This is published under the terms of CC BY-NC.

Downloaded: 550 times

Abstract

Rapid advances have occurred in Internet of Things technologies. Among Internet of Things-related applications, Internet of Vehicles (IoV) is regarded as integral infrastructure for next-generation intelligent transportation systems. IoV requires vehicles to perceive their surroundings reliably. In particular, researchers have focused on LiDAR sensing because it is robust in extreme weather. However, IoV sensing data are transmitted between vehicles and the cloud, and LiDAR requires a large quantity of data; thus, communication for cloud computing might be challenging. To address this difficulty, a LiDAR-based detection method for an IoV edge node is proposed. Small-object detection through LiDAR sensing is difficult because of the sparsity of point clouds. Although some researchers have attempted to solve this problem by fusing raw point cloud details, existing approaches still reduce model efficiency and memory cost, which is unsuitable for IoV. To overcome the problem, this paper proposes a novel model that enhances three-dimensional (3D) structural information for preserving the details of voxel features while maintaining the high efficiency and low memory usage of voxel-based methods. Experimental results indicate that the proposed method outperforms state-of-the-art LiDAR-based 3D detection methods on the widely used KITTI dataset and achieves competitive performance for all classes.