The agricultural sector is facing issues such as a decline in available workforce and increasing demand for crops, which heightens the need for technological innovations, for instance, robots that can improve production processes. Asparagus harvesting is one area where manual labor is highly demanding, and the delicate nature and unpredictable growth of asparagus make automation challenging.
In this master's thesis, we focused on improving an existing mobile robotic system for asparagus picking, with an emphasis on the detection and localization of asparagus. We prepared the mobile platform for fieldwork and improved the guidance of its parallel robot.
We focused on the process of detecting asparagus using robotic vision. We modified the sensor holder to allow for the installation of depth cameras on the mobile platform, in addition to the laser scanner. We enhanced the existing algorithm for identifying asparagus using data from the laser scanner, to enhance detection reliability.
We also added a process for recognizing and tracking asparagus based on color and depth frames from the depth camera using the YOLO algorithm. The algorithm detects and segments asparagus in color frames and then tracks them using the ByteTrack algorithm as they move. We optimized this process to run with sufficient speed for real-time operation. We used the detection data to determine the height of the asparagus and the picking point.
This work's results include the validation of asparagus picking in the field and a comparison of the detection capabilities of laser scanner and depth camera methods.
|