YOLOEB: a lightweight method for identifying violations of electric bicycles

Abstract With the rise in traffic accidents due to the popularity of electric bicycles, automatic violation detection has become difficult. Machine vision-based detection faces challenges such as labor-intensive data annotation and decreased accuracy. This study presents the YOLOEB algorithm, which combines YOLOv7 and RepVGG block reparameterization to improve detection accuracy while maintaining inference time. YOLOEB uses Resnet-50 for classification and regression positioning for detection boxes. When evaluated on the Dataset-Det, YOLOEB achieved 98.5% detection accuracy and 97.2% recall rate, reducing annotation efforts and increasing processing speed to meet practical application requirements.