Local Texture Based Borderline Detection of Mowing

Abstract

With the development of automatic robots technology, it is using for more and more fields, mowing is still done manually. One of the important reasons of this is that it is difficult to automatically plan the mowing path. Therefore, we proposed a method for finding the boundary line between the cut grass and the uncut grass based on GLCM (grey-level co-occurrence matrix) for this problem. The features we extracted were energy, contrast, correlation and entropy by each GLCM, then using the average of each feature to segment the image of the grass. We used these features to find a boundary line that maximizes the difference in features on both sides. Finally, GA (genetic algorithm) is used to solve the maximization problem. The experiment shows that it is feasible to distinguish between the cut grass and the uncut grass using GA and GLCM features.