Multiobject Tracking in Videos Based on LSTM and Deep Reinforcement Learning

Table 1

The quantitative comparison results of our tracker with other state-of-the-art trackers.

Method	Sequence	MOTA%	MOTP%	FP	IDSW
RNN-LSTM	PETS09-S2L2	38.3	71.6	1016	320
SiameseCNN		47.5	72.6	341	796
MDPSubCNN		34.5	69.7	672	282
LP_SSVM		41.5	70.5	629	212
LSTM_DRL (ours)		45.8	72.9	354	255
RNN-LSTM	ADL-Rundle-3	23.7	72.0	2193	158
SiameseCNN		39.7	72.9	191	33
MDPSubCNN		44.9	79.6	793	56
LP_SSVM		28.0	72.9	1855	81
LSTM_DRL (ours)		45.2	75.1	651	30
RNN-LSTM	TUD-Crossing	57.2	71.1	81	43
SiameseCNN		73.7	73.0	85	8
MDPSubCNN		78.9	76.7	32	6
LP_SSVM		60.0	74.2	48	18
LSTM_DRL (ours)		79.1	75.9	30	11
RNN-LSTM	AVG-Town Center	13.4	68.8	1206	299
SiameseCNN		19.3	69.0	698	142
MDPSubCNN		49.5	70.1	1381	121
LP_SSVM		14.7	70.1	459	123
LSTM_DRL (ours)		38.6	71.0	598	101
RNN-LSTM	Venice-1	12.7	71.7	686	56
SiameseCNN		22.3	73.0	322	4
MDPSubCNN		15.9	72.4	843	47
LP_SSVM		17.8	73.0	696	23
LSTM_DRL (ours)		23.4	73.8	302	6
RNN-LSTM	ETH-Jelmoli	34.8	73.3	314	59
SiameseCNN		42.3	72.8	315	30
MDPSubCNN		32.9	73.6	639	22
LP_SSVM		39.5	74.4	224	17
LSTM_DRL (ours)		43.9	75.1	213	13
RNN-LSTM	ETH-Linthescher	12.4	74.7	164	49
SiameseCNN		16.7	74.2	93	27
MDPSubCNN		27.2	74.7	191	48
LP_SSVM		15.6	75.6	41	11
LSTM_DRL (ours)		27.1	75.4	52	14
eRNN-LSTM	ETH-Crossing	21.1	75.5	27	7
SiameseCNN		27.5	74.1	20	4
MDPSubCNN		28.8	74.7	59	0
LP_SSVM		24.9	75.6	10	2
LSTM_DRL (ours)		29.6	77.9	12	4
RNN-LSTM	ADL-Rundle-1	−2.2	69.9	4213	241
SiameseCNN		25.6	71.6	1999	33
MDPSubCNN		16.2	71.5	3157	49
LP_SSVM		14.0	71.9	3507	69
LSTM_DRL (ours)		27.9	72.8	1846	39