Methodology for the objects recognition in range images using Kinect sensor

摘要

This paper presents the methodology used for the detection and recognition of some objects from the analysis of range images obtained from the Kinect sensor. Thus, is carried out the integration of a number of methodologies that are used daily in processes of vision and artificial intelligence, and from these, are intended to illustrate how efficient it can be the right choice of existing methods for the best results. In the segmentation stage it has been used the well-known Maximally-stable Extremal Region Extraction Algorithm, then, geometric characteristics of each object are extracted, and an evaluation is made of different classifiers. After the experimental tests it was possible to obtain a recognition rate of 99.4% through a multilayer perceptron neural network. All processing is performed using the openCV and openNI libraries.