Implementation and optimization of AKAZE algorithm in ARM processors and studying the algorithm acceleration using NEON

In this paper, AKAZE feature detection algorithm is implemented on a 7th generation ARM processor while it is tried to optimize the execution speed. As a first step, parts of AKAZE algorithm that are time-consuming and considered as bottleneck, are identified and rewritten using the C++ to increase the speed of the algorithm. Next, use of the NEON instructions in the ARM processor will be studied in the improvement of the execution speed of the algorithm, and parts of the program that can run simultaneously will be implemented using the specific commands of NEON. This study shows that rewriting time consuming parts of the algorithm will be very effective in terms of speed and the execution time of the algorithm with the proposed method is reduced up to 60 times compared to the existing codes. In addition, use of the NEON instructions in specific parts of the algorithm will be effective in improving the speed of program. The speed increase is very important in applications that need to process real time video images, and therefore can be useful in security systems and fast processing of camera images such as face recognition, license plate recognition, etc. required in FAVA operations.