Waypoint Generation based on Crop Row Detection Using Unet and Hough Transform

In autonomous agriculture, using wheeled mobile robots for various tasks requires the precise generation of waypoints to accurately define the robots' paths. This article presents a method that utilizes aerial imagery to detect crop positions and establish waypoints. In the image processing stage of this paper, crop detection was assessed using two methods of colorfiltering-based and Unet-based image segmentation, with Unet outperforming with 96% accuracy and 84% of IoU and showing satisfying results on different fields, considering the presence of unwanted vegetation. Subsequently, two crop row detection processes, including Linear Regression and the Hough Transform, were analyzed, presenting the convenient implementation of the Hough Transform compared to linear regression while having more accurate predictions. A mathematical conversion factor was derived to convert the position of pixels selected as path points to the global coordination systems. Finally, a specialized hardware setup comprising a high-resolution camera and a radio transmission system has been developed to capture and transmit live images of the agricultural field, enabling the system to support real-time crop detection and waypoint generation. Lastly, the potential of the introduced system is discussed to be implemented in future works.