基于LCD-YOLO的车辆检测算法

Research on vehicle detection algorithm based on LCD-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（Lightweight Car Detection-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(Frequency-Adaptive Dilated Convolution,FADC)优化YOLOv8s中的C2f（CSP bottleneck with 2 convolutions），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，能够有效提高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%，更好满足车辆的

Abstract:

For the problems of low detection accuracy, large number of parameters, and large computation of the existing YOLOv8s vehicle detection model, a LCD-YOLO (Lightweight Car Detection-YOLO) lightweight vehicle target detection algorithm based on improved YOLOv8s is proposed. The algorithm uses Frequency-Adaptive Dilated Convolution (FADC) to optimize C2f (CSP bottleneck with 2 convolutions) in YOLOv8s to improve the feature fusion ability, and uses shared convolutional layers to reduce the number of network convolutions. to reduce the number of parameters to achieve model lightweighting; through the dynamic focusing of the bounding box regression loss calculation method, it 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, and the amount of network parameters is reduced by 14.9% and the amount of computation is reduced by 10.9%, which can better satisfy the actual detection needs of vehicles.