	24
KITTI [72]	—	330000	Mobile	180 GB	Pedestrian/vehicle	Vehicle-mounted camera; up to 15 vehicles and 30 pedestrians in each image; various degrees of occlusion and truncation	>3
MOTs [73]	977	65213	Mobile	10870	Pedestrian/vehicle	Pixel-level relabeling on the KITTI_Tracking and MOTS challenge	>10
DukeMTMC [56]	1404	65213	Stable	36411	Pedestrian	Large HD video dataset, typical MOT scenario	8