Presentation 2018-12-07
Hardware Oriented Object Recognition Neural Network using Depth Image
Yuma Yoshimoto, Hakaru Tamukoh,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In recent years, deep learning using Convolutional Neural Network (CNN) has attracted attention as a powerful method for general object recognition. General object recognition is a task for recognizing objects from pictures of a real-world as a general name. General object recognition algorithm is required for service robots or embedded systems. In these cases, real-time processing is required, however, real-time processing by software is difficult. Therefore, high-speed operation by hardware is required. A Field Programmable Gate Arrays (FPGA) is a high-performance device with low-power consumption. To implement an algorithm into FPGAs, a hardware-oriented algorithm is required to reduce hardware resource utilization, because the resources of FPGA are limited. In this paper, we propose a dq{Binarized Dual Stream VGG-16 (BDS-VGG16)} which is one of the binarized convolutional neural networks using Binarized Neural Network (BNN) as a hardware-oriented algorithm for FPGA implementation. The proposed network is based on a Dual Stream VGG-16 (DS-VGG16) which realizes high accuracy by using RGB images and Depth images. The BDS-VGG16 is a combination of the DS-VGG16 and a Binarized Neural Network (BNN). In the experimental results show that the proposed algorithm is high accuracy. Also, the proposed method is efficient for FPGA implementation because the BDS-VGG16 completely remove the multipliers from the ordinary algorithm: therefore, the proposed algorithm can be implemented without multipliers.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Generic Object Recognition / FPGA / CNN / Depth Images
Paper # SIS2018-32
Date of Issue 2018-11-29 (SIS)

Conference Information
Committee SIS
Conference Date 2018/12/6(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Hagi Civic Center
Topics (in Japanese) (See Japanese page)
Topics (in English) Smart Personal Systems, etc.
Chair Takayuki Nakachi(NTT)
Vice Chair Noriaki Suetake(Yamaguchi Univ.) / Tomoaki Kimura(Kanagawa Inst. of Tech.)
Secretary Noriaki Suetake(Kyushu Inst. of Tech.) / Tomoaki Kimura(Tokyo Metropolitan Univ.)
Assistant Takanori Koga(National Inst. of Tech. Tokuyama College) / Hideaki Misawa(National Inst. of Tech., Ube College)

Paper Information
Registration To Technical Committee on Smart Info-Media Systems
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Hardware Oriented Object Recognition Neural Network using Depth Image
Sub Title (in English)
Keyword(1) Generic Object Recognition
Keyword(2) FPGA
Keyword(3) CNN
Keyword(4) Depth Images
1st Author's Name Yuma Yoshimoto
1st Author's Affiliation Kyushu Institute of Technology(KIT)
2nd Author's Name Hakaru Tamukoh
2nd Author's Affiliation Kyushu Institute of Technology(KIT)
Date 2018-12-07
Paper # SIS2018-32
Volume (vol) vol.118
Number (no) SIS-346
Page pp.pp.55-60(SIS),
#Pages 6
Date of Issue 2018-11-29 (SIS)