Towards Industrial Scenario Lane Detection: Vision-Based AGV Navigation Methods

In this paper, we propose to apply lane detection to the automated guided vehicle (AGV) platform in an industrial environment to realize automatic driving of AGV based on visual navigation. Similar to the lane detection of urban roads or highways, we present two methods for AGV navigation band detecting. The first method is based on traditional image processing which combines an adaptive threshold method with Sobel operator to detect the navigation band. The other method is based on deep learning, which implements end-to-end edge detection by LaneNet and it overcomes the problem of uneven illumination that the first method is struggling with. The RANSAC algorithm is finally used to fit the edge lines from the detected edge pixels in order to generate the AGV control commands to achieve robust real-time AGV automated driving.