The manipulation of deformable objects is an important problem in robotics and arises in many applications including biomanipulation, microassembly, and robotic surgery. For some applications, the robotic manipulator itself may be deformable. Vision-based deformable object tracking can provide feedback for these applications. Computer vision is a logical sensing choice for tracking deformable objects because the large amount of data that is collected by a vision system allows many points within the deformable object to be tracked simultaneously. This article introduces a template based deformable object tracking algorithm, based on the boundary element method, that is able to track a wide range of deformable objects. The robustness of this algorithm to occlusions and to spurious edges in the source image is also demonstrated. A robust error measure is used to handle the problem of occlusion and an improved edge detector based on the Canny edge operator is used to suppress spurious edges. This article concludes by quantifying the performance increase provided by the robust error measure and the robust edge detector. The performance of the algorithm is also demonstrated through the tracking of a sequence of cardiac MRI images.
Bibliographical noteFunding Information:
Acknowledgements I would like to thank Dr. Brian Mullan of the Division of Diagnostic Radiology at the University of Iowa for providing the Cardiac MRI images. This research was supported in part by the US National Science Foundation through grant numbers IIS-9996061 and IIS-0208564. Michael Greminger was supported by the Computational Science Graduate Fellowship (CSGF) from the US Department of Energy.
- Artificial neural networks
- Boundary element method
- Deformable object tracking
- Edge detection
- Robust statistics
- Robust tracking