基于注意力机制的前方道路场景识别算法  

Front Scene of Road Recognition Algorithm Based on Attention Mechanism

作　　者：温爽 李学伟[2] 张同宇 乐海丰 WEN Shuang;LI Xuewei;ZHANG Tongyu;LE Haifeng(College of Urban Rail Transit and Logistics,Beijing Union University,Beijing 100101,China;School of Economics and Management,Beijing Jiaotong University,Beijing 100044,China;Beijing Key Laboratory of Information Service Engineering,Beijing Union University,Beijing 100101,China)

机构地区：[1]北京联合大学城市轨道交通与物流学院,北京100101 [2]北京交通大学经济管理学院,北京100044 [3]北京联合大学北京市信息服务工程重点实验室,北京100101

出　　处：《北京联合大学学报》2022年第3期38-43,共6页Journal of Beijing Union University

基　　金：中国国家铁路集团有限公司科技研究开发计划课题(K2019Z006)。

摘　　要：随着我国机动车保有量逐年上升,而路网的建设相对缓慢,由此造成行车拥堵问题日趋严重。准确快速地识别出路口和路段等道路场景,对车辆安全驾驶具有很好的辅助作用。为了通过实时性较强的辅助驾驶功能提高机动车行驶的安全性,针对网络规模较小的ResNet18模型进行改进,加入空间注意力模块进行模型优化。实验结果表明,在复杂交通路况下,基于注意力机制改进后的ResNet18模型在道路场景数据集上的识别准确率达到91.5%,推理时间为4.935 ms,在保持实时性的同时提高了图像识别准确率,满足辅助驾驶需求。With the rising number of motor vehicles in our country year by year,the construction of the road network is relatively slow,resulting in the more and more serious problem of traffic congestion.Accurate and fast identification of road scenes such as intersections and road sections is a good aid to safe vehicle driving.In order to improve the driving safety of motor vehicles through the real-time auxiliary driving system,the ResNet18 model with a small network size is optimized by adding a spatial attention module.The experimental results show that under complex traffic conditions,the improved ResNet18 model based on attention mechanism achieves 91.5%recognition accuracy on the road scene dataset with an inference time of 4.935ms,which improves the image recognition accuracy while maintaining real-time and satisfies the assisted driving requirements.

关 键 词：智慧交通 卷积神经网络 图像识别 辅助驾驶 

分 类 号：U471.1[机械工程—车辆工程]