Language：English   Publisher：The Society of Instrument and Control Engineers  

This paper proposes a new device to capture dense point cloud of human contour using horizontal LIDAR and vertical LIDAR with mirrors on pan rotational mechanism. This paper also reports a new method to estimate simple behavior such as standing and sitting online from only single vertical scan data. The combination of rotation mechanism and vertical LIDAR enlarge measurement area to overall room. Both mirrors use and pan rotation control enhance density of point cloud. The horizontal LIDAR provides robust tracking of human without missing. The experiments about performance of the device and the method demonstrated that our device and method contain sufficient feasibility to monitor human behavior in human living environments. The authors also demonstrated online recognition of simple human behavior such as standing and sitting from only single vertical scan points using pattern recognition technique.

DOI： 10.9746/jcmsi.7.12

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.20965/jrm.2013.p0005

Scopus

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

DOI： 10.1145/2494091.2494153

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/huc/ubicomp2013ap.html#FukuiMWSS13

Publishing type：Research paper (scientific journal)  

DOI： 10.20965/jrm.2013.p0726

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1080/01691864.2013.776939

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/ar/ar27.html#FukuiWSS13

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/IROS.2013.6696738

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iros/iros2013.html#FukuiSKWSS13

Publishing type：Research paper (scientific journal)  

DOI： 10.20965/jrm.2012.p0320

Scopus

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-642-33885-4_25

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/eccv/eccv2012w3.html#OdashimaSKFS12

Publisher：Springer  

DOI： 10.1007/978-3-319-00065-7_2

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

researchmap

Other Link： https://dblp.uni-trier.de/rec/conf/icpr/2012

Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

DOI： 10.1145/2370216.2370439

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/huc/ubicomp2012.html#ShimosakaMTFS12

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-642-33885-4_26

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/eccv/eccv2012w3.html#KanekoSOFS12

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/AIM.2012.6265946

Scopus

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：JMLR.org  

researchmap

Other Link： https://dblp.uni-trier.de/rec/conf/acml/2012

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-642-31205-2_5

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/pervasive/pervasive2012.html#TominagaSFS12

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-319-00065-7_62

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iser/iser2012.html#FukuiWSS12

Language：English   Publisher：Society of Automotive Engineers of Japan, INC  

In Japan, a report has shown that being hurry is the greatest category of the driver mental states during caraccidents. In this paper, we present the analysis and detection algorithm especially developed for hurry driving in several driving situations. The most effective driving feature in each situation is selected based on the comparative statistical analysis of the driver behavior in the normal and hurry driving conditions. It is found that the features that classify the hurry driving well in most situations are the time headway, the host vehicle velocity and the longitudinal acceleration. Through these features, the appropriate classification rules are determined, which allows us to later develop the hurry-driving detection algorithm based upon the driving-state transition model. The algorithm is then validated by offline simulations with collected driving data. The results clearly confirm the practicability and effectiveness of the proposed algorithm.

DOI： 10.20485/jsaeijae.2.3_69

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

DOI： 10.1145/2030112.2030154

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/huc/ubicomp2011.html#FukuiWGSS11

Language：Japanese   Publisher：The Japan Society of Mechanical Engineers  

In this paper, we propose developed three-dimensional distance measurement device for human posture estimation in indoor environment. Our developed device is composed of a horizontal rotary mechanism and a vertical two-dimensional plane distance measurement sensor with mirrors. In addition, our developed device can scan a person fast and densely by tracking human location wherever a person is located at any room place. Because of these advantages, our developed device can measure human shape robustly. In experimental result, our developed device can measure three-dimensional human surface model during tracking a person.

DOI： 10.1299/jsmermd.2011._1p1-e01_1

CiNii Research

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/SII.2011.6147537

Scopus

researchmap

DOI： 10.20965/jrm.2010.p0221

researchmap

Language：Japanese   Publisher：一般社団法人 人工知能学会  

多数の独居高齢者の住居に設置した焦電センサの長期蓄積データから得られる生活の傾向から，その特徴を端的に表す基準を自動的に抽出し，類似した生活を行う人の生活・異変発生履歴をも参照して生活異変を検知する手法を提案する．見守りサービスのコールセンターで稼働させることを想定し，オペレータや看護士にとって分りやすい検知結果提示が可能な生活特徴に着目した生活異変検知を行うシステムを開発している．

