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

Table 1

Summary and evaluation of previous studies on traditional data association.

Author/organization	Years	Technical characteristics	Dataset	MOTA↑(%)	MOTP↑(%)	IDs↓
Shuoyuan Xu/Cranfield University	2018	JPDA applied to UAV [34]	TUD; PETS	56.3; 32.7	—	—
Bićanić/University of Zagreb	2019	Combined with JIPDA and appearance-based tracker [35]	3DMOT2015	55.9	64.0	486
Anton Andriyenk/TU Darmstadt	2011	Formulate MOT as minimization of a continuous energy function [38]	terrace2	88.1	78.1	11
			TUD	60.5	65.8	7
			PETS	81.4	76.1	15
			ped1-c2	48.0	75.5	4
Anton Milan/TU Darmstadt	2014	Formulate MOT as minimization of a continuous energy function [39]	PETS-S2L1	90.6	80.2	11
			PETS-S2L2	56.9	59.4	73
			PETS-S2L3	45.4	64.6	27
	2016	Discrete-continuous energy function minimization [40]	Datasets^1	56.3	66.0	42
Guoyu Zuo/Beijing University of Technology	2018	MDP object tracking method [41]	PETS-S2L1	88.1	68.7	5
Jianbing Shen/Beijing Institute of Technology	2018	Choose the best among small trajectories [42]	TUD-Crossing	60.2	77.2	32
			PETS09-S2L2	21.3	70.7	251
Songhao Zhu/Nanjing University of Posts and Telecommunications	2016	Hierarchical association MOT trajectory generation method [43]	Datasets^2	—	—	25
Ali Taalimi/University of Tennessee	2017	Novel hierarchical approach, network flow [44]	MOT2015	33.9	71.4	12.1
Junying Liu/Beihang University	2018	Hierarchical MOT [45]	PETS-S2L1	91.04	87.85	18
			PETS-S2L2	52.83	79.54	265
			ETH	67.09	81.05	85
			KITTI	72.89	90.61	119
Hongbin Liu/Shandong University	2018	Hierarchical data association using main-part and spatial-temporal feature models [46]	CAVIAR	88.6	89.4	3
			Parking Lot	76.2	73.9	34
			MOT15	17.5	70.9	683
Weigang Lu/Jiangnan University	2019	Hierarchical network flows [47]	MOT16	68.2	60.2	953

¹S2. L1, S2. L2, S2. L3, S2. L1-2, S2. L2-1, Stadtmitte. ²TUD, PETS2009.