For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
FPGA Implementation and Evaluation of a Real-Time Image-Based Vibration Detection System with Adaptive Filtering
Taito MANABE Kazuya UETSUHARA Akane TAHARA Yuichiro SHIBATA
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences
Publication Date: 2020/12/01
Online ISSN: 1745-1337
Type of Manuscript: Special Section PAPER (Special Section on VLSI Design and CAD Algorithms)
optical flow, Lucas-Kanade, BMFLC, FPGA, real-time,
Full Text: PDF(2.6MB)>>
This paper shows design and implementation of an image-based vibration detection system on a field-programmable gate array (FPGA), aiming at application to tremor suppression for microsurgery assistance systems. The system can extract a vibration component within a user-specified frequency band from moving images in real-time. For fast and robust detection, we employ a statistical approach using dense optical flow to derive vibration component, and design a custom hardware based on the Lucas-Kanade (LK) method to compute optical flow. And for band-pass filtering without phase delay, we implement the band-limited multiple Fourier linear combiner (BMFLC), a sort of adaptive band-pass filter which can recompose an input signal as a mixture of sinusoidal signals with multiple frequencies within the specified band, with no phase delay. The whole system is implemented as a deep pipeline on a Xilinx Kintex-7 XC7K325T FPGA without using any external memory. We employ fixed-point arithmetic to reduce resource utilization while maintaining accuracy close to double-precision floating-point arithmetic. Empirical experiments reveal that the proposed system extracts a high-frequency tremor component from hand motions, with intentional low-frequency motions successfully filtered out. The system can process VGA moving images at 60fps, with a delay of less than 1 µs for the BMFLC, suggesting effectiveness of the deep pipelined architecture. In addition, we are planning to integrate a CNN-based segmentation system for improving detection accuracy, and show preliminary software evaluation results.