FPGA implementation of 2D cross-correlation for real-time 3D tracking of deformable surfaces

3D surface measurements have many industrial and medical applications. We have previously used 3D surface deformation and tracking to identify mechanical properties of the skin. To be able to detect dynamic changes in the surface of the skin we need to have a real-time 3D measurement system. A significant portion of the computation time for tracking the changes is spent during 2D cross-correlation of surface images. This study focuses on improving cross-correlation speed by taking advantage of parallel computation in field programmable gate arrays (FPGAs). We have implemented variable size 2D cross-correlation computations using the Xilinx System Generator tool in the Virtex-6 LX240T FPGA. We have also proposed a hierarchical approach for finding the cross-correlation peak in order to efficiently use our method for different image sizes. Furthermore, the use of RAM blocks instead of shift registers in our design has lowered the resource requirements compared with other FPGA implementations. Preliminary results for our special design indicate better than 200 times speed up compared with a PC with an Intel Xeon E5620 CPU (2.4 GHz clock speed, 4 cores and 8 threads) and 12 GB DDR3 RAM and also 190 times speed up in comparison to an NVidia GForce GT 525M as the graphics processing unit (GPU).