Real-time Vehicle Detection System for Intelligent Transportation using Machine Learning

Abstract

Vehicle detection plays an important role in analyzing traffic flow data for efficient planning in intelligent transportation. Machine Learning technology has been increasingly used for vehicle detection in traffic flows. However, adverse weather conditions bring challenges for 2D vehicle detection. There is a lack of research on real-time vehicle detection using 3D LiDAR point clouds, which are more resistant to adverse weather conditions. In this paper, we proposed a system for collecting real-time traffic data using both 2D and 3D LiDAR cameras, processing the collected data for vehicle detection, and providing a web-based service with statistical traffic flow data visualization and 2D real-time vehicle detection stream display. We generated 1980 images from the 2D traffic flow videos that were collected in California Highway, and trained a 2D machine learning model on Darknet using YOLO algorithm. Approximately, 7000 frames of LiDAR point cloud data were labeled and pre-processed, and a new deep learning model for 3D vehicle detection was proposed. Compared with YOLO’s original pre-trained mode, our 2D machine learning model improved the vehicle detection that 6 different types of vehicles could be classified with an average precision of 89.25%.