Feature‐based detection and classification of moving objects using LiDAR sensor. Issue 7 (26th March 2019)

• Record Type: Journal Article
• Authors: Guo, Ziming; Cai, Baigen; Jiang, Wei; Wang, Jian
• Is Part Of: IET intelligent transport systems. Volume 13:Issue 7(2019)
• Page Start: 1088
• Page End: 1096
• Publication Date: 2019-03-26
• Subjects: object detection -- image segmentation -- image classification -- optical radar -- image matching -- feature extraction; vehicle demonstrator -- object matching framework -- feature‐based detection -- autonomous driving -- single LiDAR sensor -- moving vehicles -- object geometry -- reflection intensity -- object classification -- point cloud segmentation -- feature reference ranges -- hand‐labelled object class -- feature‐based classification; Intelligent transportation systems -- Periodicals; Electronics in transportation -- Periodicals; 388.31205
• DOI: 10.1049/iet-its.2018.5291
• Languages: English
• ISSNs: 1751-956X

Abstract

Abstract : Detection and classification of moving objects is essential for autonomous driving. To tackle this problem, this paper proposes an object classification method at detection level using a single LiDAR sensor. The aim is to extract and classify all the moving vehicles, bicyclists, and pedestrians in front of the sensor. First, the point clouds are segmented to produce distinct groups of points representing different objects, where the line segments are extracted. A segmentation combination strategy is conducted to address the over‐segmentation caused by occlusion. Then, considering the object geometry and reflection intensity, several features for classification are defined and extracted from different hand‐labelled object classes. Reference ranges of all the features are generated on a set of experiment samples. Finally, the object class can be decided by checking if its features match with the existing feature reference ranges of a certain class. The proposed method was evaluated using datasets gathered by the vehicle demonstrator, and the experiment results show considerable improvement of classification performance compared to an existing object classification method based on the object matching framework.