A Vision System Based on TOF 3d Imaging Technology Applied to Robotic Citrus Harvesting

Abstract

This study was conducted to develop a fast machine vision system based on TOF (time of flight) three-dimensional (3D) imaging technology for automatic citrus recognition and location. Supported algorithms were specifically developed and programmed for image acquisition and processing. An adaptive filter coupled between partial differential equations filter and shock filter was presented to remove noise and sharpen edges efficiently. Image analysis algorithms integrated both range and amplitude information to generate regions of interest and parameters that are characteristic or very likely to belong to spherical objects. The three-dimensional position of the fruit and radius are obtained after the recognition stages. The total classification results showed that approximately 81.8% were detected and false detection occurred only once. On the whole, the process of imaging, recognition and location consumes less than 50 ms/fruit.