Three-dimensional object recognition: statistical approach

Abstract

The design of a general purpose artificial vision system capable of recognizing arbitrarily complex three-dimensional objects without human intervention is still a challenging task in computer vision. Experiments have been conducted to test the ability of incorporating the knowledge of how human vision system works in a three-dimensional object recognition system. Firstly, the process of shape outline detection and secondly, the use of multiple viewpoints of object. Shape outline readings are put through normalization and dimensionality reduction process using an eigenvector based method to produce a new set of readings. Through statistical analysis, these readings together with other key measures, namely peak measures and distance measures, a robust classification and recognition process is achieved. Tests show that the suggested methods are able to automatically recognize three-dimensional objects from multiple viewpoints. Finally, experiments also demonstrate the system invariance to rotation, translation, scale, reflection and to a small degree of distortion.