Semantic web-based video surveillance systems can provide strong decision-making support for managers, and they have high requirements for real-time and precision of vehicle detection models in complex night scenes. To address this issue, a lightweight nighttime vehicle detection method (MC-YOLO) integrating MobileNetV2 and YOLOV3 is proposed. Firstly, in the preprocessing stage, image enhancement is performed on nighttime images to facilitate model feature extraction. Then, the lightweight network MobileNetV2 is used to extract feature by replacing the backbone network DarkNet53 in YOLOv3, thus accelerating the speed of target detection. Finally, after the convolution operation of the backbone network, a convolution block attention module is added to enhance the important feature information and suppress the secondary features, thereby improving the detection precision. The experimental results on the BDD100K dataset show that the proposed MC-YOLO model has a precision of up to 92.75%, which is superior to several other advanced comparative models.