Stereo-vision-based crop height estimation for agricultural robots

Abstract

The objective of this study was to develop a machine-vision-based height measurement system for an autonomous cultivation robot. The system was developed with a simple stereo camera configuration to facilitate practical field applications and was used to acquire accurate height measurements of various field crops. The acquired stereo images were converted to disparity maps through stereo matching, and the disparity of each pixel was calculated to determine the depth of the distance between the camera and the crop. Depth maps were used to determine the edges of regions of interest (ROI) and the crop regions were segmented using the edges located in the expected crop region closest to the camera without using any additional labelling. The crop height was calculated using the highest points in the ROI. This approach was tested on five crops, and the results showed that the system could detect the target crop region even when objects overlapped in the acquired images. Furthermore, the crop heights estimated with the developed system showed strong agreement with actual crop heights measured manually, with the R2 ranging from 0.78 to 0.84. These results indicate that the developed algorithm is capable of measuring crop heights in various ranges for agricultural robot applications.