2021 Fiscal Year Final Research Report
Deep Learning for Task-Driven Image Processing and its Application to Distant Pedestrian Detection
Project/Area Number |
19K12129
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 61030:Intelligent informatics-related
|
Research Institution | Toyota Technological Institute |
Principal Investigator |
Ukita Norimichi 豊田工業大学, 工学(系)研究科(研究院), 教授 (20343270)
|
Co-Investigator(Kenkyū-buntansha) |
Muhammad Haris 豊田工業大学, 工学(系)研究科(研究院), ポストドクトラル研究員 (60816643)
|
Project Period (FY) |
2019-04-01 – 2022-03-31
|
Keywords | 画像超解像 / 物体検出 / 画像処理 |
Outline of Final Research Achievements |
As an examples of task-driven image processing, we put our focus on tiny object detection and image super-resolution as a task and an image processing algorithm, respectively. Different from previous approaches in which these image processing and task are achieved independently, our proposed framework (i) integrates these two sub-problems in a single neural network and (ii) trains this network so that the reconstructed super-resolution image is optimized for improving the tiny object detection task. With this framework, we realize image super-resolution for generating images that are understandable by artificial intelligence instead of image super-resolution for images that are observed by human.
|
Free Research Field |
画像認識
|
Academic Significance and Societal Importance of the Research Achievements |
タスク指向超解像のための基礎技術として,汎用的な画像超解像を研究し,世界的な協議会でも上位入賞する性能を実現した.画像超解像は,人が鑑賞する画像を生成するという従来型のタスクにおいても,記録媒体に保存する画像ファイル容量の圧縮や,Youtubeや遠隔会議における映像配信など,多様な応用において実用的な技術である. また,本研究で実現した微小物体検出は,車載画像における遠方物体検出などセキュリティや安全を目的にした多様な分野に波及する技術である.
|