This study proposes a method to detect vehicles by enhancing the YOLOv5 network 1
structure and incorporating RetinexNet to address the problem of limited detection capabilities of 2
target identification algorithms in low illumination conditions, such as at night time. CBAM attention 3
module is implemented in the network’s Neck detection layer to extract the vehicle’s primary features, 4
reduce the extraction of unused features, and improve the vehicle’s detection performance. The 5
DIoU is introduced as the loss function of the model to solve the imprecise location of the prediction 6
box and speed up the convergence of the model. RetinexNet is applied to enhance the detection 7
capabilities of low-illumination images by enhancing and denoising them. The experimental results 8
indicate that the enhanced model detects vehicles with 90% accuracy in low-light conditions and has 9
a strong detection performance overall.