Intelligent Low-Light Vehicle Detection and Behavior Analysis Using Enhanced YOLOv8 for Night-Time Traffic Monitoring

Abstract

The detection and behavior analysis of vehicles in lowlight conditions is a prominent challenge in modern intelligent transportation systems (ITS) due to the poor illumination, motion blur and high noise levels present within night- time surveillance footage. Precise indication of vehicles in this kind of surroundings is crucial for controlling traffic, preventing accidents and managing urban mobility. In this paper, we propose the intelligent low-light vehicle detection and behavior analysis system using YOLOv8 deep learning framework. It includes an enhanced low-light image enhancement module which is coupled with the YOLOv8 object detection model designed to improve visibility and feature extraction, for nighttime scenes. First, the input video frames are processed with adaptive illumination correction algorithm to achieve enhanced contrast and suppression of noise free background while useful structural information is preserved. These enhanced frames are sent to the YOLOv8 detection network, which detects vehicles like cars, buses, trucks, and motorcycles with high accuracy. Besides detection, it also has the ability to analyze behaviors by observing vehicle trajectories and looking for abnormal behavior (lane change, over-speeding and erratic movement). Publicly accessible night-time traffic datasets were used for experimental evaluation. The results of our proposed method showed a 96.4 % detection accuracy, precision 95.2%, recall 94.6% and F1-score of the model : 94.9% which is significantly superior to YOLOv5 & Faster R-CNN (Existing Detection Models) at low light conditions." Case Study 1: Vehicle Detection and Inter-vehicle Behavior Assessment for Night-time Traffic Data Collection As a case study, we integrated the low-light enhancement block with YOLOv8 to assess the reliability of vehicle detection in nighttime traffic monitoring applications by evaluating its behavior in low-light conditions.