Design team members: Timothy Noronha, Martin Rozee

Supervisor: Prof. Paul Fieguth

Background

Competitions entered by the Waterloo Arial Robotics Group (WARG) provide just one of many situations where an automatically generated three-dimensional mapping system could be of use. Three-dimensional maps of a scene can be used to optimize navigation routes, as well as to provide localization information about various obstacles or other regions of interest.

Project description

A 3D scene reconstruction system is proposed which can use data from various points in a scene to reconstruct a virtual representation of that scene. The data input into the system consists of a stereo image pair, position co-ordinates, and orientation information. All of this information can be obtained from sensors planned or currently implemented on WARG vehicles, though the scene reconstruction system will not be limited to this application.

To perform the task, the sytem must convert the stereo image pair to depth data representing the distance of objects from the camera. This is done in a way similar to how humans perceive depth. The depth data is used in conjunction with the position and orientation information to localize where in the world the observed obstructions exist. The final step is to store this information in a way that can be updated as new information becomes available (from additional viewpoints), and such that it can be easily determined how certain the system is that a given volume is empty or obstructed.