A Cloud-based Real-Time Traffic Monitoring System with Lidar-based Vehicle Detection

The development of a traffic monitoring system to analyze and improve traffic flow is fundamental to the efforts of efficient traffic management, enabling smarter and safer transportation infrastructure and contributing overall to applications of intelligent transportation systems (ITS). There are real-time traffic monitoring systems for tracking and analyzing; however, the accuracy of vehicle detection and classification in these systems can be influenced by adverse weather conditions, resulting in so gaps in the traffic data. In this paper, we propose a cloud-based system that supports vehicle detection using machine learning models, presents the detected vehicles in the real-time stream and statistical data of traffic flow, and supports machine learning model and camera management. Data collection using a combination of 2D and 3D cameras was experimented with various configurations and locations to ensure data integrity and clarity. The collected datasets were labeling for 2D and 3D machine learning model training, checking model accuracy and creating a frontend/backend to create a web application to interface data stats/stream. We successfully deployed a website that can stream a 2D live video of traffic flow, achieving 2D vehicle detection model accuracy of 89%, and data is collected in real-time and streamed onto AWS using a C++ plugin for Amazon Kinesis Video Streams. We also store statistical data in a database that is then visualized for traffic flow analysis.