Date of Award
Master of Science in Computer Engineering
Department of Electrical and Computer Engineering
John M. Pecarina, PhD.
This paper describes our hardware accelerated, FPGA implementation of SURF, named Discrete SURF, to support real-time homography estimation for close range aerial navigation. The SURF algorithm provides feature matches between a model and a scene which can be used to find the transformation between the camera and the model. Previous implementations of SURF have partially employed FPGAs to accelerate the feature detection stage of upright only image comparisons. We extend the work of previous implementations by providing an FPGA implementation that allows rotation during image comparisons in order to facilitate aerial navigation. We also expand beyond feature detection as the complete Discrete SURF algorithm is run on the FPGA, rather than piped into processors. This not only minimizes overhead and increases the parallelization of the algorithm, but also allows the algorithm to be easily ported to different FPGAs. Furthermore, the Discrete SURF module is a logic-only implementation that does not rely on external hardware which therefore decreases the overall size, weight and power of the device while also allowing for easy FPGA to ASIC conversion. We evaluate the Discrete SURF algorithm in terms of performance against the original SURF and upright SURF algorithms implemented in OpenCV. Finally, we show how Discrete SURF is more compatible with an aerial navigation scenario than previous works, since rotation invariance must be considered in addition to scale.
DTIC Accession Number
Leighner, Andrew C., "FPGA Accelerated Discrete-SURF for Real-Time Homography Estimation" (2015). Theses and Dissertations. 42.