Table 3.

Comparison with other methods on different scenes of the XJTU-IAIR traffic scene dataset.

Traffic Scenes	Methods	Frames	GT	TP	FP	FN	Precision (%)	Recall (%)	FPS
Highway	Ours	999	1998	1998	0	0	100.0	100.0	30
	He’s [14]	999	1998	1805	183	172	90.8	91.3	14
	Bertozzi’s [41]	999	1998	1886	33	79	98.3	96.0	35
	Seo’s [9]	999	1998	1711	211	78	89.0	95.6	21
Moderate Urban	Ours	999	1998	1966	30	0	98.5	100.0	18
	He’s [14]	999	1998	1946	419	31	82.3	98.4	11
	Bertozzi’s [41]	999	1998	1649	236	152	87.5	91.6	34
	Seo’s [9]	999	1998	1363	124	511	91.7	72.7	14
Heavy Urban	Ours	899	1692	1621	89	26	94.8	98.4	19
	He’s [14]	899	1692	1609	608	69	72.6	95.9	11
	Bertozzi’s [41]	899	1692	1289	386	212	77.0	85.9	31
	Seo’s [9]	899	1692	970	295	441	76.7	68.7	15
Illumination	Ours	1040	2080	2080	0	0	100.0	100.0	23
	He’s [14]	1040	2080	1725	697	305	71.2	85.0	23
	Bertozzi’s [41]	1040	2080	1882	185	32	91.0	98.3	34
	Seo’s [9]	1040	2080	1506	448	128	77.1	92.2	17
Night	Ours	804	1566	1566	10	0	99.4	100.0	35
	He’s [14]	804	1566	1491	326	75	82.1	95.2	12
	Bertozzi’s [41]	804	1566	942	379	603	71.3	61.0	36
	Seo’s [9]	804	1566	1383	104	87	93.0	94.1	24
Changing Lanes	Ours	548	1096	1096	0	0	100.0	100.0	28
	He’s [14]	548	1096	876	368	81	70.4	91.5	13
	Bertozzi’s [41]	548	1096	1015	85	5	92.3	99.5	39
	Seo’s [9]	548	1096	975	99	24	90.8	97.6	21