A ripe tomato recognition and localization system.

摘要

A real-time system for the recognition and localization of ripe tomatoes in greenhouses is developed, which includes binocular camera calibration, stereo matching, ripe-tomato recognition, and localization. It can automatically calibrate the binocular camera system, recognize and localize ripe tomatoes for the robot to pick off.;A low-cost calibration method for the binocular camera system is developed using asymmetric data flow and an asymmetric energy function. The method is capable of providing the position and orientation of the right camera with reference to the left camera with high consistency and accuracy, which requires no high-cost 3D measuring equipment. A new point sampling method is developed for the calibration algorithm, where the intersection of two straight lines is used as the sample point. An exponential evaluation function is built for line extraction, which considers more on the valuable oints in the line matching process, and makes the process consistent and fast. A new simple algorithm for binocular image rectification is developed, which transforms the original binocular image pair into an ideal one for stereo matching. The stereo matching segmentation algorithm classifies pixels into groups of inner pixels and edge pixels, which match both the characteristics and the functions of the pixels, making the stereo matching process faster and more reliable. A specialized simple equation for ripe tomato localization is derived based on the calibrated binocular camera system. Experimental results demonstrate the effectiveness of the developed system.