DOI： 10.11517/pjsai.jsai2010.0_1h2nfc3b4

CiNii Research

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/IROS.2008.4650864

Web of Science

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1007/978-3-540-85845-4_13

Web of Science

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/IROS.2007.4399181

Web of Science

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/IROS.2007.4399094

Web of Science

Scopus

researchmap

Publisher：The Japan Joint Automatic Control Conference  

DOI： 10.11511/jacc.49.0.429.0

researchmap

Language：English   Publisher：Information and Media Technologies Editorial Board  

In this paper, we propose a novel kernel computation algorithm between time-series human motion data for online action recognition. The proposed kernel is based on probabilistic models called switching linear dynamics (SLDs). SLD is one of the powerful tools for tracking, analyzing and classifying human complex time-series motion. The proposed kernel incorporates information about the latent variables in SLDs. The empirical evaluation using real motion data shows that a classifier using SVM with our proposed kernel has much better performance than the classifiers with some conventional kernel techniques. Another experimental result using kernel principal component analysis shows that the proposed kernel has excellent performance in extracting and separating different action categories, such as walking and running.

DOI： 10.11185/imt.1.314

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1142/S0219843604000368

CiNii Research

researchmap

J-GLOBAL

researchmap

DOI： 10.1109/SII.2012.6427320

Scopus

researchmap

Language：Japanese   Publisher：東京 : 電子情報通信学会  

CiNii Books

CiNii Research

researchmap

Other Link： http://id.ndl.go.jp/bib/023370780

DOI： 10.20965/jrm.2011.p0494

Scopus

researchmap

Language：Japanese  

本論文では動画群の中から特定のシーンをマニュアルで探す操作を効率的に行えるようにするために複数の動画を同時に操作できるシステムを提案する．システムの操作をユーザの手の動きに着目したジェスチャで行うことを考え，環境設置型のセンサを用いて手先の三次元位置をリアルタイムで推定する．片手の動きの速度に着目したジェスチャデザインにすることで動きと動画の操作を結びつける．また提示部の表示構成をドック型にするなどの工夫により同時・複数人でも使いやすいインタフェースとした．このシステムを用いることによって動画のシーン探索の効率が向上することを実際の使用シーンに則した実験により示す．At video sharing sites, users have tasks to look for specified scenes manually from some videos after searching with keywords from a lot more movies. In this paper, we propose a system which allows users to operate multiple movies at the same time using hand gestures. We use environment-embedded sensors for measurement of the hand position in 3d space to recognize gesture and designed effective gesture and dock styled presentation part to realize simultaneous operation of multiple videos for one or more users. Experimental result shows the effectiveness of the method.

CiNii Books

researchmap

Language：Japanese  

混雑環境において人数推定を行うことは、都市環境における安全面や商業面の観点から役立つ．近年，人周りの領域の面積などといった，画像全体から得られる特徴量を用いて，回帰により人数推定を行う手法が注目されている．この手法の推定精度は，用いる人周りの領域抽出手法・特徴量に依存する．そこで本研究は，適切な領域抽出手法・特徴量を自動的に選出する学習法を提案する．提案手法は，特徴量選出の 2 段階のアルゴリズムから成り立ち，それぞれの段階に疎な解を持つことでしられる l1 ノルム正則化を用いる．Lars という正則化パスを効率的に求める考え方により，提案手法はハイパーパラメータ決定を含む学習時間が他の手法より短い．本研究では，公開データセットを用いた実験を通し，提案手法の有用性を検証した．Counting people in crowded environments is valuable for applications such as traffic analysis, safety in urban or advertising. Recently, number of people in the scenes can be estimated using holistic image features. The performance of this approach depends on the combination of motion segmentations and image features. Therefore, we propose a learning method which selects only useful features and segmentations from a large number of those automatically. Our mainly consists of two stages. Our method contains l1-norm regularization, which is known as a shrinkage and selection technique, in two stages. Thanks to using Lars methods, which calculate a regularization path efficiently, the proposed approach can determines a hyper-parameter under less training time. Experimental results show our sufficient estimation accuracy and reduction of required features within less training time.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

We proposed an imploved localization method using simple map that rider can obtain easily before driving, laser range data and wheel rotation. Our method consists of (a) localization on the simple map, (b) incremental map refinement, (c) resizing the simple map to use multiple size maps. Experiments showed that our method improve localization accuracy using prior information by the simple map.

