Improved multi object detection in night vision environment for autonomous vehicle using EnPreNiNet model

Abstract

Autonomous vehicles have emerged as a prominent research area due to their wide range of applications in intelligent transportation systems. A fundamental objective of autonomous driving is the accurate detection and classification of multiple objects in real-world environments. Although numerous deep learning algorithms have demonstrated excellent performance under normal weather conditions, their detection accuracy deteriorates significantly in adverse environments, particularly under night vision conditions. The primary challenges include uneven pixel intensity distribution, low contrast, motion blur, partial occlusion, poor illumination, and information loss, which make it difficult to distinguish foreground objects from the background. Consequently, critical road entities such as pedestrians, vehicles, buildings, roads, and streetlights are often inadequately represented, leading to segmentation failures and reduced multi-object detection performance. To overcome these limitations, this paper proposes a novel deep learning ensemble framework, EnPreNiNet, integrated with an improved image enhancement technique called the GHE Ensemble model. The GHE Ensemble model enhances night vision images by reducing noise, improving contrast, and achieving a more uniform distribution of pixel intensities, thereby preserving essential scene information. The enhanced images are subsequently processed by the proposed EnPreNiNet framework, which combines Enhanced Faster R-CNN and Enhanced YOLOv5 to achieve robust multi-object detection and classification in night vision environments. Experimental results demonstrate that the proposed framework achieves a mean Average Precision (mAP) of 76.3%, significantly outperforming existing state-of-the-art methods for multi-object detection under low-light and night-time driving conditions.