Integration of visual information processing system for spatial perception by motion stereo vision

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K18044
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Electron device/Electronic equipment
• Research Institution: Tohoku University
• Principal Investigator: Akima Hisanao 東北大学, 電気通信研究所, 助教 (40707840)
• Research Collaborator: KAWAKAMI Susumu
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 運動立体視 / 空間認識 / LSI / 神経回路網モデル / Hough変換 / 視覚情報処理

Outline of Final Research Achievements

Spatial perception is indispensable to achieve such as self-driving cars and autonomous mobile robots. The purpose of this research is to construct a spatial perception system with high power performance ratio by developing an LSI for the neural network model of motion stereo vision in which spatial perception is performed based on visual motion. The optimal virtual connection schemes were adopted for each stage of the model to implement a huge number of neural connections into a limited chip area. A real-time spatial perception system was also developed for actual environment testing. As a result, the optical flow corresponding to an approaching object was successfully detected in 30 frames per second, and the power performance ratio of more than 100 times higher than a conventional CPU was achieved.

Free Research Field

工学