Computer Vision-Based Affordable High-Density Traffic Monitoring System

Abstract

Vehicle detection and traffic monitoring system play a significant role in many areas of transportation infrastructure and to reduce traffic congestion. In this research, an affordable computer vision-based amazingly simple method has been proposed to detect and track vehicles, and also monitor high traffic density using CCTV cameras. Traffic monitoring system’s infrastructure is expensive to build and implement. An affordable traffic monitoring system has been proposed which when implemented will be able to autonomize traffic monitoring system consequently reduce traffic congestion. Still image and traffic footage dataset were collected and input into the model. Vehicles were detected by YOLOv3 algorithm which was modified to detect 5 classes of vehicles only. Traffic density was estimated with respect to time and the whole model was implemented in Raspberry Pi. A series of experiments were conducted to validate the accuracy and affordability of the model. Highest model accuracy obtained from vehicle detection from still image was 92%, vehicle detection from traffic footage was 88%, traffic density estimation was 88%. When compared with existing model, the proposed model was 3 times more affordable. Variations in lighting and angles of camera position in real life can affect the vehicle detection and identification process which will provide further scopes of work. This research was aimed to contribute a little to pave the pathways which will help traffic monitoring and ultimately reduce traffic congestion