Real-Time Color Object Recognition and Navigation for QUARC QBOT2

Abstract

QUARC QBOT2 is a ground robot manufactured by Quanser which has a combination of Mechatronics and Robotics courseware. Object detection can be done on QBOT2 by using Microsoft Kinect which is attached to the robot for vision. There are several object recognition techniques introduced which use image properties such as the ones that depend on RGB or HSV or other space properties. Other techniques are based on local binary patterns and binary sampling. Another approach is image descriptors which are based on detecting key points such as corners then matching the key points. One of the most important challenges in real-time applications is the delay. Thus, an improved technique has to be used for detection in order to achieve robustness plus accuracy in localization for the model object. In this work, an improved object recognition technique will be applied and tested on the QBOT2 and evaluated in terms of speed plus accuracy, view point, scale, rotation and illumination invariance.