• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to previous page

Building a Video Search Engine based on the Perception of Spatio-temporal Relations

Research Project

Project/Area Number 19K12028
Research Category

Grant-in-Aid for Scientific Research (C)

Allocation TypeMulti-year Fund
Section一般
Review Section Basic Section 61010:Perceptual information processing-related
Research InstitutionKindai University

Principal Investigator

Shirahama Kimiaki  近畿大学, 理工学部, 准教授 (30467675)

Project Period (FY) 2019-04-01 – 2022-03-31
Project Status Completed (Fiscal Year 2021)
Budget Amount *help
¥4,290,000 (Direct Cost: ¥3,300,000、Indirect Cost: ¥990,000)
Fiscal Year 2021: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2020: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
Fiscal Year 2019: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
Keywords映像検索 / 物体の時空間関係 / グラフたたみ込み / 記憶伝達 / 強化学習 / TRECVID / 物体の時空間的関係 / 大規模映像検索 / 時空間的関係
Outline of Research at the Start

本研究では,映像中の時空間的関係に対する人間の知覚メカニズムをモデル化し,高次の意味に適合する映像を高精度に検索可能なシステムを開発する.具体的には,時空間的に近い物体の位置関係を反映したベクトルを抽出し検索を行う手法,フレーム間の意味的な連続性を考慮しながら情報(記憶)を伝達させて,映像をコンパクトなベクトルに圧縮し検索を行う手法,データから人間が無意識に行っている記憶伝達メカニズムを模倣する関数を学習する手法の開発に取り組む.

Outline of Final Research Achievements

This project addresses three main topics, 1) Video retrieval by considering spatio-temporal relations among objects, 2) Extraction of temporal features in a video by considering the continuities of semantic contents and 3) Learning a model that captures human memory mechanism for frames in a video. In particular, regarding the third topic, a reinforcement learning method has been developed to train a model that is based on a memory defined as a finite external storage and can update it so as to achieve the optimal understanding of contents in a video. In addition, this method has been extended to the data mining field, where a dataset containing a large number of items is targeted, and a model is trained to update a set of items in order to form statistically characteristic sets.

Academic Significance and Societal Importance of the Research Achievements

深層学習の導入によって、画像認識性能は大幅に向上したが、映像認識では、それほどの性能向上が得られていない理由として、画像認識で用いられているたたみ込みニューラルネットワーク(CNN)の演算が人間の知覚メカニズムとよく合致している一方で、時間を伴う映像に対しては、長短期記憶(LSTM)などの既存モデルの演算が、人間の時間知覚メカニズムに合致していない点が挙げられる。この問題に対して、行動心理学に基づいて、人間の意思決定をモデル化するために有用な強化学習という手法を用いて、映像の内容を適切に理解するための記憶伝達メカニズムを模倣するモデルを学習し、その有効性を実験的に示した点に学術的意義がある。

Report

(4 results)
  • 2021 Annual Research Report   Final Research Report ( PDF )
  • 2020 Research-status Report
  • 2019 Research-status Report
  • Research Products

    (13 results)

All 2022 2021 2020 Other

All Int'l Joint Research (2 results) Journal Article (3 results) (of which Int'l Joint Research: 2 results,  Peer Reviewed: 3 results,  Open Access: 3 results) Presentation (5 results) (of which Int'l Joint Research: 3 results) Remarks (3 results)

  • [Int'l Joint Research] リューベック大学(ドイツ)

    • Related Report
      2020 Research-status Report
  • [Int'l Joint Research] リューベック大学(ドイツ)

    • Related Report
      2019 Research-status Report
  • [Journal Article] Generic Itemset Mining Based on Reinforcement Learning2022

    • Author(s)
      Fujioka Kazuma、Shirahama Kimiaki
    • Journal Title

      IEEE Access

      Volume: 10 Pages: 5824-5841

    • DOI

      10.1109/access.2022.3141806

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Deep Transfer Learning for Time Series Data Based on Sensor Modality Classification2020

    • Author(s)
      Li Frederic、Shirahama Kimiaki、Nisar Muhammad Adeel、Huang Xinyu、Grzegorzek Marcin
    • Journal Title

      Sensors

      Volume: 20 Issue: 15 Pages: 4271-4271

    • DOI

      10.3390/s20154271

    • Related Report
      2020 Research-status Report
    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Rank Pooling Approach for Wearable Sensor-Based ADLs Recognition2020

    • Author(s)
      Nisar Muhammad Adeel、Shirahama Kimiaki、Li Frederic、Huang Xinyu、Grzegorzek Marcin
    • Journal Title

      Sensors

      Volume: 20 Issue: 12 Pages: 3463-3463

    • DOI

      10.3390/s20123463

    • Related Report
      2020 Research-status Report
    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Presentation] 強化学習を用いた映像フレームに対する記憶メカニズムの獲得2021

    • Author(s)
      西本光希,白浜公章
    • Organizer
      映像表現&コンピュータグラフィックス研究会(AIT)
    • Related Report
      2021 Annual Research Report
  • [Presentation] 個人差を考慮したスライドに対する瞬き率推定2021

    • Author(s)
      西本光希,佐々木海斗,白浜公章
    • Organizer
      2021年電子情報通信学会総合大会
    • Related Report
      2020 Research-status Report
  • [Presentation] マルチタスク学習を用いたセンサデータからの行動認識2021

    • Author(s)
      向井大貴,野見山哲平,白浜公章
    • Organizer
      2021年電子情報通信学会総合大会
    • Related Report
      2020 Research-status Report
    • Int'l Joint Research
  • [Presentation] Kindai University and Osaka Gakuin University at TRECVID 2020 AVS and ActEV Tasks2020

    • Author(s)
      Daiki Mukai, Ryosuke Utsunomiya, Shunsuke Utsuki, Kimiaki Shirahama, Takashi Matsubara and Kuniaki Uehara
    • Organizer
      TREC Video Retrieval Evaluation (TRECVID) 2020
    • Related Report
      2020 Research-status Report
    • Int'l Joint Research
  • [Presentation] Cross-modal Music-emotion Retrieval Using DeepCCA2020

    • Author(s)
      Naoki Takashima, Frederic Li, Marcin Grzegorzek and Kimiaki Shirahama
    • Organizer
      The Eighth International Conference on Information Technology in Biomedicine (ITIB 2020)
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Remarks] Wisteria30 / GIM-RL

    • URL

      https://github.com/Wisteria30/GIM-RL

    • Related Report
      2021 Annual Research Report
  • [Remarks] Deep Transfer Learning for Time Series Data

    • URL

      https://www.info.kindai.ac.jp/~shirahama/transfer/

    • Related Report
      2020 Research-status Report
  • [Remarks] Rank Pooling Approach for ADLs Recognition

    • URL

      https://www.info.kindai.ac.jp/~shirahama/rank_pooling/

    • Related Report
      2020 Research-status Report

URL: 

Published: 2019-04-18   Modified: 2023-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi