Experience is valuable in our life time. Nowadays, many people like to record their experiences by using digital media devices which is more efficient than ordinary writing a diary. We can capture digital media to keep in digital memory directly. In order to record personal experiences effectively, we attempt to create a wearable system to capture our experiences continuously. Due to the development of personal computer and digital media devices, we have implemented the wearable video system to record our experiences from the user's view point. The system is able to capture continuous video contents by a wearable video camera and record environmental contexts including location by GPS receiver, movement by motion sensors and physiological data by body sensors which we called this system as a "life log system".

　In this research, we focus on capture and retrieval of personal experiences by a wearable system. We develop a life log system which acts as a personal digital memory for capture and retrieval of personal experiences. The system has two stages including capture and retrieval. In the capture stage, life log data is captured by using the wearable life log system which includes a wearable video camera, a microphone, motion sensors, a GPS receiver and body sensors. We apply these wearable sensors to record some useful features for processing in the retrieval stage.

　In the experiments, the wearable system has been used to capture our experiences in daily life for more than two years. The system is used in various situations such as working, shopping, meeting, traveling, dinning, etc. We could find many interesting parts of our experiences from the life log video. Video data is recorded in MPEG format that its capacity is about 1 GB/hour. Totally, we record the video data almost 500 GB and keep them in a server. In real life, it takes a long time and it is almost impossible to watch all these video experiences. Thus we attempt to find the efficient retrieval techniques to recall these experiences from life log video.

　In the retrieval stage, we develop the video retrieval and navigation system to extract the key events from the life log video. Since the life log video is recorded continuously and has no shot or scene as other structured videos. We have to consider the practical methods to retrieve and navigate the experiences from life log data. User interfaces are designed to help people to recollect some desired events more comfortable than ordinary approach by key frame extraction and indexing techniques as a navigation system. We propose the retrieval techniques based on life log contents and contexts including audio visual and environmental data from wearable sensors. Firstly, we apply GPS data to extract some key events in traveling scenes including spatiotemporal sampling based on distance and time, speed detection and directional change detection. We find that these methods are useful to extract some key events in traveling scenes such as slowing down the speed and changing the direction. However, it has a limitation to use GPS data alone that the signal cannot be received inside the building and some obstructive location. Secondly, audio visual contents from life log video are analyzed to extract the key events with user's voice. It is advantage to acquire some information such as scenes with voice annotation and talking scenes. In the experiments, we assume that talking scene should contain some faces during talking period. We identify the talking scene by a basic face detection based on skin color. Body media armband, including physical and physiological sensors, can examine some physical activities. We also employ the body sensors to analyze physical and physiological data to extract the key events in life log video. The environmental change detection is proposed by using features from body sensors. Physiological features including heat flux, skin temperature and motion data are analyzed to detect the environmental change. Image processing based on histogram difference is used to ensure the accuracy. This technique is benefit to separate life log video into different environments that we can observe the scenes when user is moving from one place to another place.

　To extend the scope of experience representation, we also develop a media integration system to represent personal experiences. The retrieval scheme is flexible to retrieve personal experiences in form of wearable video, ordinary digital video, digital photo, physiological data, and electronic document in hierarchical order. We evaluate the system using subjective and objective evaluation. Each function is evaluated individually to see its efficiency. The overall system is evaluated by comparison with the other system in various aspects.

　Additionally, we realize Memex which was predicted by Vannevar Bush in 1945. We demonstrate an integration of our life log system and MyLifeBits system for continuous and discrete recording and retrieval of personal experiences. SenseCam, a passive capture camera is used to capture some parts of experiences. We also apply SenseCam sensor data to index the key frames of life log video. The relationship of life log video and SenseCam image is demonstrated. The system integration shows that our life log platform can gain the advantages for capture and retrieval of personal experiences.

　本論文は「A Study on Capture and Retrieval of Life Log by a Wearable System(ウェアラブルシステムによるライフログの取得と処理)」と題し，英文で書かれており，7章よりなる．情報機器の小型化，大容量化が進み，個人の日常(ライフログ)を克明に記録し，利活用する道が開きつつある．そのためには，膨大なライフログデータに対して，効率よく適切なデータへアクセスするための方策が必須となる．本論文では，映像，音響，位置，加速度などを取得するウェアラブルシステムを構築するとともに，映像を中心としたライフログデータを検索するための処理とシステムについて論じている．

　第1章は，「Introduction(序論)」であり，研究の目的，背景について述べている．

　第2章は，「State of the Art (技術動向)」であり，個人体験の取得と検索に関わる研究事例について述べるとともに，関連技術として，映像検索技術についてまとめ，そのインデキシング技術，キーフレーム抽出技術についてまとめている．

　第3章は，「Wearable Life Log Capture System(ウェアラブルライフログ取得システム)」と題し，構築したウェアラブルシステムについて論じている．そのシステムにおいては，小型PCをプラットホームとして，カメラによる映像，マイクロホンによる音響，GPSによる位置情報，アームバンドによる加速度，熱流束，温度が取得可能であり，そのシステム構成について詳述している．

　第4章は，「Life Log Retrieval System(ライフログ検索システム)」と題し，ウェアラブルシステムで取得したデータに対する検索システムの全体の枠組みについて論じている．膨大なコンテンツを検索するために，GPSによる位置，加速度センサによる動きなどのコンテキストデータが重要であることを指摘している．また，店などの街のデータベースとの連携での検索が有用であることも示している．検索のためのインタフェースについても簡潔にまとめている．

　第5章は，「Wearable Video Indexing and Navigation(ウェアラブルビデオのインデキシングとナビゲーション)」と題し，検索のためのキーフレームの抽出処理について論じている．コンテキストデータを処理することで，行動の変化点をとしてのキーフレームを抽出する．具体的には，位置，移動速度，移動方向に関しての抽出を行い，地域のランドマークの検出度合いを評価している．さらに，音声の検出による会話シーンの抽出，アームバンドの熱流束を用いた環境移動の検出(屋内と屋外の移動，部屋から廊下の移動など)を行い，行動における区切り目としてのキーフレームの検出に多様な選択を持たせている．

　第6章は，「System Integration(システム統合)」と題し，ウェアラブルシステムで記録した映像等のデータばかりでなく，デジタルカメラやカムコーダといった既存メディアによる記録データも併せて扱う統合システムを構築している．ウェアラブルシステム以外のデジタルカメラの静止画やカムコーダによる動画のデータは時刻情報で管理され，選択に応じて視聴することができるシステムになっている．さらに，PC上の様々なログを扱うMylifeBitsシステムを紹介し，本研究で取り扱う連続的なライフログデータとの連携について述べている．

　第7章は，「Conclusion(結論)」であり，まとめを述べるとともに，今後の課題について言及している．

　以上これを要するに，本論文では，ライフログデータに関して，取得のためのウェアラブルシステムを構築するとともに，膨大なデータに対しての効率的なアクセスの手段を開発し，既存メディアとの統合も行った．本論文で論じたライフログの処理は，今後のメディア技術の展開に寄与することが期待され，電子工学上貢献するところが少なくない．よって，本論文は博士(工学)の学位論文として合格と認められる．