CiNii Books

researchmap

Language：English  

DOI： 10.1063/1.3596654

Web of Science

Scopus

researchmap

Language：English  

DOI： 10.1109/ACPR.2011.6166650

Web of Science

Scopus

researchmap

Language：English  

DOI： 10.1109/IROS.2011.6048095

Web of Science

Scopus

researchmap

Language：English  

DOI： 10.1109/IROS.2011.6048858

Web of Science

Scopus

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

In this paper we propose a novel method for predicting a resident's behavior in a house by one's movement trajectories. The proposed method consists of 1) segmentation of trajectory data into staying or moving and classification of segments and 2) prediction by time-series association rules from transition events of each segment. The experimental results using real resident's trajectory data of almost two years demonstrate that 1) typical staying location are affected by floor plan and furniture layout and 2) prediction of behavior by the proposed method are possible.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper describes development of search and visualization system that deals with heterogeneous time-series activity sensor data in distributed databases. We have collected different types of activity data from living enviroments of elderly people in south area of Kashiwa city: walking data by pedometers, indoor actions by pyroelectric sensors and transportation utility data by on-demand bus reservation. Different search methods are prepared for all databases and these methods extract activity history data in a unified form. This mechanism enables the search system to deal with the heterogeneous activity databases uniformly. In addition, the system provides the method for extracting person group by statistics of extracted activity history. The system facilitates activity analysis through searching action history data from diffrent data sources and visualizing these data comprehensibly. Two experiments showed that the system enables cross-sectional search from heterogeneous time-series activity databases: 1) Identification of the same person from different activity databases and 2) Extraction of person group from activity history data.

CiNii Books

researchmap

Language：Japanese  

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

In this paper, we propose a novel framework for personalized driver behavior modeling with Bayesian learning methods. Specifically, the proposed framework, formulated as Dirichlet process mixtures, achieves simultaneous categorization, identification and adaptation of driver models from driving records of multiple drivers. We also provide a histogram quantization technique to accelerate the speed of inference process. The experimental results using driving records of multiple drivers under car following situation show that our method is superior over the other conventional methods in view of accuracy and inference efficiency.

CiNii Books

researchmap

Language：Japanese   Publisher：人工知能学会  

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper deals with sensor arrangement for activity classification of the people, who lives alone, with pyroelectric sensors, so that the classification system acquires as high performance with as small sensor numbers as possible. We suggest some heuristic algorithms for approximate optimization of sensor combination, and some machine learning algorithms for exact optimization of feature selection which corresponds to that of sensor selection. For some examinations, we confirmed the significance of automatic arrangement system by comparing the quasi-optimized sensor arrangement acquired by above algorithms with the arrangement by human judge.

CiNii Books

researchmap

DOI： 10.1109/INSS.2010.5573463

Scopus

researchmap

DOI： 10.1109/ICPR.2010.1066

Scopus

researchmap

Language：English  

DOI： 10.1109/ICME.2010.5583546

Web of Science

Scopus

researchmap

Language：English  

DOI： 10.1109/IROS.2010.5651516

Web of Science

Scopus

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

Language：English  

DOI： 10.1080/00423111003668229

Web of Science

Scopus

researchmap

Language：English  

DOI： 10.1109/IROS.2009.5354135

Web of Science

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

This paper deals with the optimization of sensor arrangement for activity recognition of the people living alone with pyroelectric sensors. This system sets some laser range finders in the house during a few beginning weeks in order to simulate pyroelectric sensor at any place by acquiring the life-flow which is unique for the person, then it searches the optimal arrangement by comparing any arrange pattern virtually. In this paper, we show that the system can acquire approximated optimal sensor arrangement for practical calculation order by comparing some kinds of typical approximation optimization, and considers the selected way.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

This paper proposes an object detection method in indoor environments. With object placement and removal, the input image changes stably, so we apply background subtraction for object detection. In order to distinguish object placement and removal from background subtraction result, layered background model is employed. The stable change of input image is also caused by "temporary stable" regions like human regions, so classifying "placed object" regions and "temporary stable" regions is needed. To achieve a robust classification under occlusion, we employ feature point based tracking and partial frame subtraction. Our experiment demonstrates the system detects object placement and removal appropriately in sufficient frame-rates.

CiNii Books

researchmap

Language：English  

DOI： 10.1163/156855309X408763

Web of Science

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

Laser based tracking systems have been developed for mobile robotics and intelligent surveillance areas. Existing systems estimate only human positions. In this paper, we propose a method for human pose estimation represented by human head and waist position using only laser range finders. Two features of human contour are extracted from laser scanner data scanning on the height of torso. This method estimates human pose by using these features in the Bayesian filtering framework. Moreover, we apply particle filter with multiple transition models. Our experimental results demonstrate the effectiveness of our method against movements of multiple people.

CiNii Books

researchmap

Language：English  

Web of Science

researchmap

Language：English  

DOI： 10.1109/ROBOT.2009.5152847

Web of Science

Scopus

researchmap

Language：English  

DOI： 10.1109/ICME.2009.5202594

Web of Science

Scopus

researchmap

DOI： 10.1145/1667780.1667799

Scopus

researchmap

Language：English   Publisher：一般社団法人 日本機械学会  

Rear-end collision in the vehicle-following situation has captured vast interests of active safety researchers. This paper presents one of the major causes of this collision, a hurry driving. The behavior of the driver during the normal and the hurry driving has been analyzed comparatively, and the statistical models of the driver behavior are developed. These model features are the moving average and the moving standard deviation of the time headway. A classifier for hurry driving diagnosis is produced based on the bivariate distribution model between these features. The results clearly distinguish normal and hurry conditions in terms of time headway and its standard deviation.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

Successful multi-target tracking requires locating the targets and labeling their identities. For the tracking systems, the latter is more challenging. In this paper, we propose a method for multiple persons tracking using multiple cameras and laser range finders. Our method estimates 3D positions of human body and head, and labels their identities. The method is composed of multiple particle filters, and each particle filter tracks each person correctly by integrating information from laser range finders and the target-specific information from multiple cameras. Integration of these two types of sensors enables complement of each weak point and the correct tracking of the target. Moreover, we develop a new particle filter framework that tracks the human head by using the estimated human body position simultaneously. Our experimental results demonstrate the effectiveness and robustness of the method in multiple persons tracking.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper describes a method for estimating the position/pose of a mobile robot and the positions of people in a previously mapped environment. In the proposed method, we acquire LRF scan segments not corresponding to occupied grids of the known map and consider these segments as environmental changes caused by people and other moving-objects. By using these LRF scan segments and particle filter framework, our method estimates the robot and multiple people robustly. In addition, other moving-objects except people are detected by tracking these LRF scan segments. In our experiment, positions of the robot and people were estimated about 10cm accuracy.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper reports a method of object location estimation with Active RFID tag in real indoor environment. A robust indoor object localization system is realized based on the complementary combination of 1) received signal strength indicator(RSSI) and vibration data acquired from Active RFID tag, and 2) human behavior detected from various kinds of sensors embeded in environment. Experiment results show that our object localization algorithm can provide high performance in real living environment.

CiNii Books

researchmap

DOI： 10.1145/1667780.1667828

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人 日本ロボット学会  

In this paper, a novel approach is proposed to recover human body pose from 3D voxel data. The use of voxel data leads to viewpoint-free estimation, which benefits in that reconstruction of a training model is needless in different multi-camera arrangements. The chief advantage of our approach is speed, which enables real-time processing when capturing 8 VGA size images in 30 [fps] . Our approach is mainly based on an example-based approach. Human posture candidates are constructed beforehand, and the most appropriate posture candidate is selected per frame by comparing the similarity between 3D voxel data and posture candidates. Derivation of similarity is formulated by introducing a histogram-based feature vector that represents the 3D context of human body. In addition, a fast near-neighbor search metric is installed prior to the evaluation process, to reduce the computational cost and ensure real-time processing. Estimation stability is also improved by a motion graph, which adds a smoothing effect to the motion sequence. We demonstrate the effectiveness of our approach with experiments on both synthetic and real image sequences.

DOI： 10.7210/jrsj.26.913

CiNii Books

researchmap

Language：Japanese  

CiNii Books

researchmap

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper describes a novel wireless sensor network system that provides both functions of sensor value transmission and sensor localization. ZigBee technology is utilized as wireless network protocol in our system because of its advantages in terms of power consumption and easiness of self-organizing mesh network. For location estimation, Received Signal Strength Indicator (RSSI) is leveraged and a robust grid based sensor localization algorithm is proposed. A testbed location aware ZigBee-based system is successfully evaluated in terms of its performance of sensor value transmission and sensor localization functions.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

In this paper, we propose a household object recognition system based on computer vision. In our system, detailed images of object are acquired with four pan-tilt cameras fixed at the ceil of the room. The object is identified from these images, and the object's position is managed by using the recognition result. Also, the system presents the position of the household objects by user's request. This paper presents a robust object recognition method for changes of object size, angle, direction in images by using SIFT feature and object template images captured from various directions. Our experiment with 41 objects showed that precision was 0.94, recall was 0.66, and F-measure was 0.75.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper describes a training course for 3rd-grade undergraduate students at the department of Mechano-Informatics, the University of Tokyo. This course is designed to facilitate students learn broad regions of information engineering and robotics. It contains robot's behavior planning, controller design, electronics for robotics engineer, and software programming on embedded systems, computer vision and 3D computer graphics. At the closing stage of the course, a creative project is organized to improve engineering skills of the students. In this paper, we introduce syllabuses of each issue in the course, and also report interesting examples and discussions of creative project.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This papar presents a software interface that enables users to control home robots remotely. To control robots easily and safely, we propose users using a remote control interface based on Mixed Reality. The software interface displays the simulation how robots will move and requires confirmation of their operations. The simulation is one of the easiest ways for users to acquire the result from their operation, and confirmation will prevent robots from acting out of users' expectation. This interface is also useful for developers for robots. It visualizes the sensors which are attached to the robots and it also visualizes status of robots. When the developers utilize these data, it will help them to understand the meaning of these data. We implemented the software interface and demonstrated it. It enabled us to control home robots easily and safaly.

CiNii Books

researchmap

Language：English  

DOI： 10.1109/INSS.2008.4610899

Web of Science

Scopus

researchmap

Language：English  

Web of Science

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper presents a behavior labeling algorithm for time-series data, extracted using home sensors. In the previous work, we proposed a time-series data clustering method based on Hidden Markov Models (HMMs). However, change points of behaviors are not clearly segmented in our former method and short length behaviors tend to be misrecognized. In this paper, we propose a novel behavior labeling algorithm by introducing Singular Spectrum Transformation (SST) to detect change-points robustly. This algorithm enables more precise behavior labeling because of clear change-point detection.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper describes an algorithm of behavior labeling and anomaly detection for elderly people living alone. In order to grasp the person's life pattern, we equip some pyroelectric sensors into the house and measure the person's movement data all the time. From those sequential data, we extract two kinds of information, time and duration of each behavior, and make the system calculate two-dimensional probabilistic density function of them. By using this function, the system classifies behavior labels and detects anomaly. In addition to these two kinds of information, we consider another kind of informaion, behavior transition patterns, at anomaly detection. Here, we assume local anomaly and global anomaly. The former means the rare behaviors and the latter means the changes of life pattern. The algorithm is confirmed through the experiment on about 400 days real behavior data.

CiNii Books

researchmap

Language：Japanese   Publisher：The Japan Society of Mechanical Engineers  

This paper describes a robot programming course for 3rd-grade undergraduate students at the department of mechano-informatics, the university of Tokyo. This course includes many issues for building a mobile robot's behavior programming: Programming in embedded Linux environment, image processing, inverse kinematics for an arm and an omnidirectional mobile robot, etc. These issues are integrated as mobile robot's behaviors where robots explore the environment using vision and other sensors and act on the environment using omnidirectional wheels and an arm.

DOI： 10.1299/jsmermd.2008._2A1-J21_1

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本ロボット学会  

In this paper, we propose a robust online action recognition method based on boosted sequential classification. Our method utilizes boosting algorithm that is one of ensemble learning algorithms. This algorithm is also known as a feautre selector and has been utilized in the fields of image processing and natural language processing in recent years. Based on the boosting scheme, our method can automatically and efficiently select significant features for action recognition. Additionally, the method leverages temporal dependency of actions based on Ising model to improve recognition performance. We evaluated our method to action recognition, such as walking and running, using motion capture data only with posture features. In the result, our method can classify the actions more robustly than the method that does not utilize temporal dependency of actions.

DOI： 10.7210/jrsj.25.906

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

We have been working on a marker-less motion capture system that works in a multiple camera environment. This system assumes 3D voxel data to be the input, while discrete human posture data is assumed to be the output. The discreteness of human posture data is provided by an example-based approach, which constructs human posture candidates from a large motion capture database beforehand. During the estimation process, the most appropriate candidate will be chosen through a simple similarity calculation between voxel data and posture candidates. This approach will drastically reduce the computational cost compared to conventional methods, but increase of candidates will possibly lead to considerable computational cost. Therefore, prior to the similarity calculation phase, we introduce a near-neighbor search metric, which drastically decreases the similarity caculation frequency and the total computational cost. In this paper, we present a novel near-neighbor search metric, which is capable of dealing with much more candidates than the metric presented before, and yet maintaining the speed needed for online processing.

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

We have been working on a marker-less motion capture system that works in a multiple camera environment. This system assumes 3D voxel data to be the input, while discrete human posture data is assumed to be the output. The discreteness of human posture data is provided by an example-based approach, which constructs human posture candidates from a large motion capture database beforehand. During the estimation process, the most appropriate candidate will be chosen through a simple similarity calculation between voxel data and posture candidates. This approach will drastically reduce the computational cost. compared to conventional methods, but increase of candidates will possibly lead to considerable computational cost. Therefore, prior to the similarity calculation phase, we introduce a near-neighbor search metric, which drastically decreases the similarity caculation frequency and the total computational cost. In this paper, we present a novel near-neighbor search metric, which is capable of dealing with much more candidates than the metric presented before, and yet maintaining the speed needed for online processing.

CiNii Books

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

In this paper, we propose a robust online action recognition method based on boosted sequential classification. Our method utilizes boosting algorithm that is one of ensemble learning algorithms. This algorithm is also known as a feautre selector and has been utilized in the fields of image processing and natural language processing in recent years. Our method automatically and efficiently selects significant features for action recognition. Additionally, the method assumes interdependency between action labels based on Markov random fields. We evaluated our method to action recognition, such as walking and running, using motion capture data only with posture features. In the result, our method classified the actions more robustly than other methods that do not utilize interdependency between action labels.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

This paper presents an online supervised segmentation method of daily actions like walking and sitting. The proposed method utilizes switching models of two dynamical systems to represent segmental boundary of human motion. It consists of three phases based on the pattern of time-series likelihood of the switching models. 1)Candidates of the segmental points are detected without omission. 2)The candidates are refined by matching algorithm using time-series likelihood of the correct segmental boundary. 3)The segmental points are output from the candidates by considering latency due to the decision. The experimental results using real motion capture data show that the proposed method are effective for online segmentation of daily actions.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

In this paper, we propose a robust online action recognition method based on boosted sequential classification. Our method utilizes boosting algorithm that is one of ensemble learning algorithms. This algorithm is also known as a feautre selector and has been utilized in the fields of image processing and natural language processing in recent years. Our method automatically and efficiently selects significant features for action recognition. Additionally, the method assumes interdependency between action labels based on Markov random fields. We evaluated our method to action recognition, such as walking and running, using motion capture data only with posture features. In the result, our method classified the actions more robustly than other methods that do not utilize interdependency between action labels.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本ロボット学会  

In this paper, we propose a robust online action recognition algorithm with a segmentation scheme that detects start and end points of action occurrences. Specifically, the alogorithm estimates reliably what kind of actions occurring at present time. The algorithm has following characteristics. (1) The algorithm incorporates human knowledge about relations between action names in order to toughen the recognition, thus it labels robustly multiple action names at the same time. (2) The algorithm uses time-series Action Probability that represents the likelihood of each action occurrence at every frame time. The Action Probability is obtained from time-series human motion using support vector machine. (3) The algorithm can detect robustly and immediately the segmental points using classification technique with hidden Markov models (HMIs) . The experimental results using real motion capture data show that our algorithm not only prevents the system from making unnecessary segments due to the error of time-series Action Probability but also decreases effectively the latency for detecting the segmental points.

DOI： 10.7210/jrsj.25.130

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

Human motion has individual differences in motion trajectories and velocity of parts of the body by age, sex, and physical frame. This sometimes declines action recognition performance when the target subject is changed. In this paper, we propose a method for constructing the action recognition model that fits each person's motion. In the proposed framework, the existing recognition model for some person is modified to fit new person's motion. Only when the label predicted by the existing model is wrong, the model is modified with corrected labels. So, little training data are needed and the recognition model can be constructed more effectively. Empirical result using labeled motion data shows the effectiveness of the proposed framework.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper proposes a method for multiple persons tracking using sensing floor and multiple cameras. Our method estimates 3D position and ID of targets. To integrate sensing floor and multiple cameras, our method uses particle filter framework. The particle filter framework is known to be effective for tracking people. In this framework, our method adopts a load on sensing floor, color histogram and background subtraction of cameras images for hypothesis evaluation. Our experimental results demonstrate the effectiveness and robustness of the method against complicated movements of multiple persons.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

We are currently attempting to construct a mixed reality for home usage. The main components of the system include the active projector placed on the ceiling, a room where the projector is to be located, and a PC for controlling the output. By transferring the position at which a movie is projected so that the background of the movie scene stays at the same position, the motion of the object moving in the movie would be reproduced in the real world. Our method involves the following procedures: Input of a movie file, analysis of the background motion, transfer of the position of projection, and correction of the distorted image. This paper focuses on the latter two parts: Transfer of the position of projection and the correction of image distortion.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

An online approach is proposed to recover human body pose from 3D voxel data. The use of voxel data leads to viewpoint-free estimation, which benefits in that retraining is redundant in different multi-camera arrangements. Other advantages of our approach are speed and robustness. These are provided by an example-based method, applied by extracting posture labels from a large motion capture database. During the online process, only a similarity evaluation is needed between posture labels and online voxel data. The metric is formulated by introducing a histogram-based feature vector for representing the context of 3D volume. Estimation stability is improved by a precomputed graphical model of motion, which adds a smoothing effect to the motion sequence. We demonstrate speed and robustness of our approach with experiments on both synthetic and real image data.

CiNii Books

researchmap

Language：English  

DOI： 10.1109/ROBOT.2007.364058

Web of Science

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

In this paper, we propose two components, behavior description and anomaly detection algorithm in daily life. To begin with, action labels are assigned for each data-segment, using HMM (Hidden Markov Model) and k-means method. In traditional method, HMMs are composed for all data-segments. This paper improved our previous method, and succeeded to reduce calculation time. In anomaly detection step, typical action data are acquired, using probabilistic density of occurrence and successive time. Each probabilistic density is composed based on accumulating labeled-data, using SDLE (Sequential Discounting Laplace Estimation) and SDEM (Sequential Discounting Expectation and Maximization) algorithms. When new data come, if typical action data is changed largely, the data are detected anomaly.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

We developed an automatic behavior labeling system for tremendous data such as life-logs gathered through sensing human life. In this paper, the following advantages of the system were emphasized: 1) labeling behavior exploiting human knowledge and experience 2) relaxing constraints and selecting features freely as to constructing a labeling model, 3) helping find effective features for each behavior label, 4) dealing with incomplete data efficiently. The sequential labeling model with above features could be realized using Conditional Random Field (CRF) framework. We describe features of CRF, procedures of selecting effective features and how to deal with incomplete data in CRF framework. Lastly the model with CRF framework was applied to the real sensing environment and we state the performance based on the model and the feasibility of CRF framework in the human behavior recognition fields.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本ロボット学会  

In this paper, we propose an online recognition method for daily actions, such as walking and standing. The proposed method has following characteristics: (1) simultaneous recognition that is able to output multiple action names when human act more than one action, such a situation as<I>human is waving hand on standing</I>, (2) modeling action classifiers with kernel methods, (3) effective optimization for the parameters of the recognition system with margin-based query learning. The characteristic (2) unifies the process for modeling and learning the classifiers, and makes us easy to incorporate prior knowledge about action. The characteristic (3) reduces the burden of process for annotating action, which is an inevitable task for supervised learning. The experimental results using real motion capture data show that the proposed margin-based query learning is very effective to achieve high performance of the recognition system with very small sized query and annotation process.

