Medial axis-based machine-vision system for orchard robot navigation

In this paper, we propose a novel medial axis-based machine vision technique for autonomous path tracking and navigation of an agricultural robot in orchards. In the proposed method, the raw color image captured by the onboard camera is converted to grayscale using the color index of vegetation extraction. The gray image is run through Gabor filters to extract the texture features before applying Principal Component Analysis technique to minimize the computational load during path coordinate extraction. The resulting image is clustered using Kmeans clustering algorithm with k = 2 as proved best based on Silhouette method. Navigation path is then extracted using medial axis algorithm from the binary image generated after k-means clustering. Fuzzy logic controller with two inputs, heading and offset, is used to smoothly track the medial axis during navigation. Experiment is carried out in an orchard environment to gauge the performance of the system. The results of field experiment clearly demonstrate that the proposed medial axis technique has the potential to accurately extract guidance path for robot navigation. The navigation performance of the system is quite satisfactory with a maximum trajectory tracking error and standard deviation of 14.6 mm and 6.8 mm respectively. The average root mean square error (RMSE) for the lateral deviation was 45.3 mm.

Automated summary

The paper proposes a novel medial axis-based machine vision technique for autonomous path tracking of an agricultural robot. The technique involves converting color images to grayscale, extracting texture features, clustering images, and using fuzzy logic controllers. Experimental results demonstrate the system's potential for accurate navigation in orchard environments.