Researching | 3D vehicle detection for unmanned driving systerm based on lidar

Optics and Precision Engineering, Volume. 30, Issue 4, 489(2022)

3D vehicle detection for unmanned driving systerm based on lidar

Xiru WU^* and Qiwei XUE

Author Affiliations

College of Electronic Engineering and Automation， Guilin University of Electronic Technology， Guilin541004， China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (15)

References (22)

Paper Information

Figures & Tables(15)

Fig. 1. Overall framework of 3D vehicle detection algorithm

Download full size

View in Article

Fig. 2. Image before filtering point cloud data

Download full size

View in Article

Fig. 3. Image after filtering point cloud data

Download full size

View in Article

Fig. 4. Image before RANSAC ground segmentation

Download full size

View in Article

Fig. 5. Image after RANSAC ground segmentation

Download full size

View in Article

Fig. 6. Frame of obstacle detection network

Download full size

View in Article

Fig. 7. Fusion sampling process

Download full size

View in Article

Fig. 8. Center point selection block diagram

Download full size

View in Article

Fig. 9. 3D vehicle detection result of experiment 1

Download full size

View in Article

Fig. 10. 3D vehicle detection result of experiment 2

Download full size

View in Article

Fig. 11. 3D vehicle detection result of experiment 3

Download full size

View in Article

Table 1. Point cloud statistical filtering algorithm

View in Article

Table 1. Point cloud statistical filtering algorithm

算法：点云统计滤波算法

步骤1　创建PCL滤波器对象。

步骤2　设置待滤波点云数据。

步骤3　统计时考虑查询点临近点数。

步骤4　根据设定阈值判断是否为离群点。

步骤5　剔除离群点。

步骤6　重复步骤3-步骤5直至所有点距离小于 $d_{m a x}$ 。

Table 2. RANSAC segmentation algorithm

View in Article

Table 2. RANSAC segmentation algorithm

算法：RANSAC算法

步骤1 在点云中随机选取三个点并计算任意一点与平面距离 $d_{i}$

$n = (P_{2} - P_{1}) \times (P_{3} - P_{1})$

$d_{i} = \frac{n^{T} (P_{i} - P_{1})}{{‖n‖}_{2}}$

其中， $P_{i}$ 为点云任意一点， $i = 4,5, \dots, N$

步骤2 设定阈值 $d_{i} = τ$ 保存点云并扫描激光雷达数据的点云数量。

步骤3 重复步骤1至步骤3共 $T$ 次，将点数量最多的点云保存下来以拟合出准确的平面参数。

$T = \frac{l o g (1 - q)}{l o g [1 - {(1 - e)}^{s}]}$

$e$ ：点云数据中异常点的比例；

$s$ ：每次迭代选取点的数量；

$T$ ：RANSAC 最大迭代次数；

$q$ ：至少一次选取到正常点的概率。

步骤4 若阈值 $d_{i} < τ = 2 σ$ 则设定为地面点并在点云数据中剔除。

$σ = \sqrt[]{\frac{\sum_{i = 1}^{N} (d_{i} {- \bar{d})}^{2}}{N - 1}}$

其中， $\bar{d} = \frac{1}{N} \sum_{1}^{N} d_{i}$

Table 3. Comparison of pretreatment models
View table
View in Article
Table 3. Comparison of pretreatment models
样本
序号
预处理前预处理后
样本
点数
邻近距离方差
样本
点数
邻近距离方差
1 16 210 0.36 0.8 15 418 0.29 0.1
2 15 447 0.59 0.7 14 501 0.49 0.3
3 17 730 0.48 0.5 16 946 0.42 0.4

Table 4. Performance comparison of 3D detection network
View table
View in Article
Table 4. Performance comparison of 3D detection network
算法检测精度时间/s
实验1 实验2 实验3
AVOD 66.47% 76.39% 60.23% 0.08
3DBN 73.53% 83.77% 66.23% 0.13
Point-GNN 79.47% 88.33% 72.29% 0.60
F-PointNet 69.79% 82.19% 60.59% 0.17
本文算法 75.32% 89.72% 74.26% 0.12

Tools

Get Citation

Copy Citation Text

Xiru WU, Qiwei XUE. 3D vehicle detection for unmanned driving systerm based on lidar[J]. Optics and Precision Engineering, 2022, 30(4): 489

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Information Sciences

Received: Jul. 8, 2021

Accepted: --

Published Online: Mar. 4, 2022

The Author Email: WU Xiru (xiruwu520@163.com)

DOI:10.37188/OPE.20223004.0489

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology