Monocular visual odometry for robotic wheelchair in a virtual environment

Abstract

This paper applies to an adaptation of a Visual Odometry (VO) technique, originally created for outdoor real datasets, to an indoor virtual environment, and the integration of this environment with a dynamic model of a wheelchair. Thus, it were used three software; Matlab® running the dynamic model of the wheelchair, generating the pose vector for each step of the trajectory, and sending these values; Blender receiving the chair's position (x,y) and orientation (β), moving the robot to the specified position, rendering the images and saving them in a directory; and an OpenCV/C++ application using these images to estimate the traveled trajectory. The objective is to show a methodology to combine a model in Matlab® with a computer graphics platform, and a very powerful monocular visual odometry technique for outdoor environments, adapted to indoor, comparing with the ground truth values provided by this simulator. In addition, to the best of our knowledge, no adaption of this monocular method for indoor environments are available yet.