Straight-line Generation Approach using Deep Learning for Mobile Robot Guidance in Lettuce Fields

This study proposed a deep learning-based approach to recognize various types of objects in images and generate optimal straight-line segments for mobile robots to perform heading corrections in complex environments. Object detection, based on a circular convolutional network framework, was utilized to identify various objects, such as watering strips, lettuce crops, or field furrows, in both the upper and lower regions of the image. Following the processing of multiple images, the center points of objects belonging to the same category were extracted, and a regression analysis method was used to generate a straight line. The slopes of these line segments are estimated, and the average value is calculated alïer determining the heading angle with the vertical line segment in the image through trigonometric operation. The flexibility and robustness of the straight-line detection system are enhanced by using the proposed approach.