基于YOLOv5s的交通信号灯检测算法

Traffic signal detection algorithm based on YOLOv5s

在自动驾驶和辅助驾驶领域,准确判断交通信号灯的状态与类别对于智能汽车的行车安全十分重要。针对城市道路交通信号灯目标小、背景环境复杂多样造成的检测难度大等问题,提出一种基于YOLOv5s的交通信号灯检测算法YOLOv5s_MCO。该算法使用MobileNetv2轻量化网络代替原主干特征提取网络,利用深度可分离卷积和逆残差结构,降低模型的参数量及计算量;然后引入卷积块注意力机制(convolutional blockattention module,CBAM),从通道和空间2个维度进行特征增强,增大网络的感受野,使网络更关注交通信号灯的目标特征,提高对小尺度目标的检测能力。实验结果表明:所提算法在自制的国内交通信号灯数据集上检测精度达到了81.89%,相较于原YOLOv5s算法提升了1.33%,同时改进后的模型大小仅为19.1 MB,检测速度达到了39.2帧/s,能够满足实时高效的检测要求。

In the field of autonomous driving and assisted driving,it is important to accurately determine the status and category of traffic signals for the driving safety of smart cars.To address the problems of detection difficulty caused by small targets of urban road traffic signals and complex and diverse background environments,a traffic signal detection algorithm YOLOv5s_MCO based on YOLOv5s is proposed.The algorithm uses MobileNetv2 lightweight network instead of the original backbone feature extraction network,and uses depth separable convolution and inverse residual structure to reduce the parameters and computation of the model.Then the convolutional block attention module is introduced to perform feature enhancement from both channel and space dimensions to increase the perceptual field of the network,so that the network can focus more on the target features of traffic signals and improve the detection ability of small-scale targets.The experimental results show that the detection accuracy reaches 81.89%on the homemade domestic traffic signal dataset,which is 1.33%better than the original YOLOv5s algorithm,while the size of the improved model is only 19.1 MB and the detection speed reaches 39.2 frames per second,which can meet the requirements of real-time and efficient detection.