Auckland Bioengineering Institute


Stereoscopic shape/strain measurement

bioinst-stereoscopic

The human brain is capable of qualitatively estimating the position and orientation of surfaces based on different views of the same scene with two eyes, as well as cues offered by cross-correlating surface patterns and textures. Quantitative measurements of 3D geometry can be obtained from multiple-camera systems. If two or more cameras are directed at a surface patterned with identifiable marker points, projective geometry can be used to estimate the 3D position of each point. In this case, however, each point must be clearly distinguishable from neighbouring points or the reliability of the method will suffer markedly. This typically requires that markers be relatively widely spaced so that only a small number of 3D points can be identified. Much more information can be obtained if the surface is patterned with random dots having average size of a few pixels in each 2D camera view. In this case every position has a unique pattern in its neighbourhood that can be readily identified.

Markers may be either projected on to the surface or fixed to the surface of the body. If the patterns are projected, it is difficult to obtain accurate information about surface deformation if the body of interest undergoes a shape change. Conversely, if patterns are fixed to the surface, both the geometry and surface strains can be accurately determined using correlation-based techniques to match different views of the same region of the body. These methods are computationally expensive, involving cross-correlation of multiple 3D geometrically transformed patterns to determine the goodness-of-fit from model-based estimates of the deforming body shape and texture to the patterns observed by multiple cameras. This goodness-of-fit measure contributes to the objective function in a nonlinear optimization of the model parameters to identify the model parameters that best characterize the current body deformation.

This project aims to implement novel cross-correlation algorithms on specialized hardware (GPU, DSP, FPGA) to rapidly determine the 3D deformation field of the neighbourhood of points on the surface of soft bodies, using information available from multiple 2D views of a randomly textured surface.
 

Project members


 

Publications


  1. Azhar, M., A.J. Taberner, M.P. Nash, and P.M.F. Nielsen. 3D material point tracking using phase based cross-correlation stereoscopy. in Twenty-sixth International Conference Image and Vision Computing New Zealand. 2011. Auckland, New Zealand.
  2. Parker, M.D., A.J. Taberner, and P.M.F. Nielsen. A thermal stereoscope for surface reconstruction of the diabetic foot. in Engineering in Medicine and Biology Society,EMBC, 2011 Annual International Conference of the IEEE. 2011.