| 摘 要: 为解决复杂交通监控场景下车辆检测精度不足的问题,提出基于 YOLOv8模型改进的CDS-YOLOv8车辆检测算法。首先,在特征融合部分采用轻量化的CCFM,降低了模型参数和计算量,增强模型对于尺度变化的适应性。其次,采用配备多种注意力机制的 DyHead检测头,以增强在复杂背景中识别低分辨率目标的性能。最后,结合SAConv模块提高显著区域的表示能力。实验结果表明,在智能交通数据集上,相较于原始 YOLOv8模型,CDS-YOLOv8的精度、召回率和mAP分别提高了3%、5.7%和5.1%,展现出较强的实用性。 |
| 关键词: YOLOv8 车辆检测 注意力机制 |
|
中图分类号: TP183
文献标识码: A
|
|
| Improving Traffic Surveillance Vehicle Detection Algorithm for YOLOv8 |
|
WANG Huiqin, QIAN Songrong
|
(State Key Laboratory of Public Big Data, Guizhou University, Guiyang 550025, China)
wanghq_gzu@163.com; qiansongr_gzu@163.com
|
| Abstract: To address the issue of insufficient vehicle detection accuracy in complex traffic surveillance scenarios,a vehicle detection algorithm named CDS-YOLOv8 is proposed based on the YOLOv8 model. Firstly, a lightweight CCFM is adopted in the feature fusion part to reduce the number of model parameters and computational load, enhancing the model’s adaptability to scale changes. Secondly, a DyHead detection head equipped with various attention mechanisms is used to improve the performance of recognizing low-resolution targets in complex backgrounds. Lastly, the SAConv module is combined to enhance the representation ability of salient regions.Experimental results show that, compared to the original YOLOv8, CDS-YOLOv8 has increased precision, recall, and mAP by 3% , 5.7% , and 5.1% respectively on intelligent traffic datasets, demonstrating strong practicality |
| Keywords: YOLOv8 vehicle detection attention mechanism |
