摘要
我国的轨道交通线路里程长,途径环境的复杂多样化,由自然灾害、人为因素、随机异物等造成轨道出现障碍物,严重影响行车安全。在客流量大,列车班次间隔时差短的今天,通过人为检测、固定安装监控点的方式随着运营线路的加长成本愈加高昂,因此,通过结合行驶的列车记录的单目实时图像,提出了一种改进的LeNet-5卷积神经网络的轨道交通障碍物检测方法,可实时识别出列车前方铁轨是否存在障碍物,为列车控制系统提供智能预警信息,避免列车与障碍物发生碰撞,保障列车行车的安全性和可靠性。
China’s rail transit lines have long mileage,the environment is complex and diversified.Natural disasters,human factors,and random foreign objects can be cause obstacles in the orbit,which seriously affects driving safety.In today’s large passenger flow and in a short interval,the method of artificially detecting and fixedly installing camera points can not meet the actual demand.Therefore,this paper proposes an improved LeNet by combining the monocular real-time images recorded by the traveling trains.The LeNet-5 convolutional neural network’s method for detecting obstacles in rail traffic can identify whether there are obstacles in the rails in front of the train,provide intelligent warning information for the train control system,avoid collision,ensure the safety and reliability of trains.
出处
《工业控制计算机》
2020年第1期63-66,共4页
Industrial Control Computer
关键词
轨道交通
卷积神经网络
障碍物识别
安全预警
rail transit
convolutional neural network
obstacle recognition
safety warning