Vision-Based Mowing Boundary Detection Algorithm for an Autonomous Lawn Mower

This study proposes a vision-based mowing boundary detection algorithm for an autonomous lawn mower. An autonomous lawn mower requires high moving accuracy for efficient mowing. This problem is solved by using a vision system to detect the boundary of two regions, i.e., before and after the lawn mowing process. The mowing boundary cannot be detected directly because it is ambiguous. Therefore, we utilize a texture classification method with a bank of filters for classifying the input image of the lawn field into two regions as mentioned above. The classification is performed by threshold processing based on a chi-squared statistic. Then, the boundary line is detected from the classified regions by using Random sample consensus (RANSAC). Finally, we apply the proposed method to 12 images of the lawn field and verified that the proposed method can detect a mowing boundary line with centimeter accuracy in a dense lawn field.