A 125,582 vector/s throughput and 95.1% accuracy ANN searching processor with Neuro-Fuzzy Vision Cache for real-time object recognition
Injoon Hong, Junyoung Park, Gyeonghoon Kim, Jinwook Oh, Hoi‐Jun Yoo · 2013
A fast and accurate Approximate Nearest Neighbor (ANN) searching processor is proposed to resolve the main bottleneck of the real-time object recognition process, the ANN searching. A new scheme, Spatio-Temporal Locality searching (STL-searching), is proposed to reduce the external memory bandwidth by at least 78x compared to Locality Sensitive Hash (LSH) scheme. However, the STL-searching suffers from low cache hit/miss decision accuracy, 52%. To improve the decision accuracy, a Neuro-Fuzzy Vision Cache (NFVC) with NFVC controller is proposed so that cache hit/miss decision can be made at 96% accuracy. It is implemented in 0.13μm CMOS process and achieves 125,582 vector/s throughput and 95.1% ANN searching accuracy, which are 2.02x and 1.32x higher than the state-of-the-art work.