摘要
目标识别是智能车感知周围环境实现智能行驶的重要技术,高精度的目标识别算法可为智能车的安全行驶提供保障,因此提出一种基于三维激光雷达点云多特征提取的车辆识别算法。将提取的激光雷达数据的12维特征与分类器相结合识别城市道路中的车辆目标。算法首先对非地面激光雷达点云进行聚类,对聚类后的每簇点云数据提取12维特征;然后根据这12维特征训练四种分类器;最后使用KITTI数据集进行仿真,比较四种分类器的性能和三种不同维度特征(12维、26维、8维特征)识别车辆目标的准确率。仿真结果表明:提取的12维特征相比较于其它两种维度的激光雷达特征,可以提高车辆目标分类的准确性,与随机森林结合的识别精度优于其他分类器方法。另外,在百度Apollo数据集的道路场景中也验证了该算法的性能,结果表明其可满足车辆识别的精度。
Target recognition is an important technology for intelligent cars to perceive the surrounding environment to achieve smart driving,high-precision object recognition algorithms can provide a guarantee for the safe driving of intelligent cars.A vehicle recognition algorithm based on multi-feature extraction of 3D LiDAR point cloud is proposed.It combines the extracted 12-dimensional features of LiDAR data with a classifier to identify vehicle objects on urban roads.Firstly,it clusters non-ground LiDAR point clouds and extracts 12-dimensional features from each cluster of point cloud data,then four classifiers are trained according to these 12 dimensional features.Finally,it uses the KITTI data set for simulation,comparing the performance of four classifiers and the accuracy of identifying vehicle objects in three different dimensional features(12-dimensional,26-dimensional,and 8-dimensional features).The simulation results show the 12-dimensional features extracted in this paper compared with the other two dimensions of LiDAR features can improve the accuracy of vehicle object classification,and the recognition accuracy combined with random forest classifier is superior to other classifier methods.In addition,the algorithm can meet the accuracy of vehicle recognition in the road scene of Baidu Apollo dataset.
作者
李欣
李京英
LI Xin;LI Jingying(School of Communications and Information Engineering&School of Artificial Intelligence,Xi’an University of Posts&Telecommunications,Xi’an 710061,China)
出处
《传感器与微系统》
CSCD
2020年第10期138-141,共4页
Transducer and Microsystem Technologies
关键词
智能车
多线激光雷达点云
多特征
车辆识别
intelligent car
multi-line LiDAR point cloud
multi-features
vehicle recognition
