Obstacle Recognition Using Multi-Spectral Imaging for Citrus Picking Robot

Abstract

To ensure that citrus picking robot operates safely in complex natural scenes, the branches must be recognized to prepare for avoiding obstacle and path planning. Due to the complexity of natural environment and the deficiency of identifying branches by the traditional methods, five narrow band filters in the Vis-NIR region were selected to capture five images. The first four principal component images were segregated the noise and extracted from the five filtered images using minimum noise fraction (MNF). The three-dimensional image data block was constructed to recognize branches using spectral angle mapper (SAM) classification. Experiment results show that the multi-spectral imaging technology combining with MNF-SAM can be used effectively to recognize branches in different lighting conditions.