A 52 mW Full HD 160-Degree Object Viewpoint Recognition SoC With Visual V ocabulary Processor for Wearable Vision Applications Yu - C h i S u, Student Member , IEEE, Keng-Y en Huang, Student Member , IEEE, Tse-Wei Chen ,M e m b e r ,I E E E

Abstract

A 1920 1080 160 object viewpoint reco gnition system-on-chip (SoC) is presented in this paper. The SoC design is dedicated to wearable vision applications, and we address several crucial issues including the low recognition accuracy due to the use of low resolution images and dr amatic changes in object view- points, and the high power consumption caused by the complex computations in existing computer vision o bject recognition sys- tems. The human-centered design (HCD) mechanism is proposed in order to maintain a high recognition rate in dif ficult situations. To overcome the degradation of accur acy when dramatic changes to the object viewpoint occur, the object viewpoint prediction (OVP) engine in the HCD provides 160 object viewpoint in- variance by synthesizing various object poses from predicted object viewpoints. To achieve low power consumption, the visual vocabulary processor (VVP), wh ich is based on bag-of-words (BoW) matching algorithm, is used to advance the matching stage from the feature-level to the object-level and results in a 97% re- duction in the required memory bandwidth compared to previous recognition systems. Moreover, the matching efficiency of the VVP enables the system to support real-time full HD (1920 1080) processing, thereby improving the recognition rate for detecting at r a ffic light at a distan ce of 50 m to 95% compared to the 29% recognition rate for VGA (640 480) processing. The real-time 1920 1080 visual recognition chip is realized on a