Three-Dimensional Pedestrian Detection by Fusing Image Semantics and Point Cloud Spatial Visibility Features

Table 1. Performance comparison of different visibility description methods on KITTI dataset
View table
Table 1. Performance comparison of different visibility description methods on KITTI dataset
Visibility code
［U，O，F］
3D pedestrian detection AP /% mAP /%
Easy Moderate Hard
［0，1，-1］ 68.76 62.49 57.74 63.00
［0.5，0.7，0.4］ 70.84 64.57 59.77 65.06

Table 2. Performance comparison of different density along height direction on nuScenes dataset
View table
Table 2. Performance comparison of different density along height direction on nuScenes dataset
Number of channels 3D pedestrian detection AP /% mAP /%
0.5 m 1.0 m 2.0 m 4.0 m
1 62.24 64.36 66.36 68.78 65.44
32 69.68 71.87 73.73 75.97 72.81

Table 3. Performance comparison of different number of frames on nuScenes dataset
View table
Table 3. Performance comparison of different number of frames on nuScenes dataset
Nunber of frames 3D pedestrian detection AP /% mAP /%
0.5 m 1.0 m 2.0 m 4.0 m
1 37.46 38.22 39.29 40.40 38.84
10 68.36 70.63 72.30 74.59 71.47

Table 4. Performance comparison of different optimized methods on KITTI validation set
View table
Table 4. Performance comparison of different optimized methods on KITTI validation set
Method 3D pedestrian detection AP /% mAP /%
Easy Moderate Hard
PP（Baseline） 70.16 63.40 57.49 63.68
PP+Vis. 71.86 64.49 58.72 65.02
PP+Img. 72.08 65.56 60.34 65.99
PP+Vis.+Img. 71.84 65.65 60.81 66.10

Table 5. Performance comparison of different methods on KITTI test set

View table

Table 5. Performance comparison of different methods on KITTI test set

Method	3D pedestrian detection AP /%			Speed /Hz
Method	Easy	Moderate	Hard	Speed /Hz
VoxelNet^［7］	39.48	33.69	31.51	4.4
AVOD^［3］	36.10	27.86	25.76	10
SECOND^［18］	51.07	42.56	37.29	20
F-PointNet^［4］	50.53	42.15	38.08	5.9
PointPainting^［6］	50.32	40.97	37.87	2.5
PointRCNN^［19］	47.98	39.37	36.01	10
Proposed method	51.07	41.36	37.83	30

Tools

Get Citation

Copy Citation Text

Lu Xiong, Zhenwen Deng, Wei Tian, Zhiang Wang. Three-Dimensional Pedestrian Detection by Fusing Image Semantics and Point Cloud Spatial Visibility Features[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0228011

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Feb. 14, 2022

Accepted: Mar. 14, 2022

Published Online: Feb. 7, 2023

The Author Email: Wei Tian (tian_wei@tongji.edu.cn)

DOI:10.3788/LOP220712

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Performance comparison of different visibility description methods on KITTI dataset

Table 1. Performance comparison of different visibility description methods on KITTI dataset

Table 2. Performance comparison of different density along height direction on nuScenes dataset

Table 2. Performance comparison of different density along height direction on nuScenes dataset

Table 3. Performance comparison of different number of frames on nuScenes dataset

Table 3. Performance comparison of different number of frames on nuScenes dataset

Table 4. Performance comparison of different optimized methods on KITTI validation set

Table 4. Performance comparison of different optimized methods on KITTI validation set

Table 5. Performance comparison of different methods on KITTI test set

Table 5. Performance comparison of different methods on KITTI test set