Article: Research on map matching of lidar/vision sensor for automatic driving aided positioning Journal: International Journal of Vehicle Information and Communication Systems (IJVICS) 2021 Vol.6 No.2 pp.121 - 136 Abstract: Aiming at the technical difficulties of map matching aided positioning based on lidar/vision sensor, the joint calibration of lidar/vision sensor and point cloud/image registration technology, as well as the dynamic environment interference removal method based on depth learning are studied. In this paper, a lightweight coding-decoding architecture is introduced. We use deep separable convolution technology to extract urban environment features and generate semantic-level feature descriptors. The similarity measurement criteria that contain semantic information and geometric state are established. Then, we perform the robust feature matching. Finally, a map matching location model based on recursive Bayesian filtering optimisation framework and an estimation method of location confidence are proposed. It realises the map assistant positioning under the complex environment of the city. In typical urban environments, the speed of feature extraction and matching is better than 0.1s, the success rate of matching is more than 95%, and the positioning accuracy of high-precision map matching is better than 20 cm. Inderscience Publishers - linking academia, business and industry through research

Title: Research on map matching of lidar/vision sensor for automatic driving aided positioning

Authors: Qing An; Xijiang Chen; Yuhua OuYang

Addresses: School of Urban Construction, Wuchang University of Technology, Wuhan, Hubei 430223, China; Wuhan Huagong Cloud Technology Co., Ltd., No. 133 Tangxunhubei Road, Wuhan, Hubei 430223, China ' School of Resources and Environmental Engineering, Wuhan University of Technology, Wuhan, Hubei 430070, China ' School of Urban Construction, Wuchang University of Technology, Wuhan, Hubei 430223, China

Abstract: Aiming at the technical difficulties of map matching aided positioning based on lidar/vision sensor, the joint calibration of lidar/vision sensor and point cloud/image registration technology, as well as the dynamic environment interference removal method based on depth learning are studied. In this paper, a lightweight coding-decoding architecture is introduced. We use deep separable convolution technology to extract urban environment features and generate semantic-level feature descriptors. The similarity measurement criteria that contain semantic information and geometric state are established. Then, we perform the robust feature matching. Finally, a map matching location model based on recursive Bayesian filtering optimisation framework and an estimation method of location confidence are proposed. It realises the map assistant positioning under the complex environment of the city. In typical urban environments, the speed of feature extraction and matching is better than 0.1s, the success rate of matching is more than 95%, and the positioning accuracy of high-precision map matching is better than 20 cm.

Keywords: laser radar; visual sensor; point cloud; depth learning; map matching.

DOI: 10.1504/IJVICS.2021.115253

International Journal of Vehicle Information and Communication Systems, 2021 Vol.6 No.2, pp.121 - 136

Received: 15 Jan 2020
Accepted: 03 Apr 2020
Published online: 26 May 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Research on map matching of lidar/vision sensor for automatic driving aided positioning

Keep up-to-date