摘要
机器人视觉避障是一个热门的研究方向。针对机器人在行进时如何通过视觉传感器进行自主避障的问题,设计了一种深度卷积神经网络,通过收集训练样本,得到一个端到端控制的自主避障网络模型,并对模型做了定量的评估比较。在现实场景中用单目摄像头行进采集时的第一视角图像和转向角,使用训练好的模型进行转向角的预测,来控制自行搭建一辆两轮自平衡车平台的行进方向。实验结果表明,采用该方法的机器人具有一定的自主避障能力。
Aiming at how the robot can autonomously avoid obstacles through the visual sensor while traveling,a deep convolutional neural network is designed,an end-to-end controlled obstacle avoidance network model is trained and quantitatively evaluated and compared by collecting training samples.In the real scene,the monocular camera is used to collect the first angle image and the steering angle while traveling,and the trained model is used to predict the steering angle to control the traveling direction of a two-wheel self-balancing platform.
出处
《工业控制计算机》
2019年第9期77-79,共3页
Industrial Control Computer
关键词
视觉避障
深度卷积神经网络
端到端控制
visual obstacle avoidance
deep convolutional neural network
end-to-end control
