Ground Plane Segmentation of Time-of-Flight Images for Asparagus Harvesting

A key task in processing time-of-flight images for the detection of asparagus spears is the identification and segmentation of the ground plane. In this paper, we aim to compare the performance of RANSAC with a modified version of an existing deterministic method, namely Hyun's method. Each method is tested on scenes with varying amounts of field clutter. Additionally, a variety of camera angles are investigated. We find that both RANSAC and the proposed method produce ground plane predictions with a root mean square error of less than 0.05m and execute at a rate of approximately 0.056s per frame. However, RANSAC was shown to be much less reliable in high clutter scenes. The camera mounting angle is found to significantly affect the density and noise of points in time-of-flight images. These factors translate to significantly worse performance for both methods at low camera angles.