【摘要】 在车路协同中，路侧感知设备可以检测出行车环境中感兴趣的目标，协助车辆进行智能决策。论文使用架设在路口的相机采集可见光图像，对行人、自行车和汽车等目标进行标注，构建了路侧目标检测数据集。论文采用最新的anchor-free目标检测方法 YOLOX作为基准模型，增加CBAM和SENet两种attention模块以加强主干网络特征，使用focal loss替换物体分支原有的交叉熵损失以解决正负样本不平衡的问题。在自建数据集上的实验结果表明，改进后的YOLOX-A模型其检测精度有明显提升，同时检测速度几乎不变，适合于路侧场景下的目标检测。更多还原

【Abstract】 In vehicle-road collaboration,roadside sensing devices can detect interesting targets in the driving environment and assist vehicles in making intelligent decisions. This paper uses a camera set up at the intersection to collect visible light images,annotates pedestrians,bicycles,and cars,and builds a roadside target detection data set. This paper uses the newest anchor-free target detection method YOLOX as the baseline model,adds CBAM and SENet as two attention modules to enhance the backbone network features,and uses focal loss to replace the original cross-entropy loss of the object branch to solve the problem of positive negative sample imbalance. The experimental results on the self-built data set show that the proposed YOLOX-A model has significantly improved detection accuracy,while the detection speed is almost unchanged,which is suitable for target detection in roadside scenes.更多还原