A practical solution for ripe tomato recognition and localisation

Abstract

This article presents straightforward and fast algorithms that achieve low-cost computation of a software/hardware process for a practical robot vision solution for ripe tomato picking in greenhouses. The solution has three steps. First, it uses a single-colour component V in a YUV colour space to segment out the ROI (region of interest) of ripe tomatoes from the background, green stems and foliage. Second, an edge extractor with a constrained curvature is developed to extract the arc edges within a certain radius range based on the ripe tomato model for calculating the central points of each ripe tomato that was detected in the first step. Finally, according to the circle---circle relationship found in the geometric analysis, the overlapping area between two ripe tomatoes is calculated, and the depth ordering of the two overlapping ripe tomatoes is determined. The key contribution of this proposed solution is the real-time recognition and localisation of ripe tomatoes that are covered by green foliage, stems and unripe tomatoes in the real world. Simplicity and practicality are the key considerations of this solution.