小尺度交通信号灯的检测与状态识别  

Detection and State Recognition of Traffic Light with Small Scale

作　　者：董晓玉 孔斌 杨静[2,3,4] 王灿[2,3,4] DONG Xiao-yu;KONG Bin;YANG Jing;WANG CAN(Anhui Jianzhu University,Hefei 230601,China;Institute of Intelligent Machines,Chinese Academy of Sciences,Hefei 230031,China;Anhui Engineering Laboratory for Intelligent Driving Technology and Application,Hefei 230088,China;Key Laboratory of Biomimetic Sensing and Advanced Robot Technology,Hefei 230031,China)

机构地区：[1]安徽建筑大学,安徽合肥230601 [2]中国科学院合肥智能机械研究所,安徽合肥230031 [3]安徽省智能驾驶技术及应用工程实验室,安徽合肥230088 [4]安徽省仿生感知与先进机器人技术重点实验室,安徽合肥230031

出　　处：《测控技术》2020年第11期45-51,共7页Measurement & Control Technology

基　　金：国家自然科学基金重大研究计划集成项目(913203002);安徽省新能源汽车暨智能网联汽车产业技术创新工程项目;中国科学院合肥物质科学研究院院长基金青年“火花”项目(YZJJ2020QN20)。

摘　　要：交通信号灯识别包括检测和状态识别,在智能交通系统中发挥重要作用。基于YOLOv3算法提出了一种交通信号灯检测与状态识别模型。针对交通信号灯相较于交通场景中其他目标具有尺度小的特性进行了算法的设计:降低骨干网络的下采样倍率以增加小尺度目标的特征描述能力;通过增大特征图的尺度来改进多尺度特征融合;引入广义交并比作为检测任务的损失函数来改进目标边界框的回归效果。同时,根据交通信号灯本身的特性,使用颜色和形状约束的方法对信号灯进行状态识别和类别验证。最后在公开的Bosch交通信号灯数据集上和实际的城区道路进行了实验验证。实验结果表明,所提出的算法能够提升交通信号灯识别的精度和召回率,识别准确率可以达到90%左右。Traffic light recognition which consists of detection and state estimation plays an important role in intelligent traffic system.A traffic light detection and state recognition model based on YOLOv3 algorithm is proposed.Some improvements were made to the traffic light for its small scale compared with other objects.Downsampling rate of the backbone was reduced to increase the ability of feature representation of small scale object.The multi-scale feature map fusion was improved by increasing its scale.Moreover,the generalized intersection ratio was introduced as the loss function of the detection to improve the regression effect of the object's bounding box.At the same time,according to the characteristics of the traffic light itself,the method of color and shape constraints was used to identify the state and verify the category of the traffic light.Finally,experimental verification was carried out on the Bosch Small Traffic Light Dataset and actual urban roads.Experimental results show that the proposed algorithm can improve the precision and recall rate of traffic light recognition,and the recognition accuracy can reach about 90%.

关 键 词：交通信号灯识别 小尺度目标 特征融合 广义交并比 颜色和形状约束 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]