Yolov10-Driven Enhanced Vehicle Detection In Low-Light On-Board Environments

Abstract

Accurate vehicle detection in low-light and nighttime environments remains a significant challenge in intelligent transportation systems and autonomous driving. Vision-based on-board detection systems often suffer from reduced performance due to poor illumination, motion blur, glare from headlights, and high levels of image noise. To address these challenges, this project proposes a YOLOv10-driven enhanced vehicle detection framework designed for real-time applications in low-light conditions.

The proposed system integrates image pre-processing techniques such as adaptive histogram equalization, contrast enhancement, and noise reduction to improve image quality before detection. The enhanced images are then processed using a fine-tuned YOLOv10 model trained on diverse low-illumination datasets. This approach improves the model’s ability to detect small, distant, and partially occluded vehicles while reducing false positives caused by noise and lighting artifacts.

The lightweight and optimized architecture of YOLOv10 ensures high detection accuracy with low computational overhead, making it suitable for deployment on resource-constrained embedded systems. Experimental results demonstrate improved performance in terms of mean Average Precision (mAP), detection speed, and robustness compared to existing models such as YOLOv8 and Faster R-CNN.

Overall, the proposed system provides a reliable and efficient solution for vehicle detection in challenging nighttime environments, contributing to enhanced road safety and intelligent transportation systems.