基于云边端协同与改进YOLOv10的高速公路车辆类型识别系统设计

Design of a Highway Vehicle Type Recognition System Based on Cloud-Edge-End Collaboration and Improved YOLOv10

摘要: 针对高速公路交通监控场景中车辆类型识别实时性不足、人工统计效率较低以及云端集中处理压力较大的问题，本文设计了一种基于云边端协同架构的高速公路车辆类型识别系统，并构建了CBAM-WIoU-YOLOv10车辆目标检测与类型识别模型。系统包括端侧视频采集、边侧模型推理和云侧结果展示与统计分析三个层级，将车辆检测与类型识别任务部署于边缘侧完成，仅向云侧上传车辆类别、置信度、检测时间和设备编号等结构化识别结果，从而有助于降低原始视频集中上传造成的网络带宽压力。针对高速公路监控图像中车辆尺度变化明显、背景干扰较强以及部分类别外观相似等问题，本文在YOLOv10m基础上引入CBAM注意力机制，以增强模型对车辆关键特征区域的关注能力；同时采用WIoU损失函数优化边界框回归过程，提高车辆目标定位精度。实验基于包含6类车辆目标的数据集开展训练与验证，车辆类别包括小型车辆、中型车辆、大型车辆、小型货车、大型货车和油罐车。实验结果表明，改进模型在6类车辆目标上的精确率(Precision)、召回率(Recall)、mAP@0.5和mAP@0.5:0.95分别达到93.45%、89.82%、96.35%和82.49%，推理速度达到78.1帧/秒(FPS)。结果表明，所提模型在检测精度与推理效率之间取得了较好平衡，能够为高速公路车辆类型识别和交通运行统计分析提供有效的模型与数据支撑。

Abstract: To address the problems of insufficient real-time performance in vehicle type recognition, low efficiency of manual traffic statistics, and high computational pressure caused by centralized cloud processing in highway traffic monitoring scenarios, this paper designs a highway vehicle type recognition system based on a cloud-edge-end collaborative architecture and develops a CBAM-WIoU-YOLOv10 model for vehicle object detection and type recognition. The system consists of three layers: video acquisition at the end layer, model inference at the edge layer, and result visualization and statistical analysis at the cloud layer. Vehicle detection and type recognition tasks are performed at the edge layer, while only structured recognition results, including vehicle category, confidence score, detection time, and device ID, are uploaded to the cloud platform, thereby helping to reduce the network bandwidth pressure caused by the centralized transmission of raw video streams. To address significant vehicle-scale variations, complex background interference, and similar appearances among certain vehicle categories in highway monitoring images, the CBAM attention mechanism is introduced into YOLOv10m to enhance the model’s ability to focus on key vehicle feature regions. Meanwhile, the WIoU loss function is employed to optimize bounding-box regression and improve vehicle localization accuracy. Experiments are conducted on a dataset containing six vehicle categories, including small vehicles, medium-sized vehicles, large vehicles, small trucks, large trucks, and tank trucks. The experimental results show that the improved model achieves a Precision of 93.45%, a Recall of 89.82%, an mAP@0.5 of 96.35%, and an mAP@0.5:0.95 of 82.49%, while maintaining an inference speed of 78.1 frames per second (FPS). These results demonstrate that the proposed model achieves a favorable balance between detection accuracy and inference efficiency, providing effective model and data support for highway vehicle type recognition and traffic operation analysis.