基于LCD-YOLO的车辆检测算法

Research on Vehicle Detection Algorithm Based on Lightweight Car Detection-YOLO

摘要:

针对现有YOLOv8s车辆检测模型的检测精度低、参数量多、计算量大的问题,提出了一种基于改进YOLOv8s的LCD-YOLO轻量化车辆目标检测算法。该算法使用频率自适应膨胀卷积优化YOLOv8s中的C2f模块,以提升特征融合能力；使用共享卷积层,减少网络卷积次数,以此减少参数量来实现模型轻量化；通过动态聚焦的边界框回归损失计算方法,能够有效增强网络对遮挡重叠目标检测能力,提高边框检测精度。在KITTI数据集上进行实验,结果表明所提算法平均检测精度提升到95.1%,相比于YOLOv8s算法检测精度提高了2.9%,网络参数量减少14.9%,计算量减少10.9%,能更好满足车辆的实际检测需求。

关键词:车辆检测; YOLOv8s; 特征融合; 轻量化

Abstract:

To address the issues of the low detection accuracy, exorbitant number of parameters, and extensive computations in the existing YOLOv8s vehicle detection model, a lightweight car detection-YOLO (LCD-YOLO) algorithm based on improved YOLOv8s is proposed for lightweight vehicle target detection. The algorithm applies frequency-adaptive dilated convolution (FADC) to optimize the cross-range partial (CSP) bottleneck with two convolutions (C2f) in YOLOv8s to enhance feature fusion ability. Shared convolutional layers reduce the number of network convolutions and parameters, thereby achieving a lightweight model. Through the dynamic focusing of the bounding box regression loss calculation method, this model can effectively improve the network's ability to detect occluded overlapping targets and improve the accuracy of border detection. Experiments on the KITTI dataset show that the average detection accuracy of the proposed algorithm is improved to 95.1%, which is 2.9% higher than that of the YOLOv8s algorithm, while reducing the number of network parameters by 14.9% and amount of computation by 10.9%, which can better satisfy the actual detection needs of vehicles.

Key words:vehicle detection; YOLOv8s; feature fusion; lightweight