摘要
卷积神经网络检测算法自提出以来,日益受到人们的关注,广泛应用于各行各业。然而,在城市轨道交通障碍物检测方面,由于列车制动距离长、响应度不够、突发状况多等问题,卷积神经网络尚未有过多的应用。针对城市轨道交通行车的特点,提出一种基于YOLOv3网络的在途列车障碍物检测方法,通过对该网络结构及训练方式的优化,提高网络对于微小目标的识别准确率,留给列车更多的响应时间,以提升该方法的实用性。经实验验证,该方法对于远距离目标识别有较高的准确率,检测速率约为31帧/s,是一种有效的列车在途障碍物检测方法。
Since the start of convolutional neural network detection algorithm,it has been paid more and more attention and widely used in various industries.However,in urban rail transit obstacle detection,convolution neural network has not been used much because of the problems of long braking distance,insuffi cient response and many different emergency scenarios.Aiming at the characteristics of transit,this paper proposes a method of obstacle detection for en-route transit train based on YOLOv3 network.Through optimizing the network structure and training mode,the accuracy rate of recognition of small targets is improved,and more response time is left for the train to improve the practicability of the method.The experiment shows that the method has a high recognition accuracy for long-distance targets,and the detection rate is about 31 frames/s,which is an effective method for the detection of obstacles for train en-route.
出处
《现代城市轨道交通》
2021年第6期19-23,共5页
Modern Urban Transit