Episodic Image Recording System for Memory Aid by Understanding Behavior and Situation
Grant-in-Aid for Scientific Research (C)
|Research Institution||HOKKAIDO UNIVERSITY|
KAWASHIMA Toshio Grad.School of Eng., Hokkaido Univ.Assoc.Prof., 大学院・工学研究科, 助教授 (20152952)
TANAHASHI Shin Grad.School of Eng., Hokkaido Univ.Inst., 工学研究科, 助手 (90250480)
AOKI Yoshinao Grad.School of Eng., Hokkaido Univ.Prof., 工学研究科, 教授 (90001180)
|Project Fiscal Year
1997 – 1998
Completed(Fiscal Year 1998)
|Budget Amount *help
¥2,200,000 (Direct Cost : ¥2,200,000)
Fiscal Year 1998 : ¥900,000 (Direct Cost : ¥900,000)
Fiscal Year 1997 : ¥1,300,000 (Direct Cost : ¥1,300,000)
|Keywords||Memory Aid / Prothesis / Episodic Memory / Understaing Behavior / 記憶 / 補綴 / エピソード / 行動理解|
In the project, we proposed a method for memory aid which utilizes image processing and data retrieval technologies, In the experimental system, image from user's view point taken by a CCD camera is analyzed by a computer to select impressive scenes from the image sequence.
The research is summarized in the following points.
(a) With the head-mount camera we recorded episodic images and evaluated the ability of these imagesto improve mental retrace of past events, In the experiments, we found the following facts.
A still image which views episode improves mental recall. Though the ability of mental recall depends on subjects, episodic image showed improvement on the recall experoment.
If the scene is complicated, events less related to the context of the situation often unrecalled. In such a case, an image which clearly discriminates the situation from the context made improvement in recall.
Structured still images of episodic scene recall the events if the scene was complicated. We organized the images taken by a CCD camera into a hierarchy of scene. The structured images dramatically improved the mental retrace.
(b) Algrothms to extract images closely related to episodes are proposed.
An algorithm which extract human-face from long-term video recordings is proposed.
An algorithm to monitor the motion of the user is proposed.
Algorithms to segment the long-term scene into hierarchy are proposed.
An algorithm to recognize human action from image is proposed. We concentrated on the algorithm to extract scenes in which the user to pick/place objects.
Research Output (15results)