Research on 3D Point Cloud Object Detection Algorithm for Autonomous Driving

Table 3

Comparison of 3D positioning performance between our algorithm and the state-of-the-art algorithm.

Method	Modality	Car			Pedestrian			Cyclist
		Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
PointPillars [22]	Only LiDAR	90.07	86.56	82.81	57.60	48.64	45.78	79.90	62.73	55.58
PointRCNN [33]		92.13	87.39	82.72	54.77	48.13	42.84	82.56	67.24	60.28
Part-A^2 [29]		94.07	85.35	75.88	59.04	49.81	45.92	83.43	68.73	61.85
PV-RCNN [39]		90.25	81.43	76.82	52.17	43.29	40.29	78.60	63.71	57.65
SE-SSD [40]		91.49	82.54	77.15	-	-	-	-	-	-
AVOD-FPN [18]	RGB + LiDAR	90.99	84.82	79.62	58.49	50.32	46.98	69.39	57.12	51.09
F-PointNet [16]		91.17	84.67	74.77	70.00	61.32	53.59	77.26	61.37	53.78
ContFuse [20]		94.07	85.35	75.88	—	—	—	—	—	—
MV3D [15]		86.62	78.93	69.80	—	—	—	—	—	—
Ours		95.01	88.32	80.57	79.67	66.89	56.36	90.12	72.41	63.21