基于改进SFA3D的道路环境目标检测  

Road Environment Object Detection Based on Improved SFA3D

作　　者：况强 KUANG Qiang(Xuzhou Mining Group Co.,Ltd.,Xuzhou Jiangsu 221110,China)

机构地区：[1]徐州矿务集团有限公司,江苏徐州1221110

出　　处：《信息与电脑》2024年第6期132-136,共5页Information & Computer

摘　　要：传统固定时长的交通信号灯逐渐无法适应社会的发展,根据路口实时车流量、人流量控制红绿灯时长的智能交通信号灯系统技术逐步被社会接受。针对目前道路环境目标检测精度不高的问题,提出引用注意力机制、优化模型损失函数以改进SFA3D目标检测框架,精确检测场景中交通参与要素目标,辅助交通信号器决策获得最佳配时方案。首先将点云转换为鸟瞰图输入,结合残差网络(Residual Network,ResNet)利用特征金字塔思想进行特征提取,再通过注意力机制对特征信息进行注意区域的特征优化提取,最后对各尺度特征线性加权求和得到目标检测信息。实验结果表明,该算法相比于原始SFA3D网络在保证实时性的同时精度有所提高,证明算法的有效性。The traditional traffic signal lamp with fixed duration cannot adapt to the society development gradually.The intelligent traffic signal lamp control system technology which controls the traffic light duration according to the real-time traffic flow and human flow at the intersection is gradually accepted by the society.Aiming at the problem that the detection accuracy of road environment targets is not high,the attention mechanism and the optimized model loss function are used to improve the SFA3D object detection framework,and the objects of traffic participation elements in the scene are accurately detected,and assist the traffic signalers to make decisions to obtain the best timing scheme.Firstly,the point cloud is converted into the input of bird's-eye view,and the feature pyramid is used to extract the features with the help of ResNet.Then,the feature information of the attention region is optimized and extracted by the attention mechanism.Finally,the object detection information is obtained by the linear weighted sum of each scale feature.Experimental results show that compared with the original SFA3D network,the algorithm can guarantee the real-time performance and improve the accuracy,which proves the effectiveness of the algorithm.

关 键 词：SFA3D 目标检测 道路环境 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]