基于YOLO网络的单目视觉教学与重复系统

Monocular Visual Teaching and Repeating System Based on YOLO

视觉教学与重复是一种处理目标需要定期巡检的重要方法。为了保持任务点处重复阶段与教学阶段有相同的相机姿态,重新设计了单目视觉教学与重复系统。系统记录教学阶段的路径并在路径上设置多个任务点。在重复阶段重复行驶路线,到达每一个任务点处时,以YOLO网络识别参照物获取其在图像中的高度,并通过针孔模型变换获得特征点的真实深度。实验表明,在0.6~3.40m距离范围内,深度估计误差约为0.02m,能够满足系统要求。

Visual teaching and repeating is an important method for dealing with targets that require regular inspections.In order to maintain the same camera pose at the repeating phase and the teaching phase at the task point,the monocular visual teaching and repeating system is redesigned in this paper.The system records the path of the teaching phase and sets multiple task points on the path.Repeat the driving route in the repeating phase.When one of task points is reached,the reference object is recognized by the YOLO network to obtain its height in the image,and the true depth of the feature point is obtained through the pinhole model transformation.Experiments show that in the range of 0.6~3.40 m,the error of depth estimation is about 0.02 m,which can meet the system requirements.