LiDAR-Based Vehicle Detection by Using DBSCAN Unsupervised Clustering Approach

Roadside LiDAR is helping to build intelligent and safe transportation. Object detection is a challenging and fundamental problem in computer vision. Moreover, the vehicle detection system is essential to Intelligent Transportation Systems (ITS). Many researchers in the transportation field spend an enormous amount of money to collect and analyze traffic data to optimize street systems. This research aims to develop a case study for a vehicle detection system in a complex roadway area based on LiDAR through an embedded system. For this purpose, an embedded GPU integrated (Nvidia JetsonTX2) with low power and high performance has been picked, which supports an unsupervised learning algorithm to be run simultaneously and a detection algorithm to be applied for point cloud recognition. We also discuss the structure of the architectures of LiDAR-based roadside systems, and lidar data processing for vehicle detection. In the real-scanned HDL-32E Velodyne 3D LiDAR dataset, our proposed method can achieve vehicle detection accuracy of up to 82.7% in several real-scene datasets. The future research directions to contribute resources beneficially to industry, academia, and government agencies for choosing appropriate LiDAR-based technologies for their vehicle monitoring systems.