3D reconstruction based on one camera applied to a rehabilitation of a human hand

Boris Simončič (2009) 3D reconstruction based on one camera applied to a rehabilitation of a human hand. EngD thesis.

PDF Download (9Mb)

Abstract

The subject of this work is a rehabilitation method for patients who suffer from arm or wrist injury with the help of a personal computer and a single web camera. Patient holds in his hand a simple object, cuboid, and moves it around. Camera records his movement while the software calculates position of the object in 3D space on the basis of color information and cuboid model. Program places this object in virtual 3D space, where another similar object is already present. The patient's task is to move the real object in the position, which matches the position of virtual object. Doing so he trains specific movements of his wrist and arm, which speeds up the recovery. This problem is thus interesting from both technical and practical point of view. The idea of rehabilitation method using computer and additional hardware is not new, but our solution differs significantly. We tried to make it accessible to as many people as possible. Thus we started with a limitation that only a personal computer and a single web camera is required. Web camera captures the scene, which is analyzed by our software in real-time. Software then tries to detect all the necessary characteristics to reconstruct the cuboid in 3D space. At the end of this work we also give the evaluation of the software, which is based on quantitative and qualitative analysis performed on a set of examples. On evaluation basis and under the condition that the user himself is acquainted with the basics of working with computers and is also willing to put in some effort on his own, we can conclude that presented solution serves its purpose good enough despite some faults. Presented solution is suitable mostly in cases, where accuracy is not critical and smaller deviations at 3D reconstruction do not thwart the process of the rehabilitation.

Actions (login required)