Forests are acknowledged as major factors in CO2 absorption, which is vital for achieving carbon neutrality. In the forestry, it is necessary to further increase productivity and the use of unmanned machinery. As a means of increasing productivity, we focused on autonomous log-loading and -unloading using forestry machinery. Instance segmentation is an image processing technique in which objects are detected in pixels following labor-intensive training; it was used to investigate autonomous log-loading and -unloading. Here, a method is proposed for obtaining the grappling position of logs by detecting individual logs using object detection. Using this method, the three-dimensional coordinates of the logs were estimated in the radial, axial, and vertical directions with maximum errors of 1.644, 1.866, and 0.407 m, respectively, and root-mean-square errors of 0.643, 0.547, and 0.150 m, respectively. The proposed method has limitations to its applications and is unsuitable for precise operations. However, it could be used effectively for searching the surrounding logs in advance of precise segmentation because the average processing time from multiple object tracking to obtaining the three-dimensional coordinates of the estimated grappling position was 0.792 s.
View full abstract