Navigating path recognition for greenhouse mobile robot based on K-means algorithm

Abstract: In a greenhouse with an unstructured environment, for the images collected by monocular vision, conventional path recognition algorithms are difficult to guarantee their robustness due to illumination variation, background reflection, shadow noise, etc. In addition, the increase of the amount of calculation of algorithms caused by the complicated background information of the greenhouse environment affects the quickness and the real-timeness of the greenhouse mobile robot autonomous navigation, which leads to the difficulty of meeting the requirement for the operation efficiency of the greenhouse mobile robot and impedes the practical application of the mobile robot technology in agricultural production. For the above problems, considering the influence of illumination conditions and complex background information in the greenhouse environment on the quality of the image segmentation, this paper focuses on the research of the color space selection and the image segmentation algorithm for a monocular vision greenhouse mobile robot. In order to not only reduce the impact of light information on the path recognition so as to improve the robustness of the algorithm, but also to enhance the accuracy of the path information recognition by adopting a novel image segmentation algorithm and meanwhile, reducing the calculation of the subsequent Hough transform so as to increase the quickness of path identification. Firstly, to ensure the robustness of the navigation path recognition algorithm in the greenhouse environment, three components H, S, and I are respectively separated from HSI color space, and the H component which has nothing to do with light intensity and can effectively restrain the effect of noise is extracted from the subsequent image processing. Secondly, to improve the rapidity of the greenhouse navigation path recognition and meet the real-time requirements of autonomous navigation operations, for the color characteristic of the greenhouse environment, the clustering segmentation of the image is performed based on K-means algorithm to achieve the respective clusters of the path and the green crop information. Then, the redundant and the interference information existing in the clustered image is eliminated by a morphological corrosion so as to obtain the complete and clear path information. Compared with a conventional threshold segmentation method, the proposed method can solve the problem of a too large memory occupation and a too long calculation time caused by the unclear segmentation information for the subsequent Hough transform, thus can enhance the rapidity of the greenhouse path recognition and meet the real-time requirements of autonomous navigation and operation of the greenhouse robot. Finally, in order to verify the effectiveness of the proposed method, the method in this paper, and the conventional method of the gray processing in RGB color space and the threshold segmentation are respectively used to process the greenhouse image information for comparison. The experiment results show that for the greenhouse robot working in the environment with a complex background and variable light, the proposed method can significantly reduce the effect of the non-uniform illumination on the navigation path recognition, that is, has a good robustness to the non-uniform illumination. Furthermore, the processing time of a single image is reduced by 53.26%, so the rapidity of the path recognition can be significantly improved.