DOI： 10.7210/jrsj.24.861

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

The system that visualize given words was built by using plenty of images existing on the Internet. This system has two devised points. Firstly, it uses not only the input query but also synonymous words of the input query when searching images from database. Thanks to that, wider range of images are able to be extracted as image candidates. Secondly, it re-rank images by using word class and appearance frequency in the page that image candidate exist of words in the acceptation text of the input query. In addition to them, image features such as shape or color are extracted from the image candidates and the important features for the input query are estimated. Weight calculated based on the important features are also used to rank the search result. Experimental results proved that images searched by the system are accurate enough in terms of subjectivity.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper describes action prediction system based on sensor data in human behavior measurement environment. The prediction is based on time-series association rule method. The method detects characteristic sequences from sensor event sequence and taught behavior to be predicted. Based on the detected sequences, our system predicts occurrence of the resident's behavior with prediction reliability before arbitrary period. The sensor events for the method are provided from sensor-embedded environments or a behavior simulator. The simulator generates virtual sensor data with input behavior sequence and sensor configuration. We evaluated the action prediction system for real sensor data in our Sensing Room and virtual sensor data. The experiments show feasibility of the action prediction on such behaviors as drinking a bottle of tea.

CiNii Books

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

Language：English  

DOI： 10.1109/IROS.2006.282059

Web of Science

Scopus

researchmap

J-GLOBAL

researchmap

J-GLOBAL

researchmap

DOI： 10.1007/11552246_2

Scopus

researchmap

Language：English  

DOI： 10.1109/IROS.2006.282400

Web of Science

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

This paper presents a human like segmentation method for daily actions. The proposed method utilizes human's teacher data of segmentation. By clustering the teacher data, wide variety of segmentation criteria of human are acquired. Each cluster is characterized by body parts it pays attention to, on the basis of tree representation of hierarchical structure of human body. We assume that latent dynamics changes at the segment-points of action, and utilize switching model of two linear dynamic systems to represent segmentation boundary.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

In this paper, we propose a rapid and robust online action recognition method. The main features of the proposed method are: 1) to select important motion features for action recognition, 2) to require very small calculation cost for recognition compared to conventional methods, 3) to exploit "Combinational Motion Features" which we propose as a new conception. We evaluated the proposed method to gate action recognition, such as walking and running, by utilizing motion capture data. In the result, the proposed method reduced parameters given by human to action recognizer and lessend human's task. In addition, the proposed method needed very small calculation cost for recognition while was capable of recognizing robustly as much as conventional action recognition method based on support vector machine. Moreover, the introduction of combinational motion features enhanced recognition performance.

CiNii Books

researchmap

DOI： 10.1109/ROBOT.2005.1570582

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人 日本ロボット学会  

This paper presents a recognition method of human daily-life action. The method utilizes hierarchical structure of actions and describes it as tree. We modelize actions by Continuous Hidden Markov Models which output time-series feature vectors extracted based on knowledge of human. In this method, recognition starts from the root, competes the likelihoods of child-nodes, chooses the maximum one as recognition result of the level, and goes to deeper level. The advantages of hierarchical recognition are: (1) recognition of various levels of abstraction, (2) simplification of low-level models, (3) response to novel data by decreasing degree of details. Experimental result shows that the method is able to recognize some basic human actions.

DOI： 10.7210/jrsj.23.957

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

CiNii Books

researchmap

J-GLOBAL

researchmap

Language：English   Publisher：一般社団法人 人工知能学会  

DOI： 10.1527/tjsai.20.197

Scopus

researchmap

DOI： 10.1109/IROS.2005.1545363

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

CiNii Books

researchmap

J-GLOBAL

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

This paper proposes a novel kernel computation algorithm between time-series human motion data that can be used for online action recognition. The proposed kernel is based on probabilistic models called switching linear dynamics (SLDs). SLDs are one of the powerful tools for tracking, analyzing and classifying human complex time-series motion. The proposed kernel incorporates information of the latent variables in SLDs by using technique of kernel design for probabilistic models called marginalized kernels. Compared with the other conventional kernels using SLDs, the main advantage of the proposed method is that the proposed method requires much less computational cost than the others. The empirical evaluation using real motion data shows that a classifier with our proposed kernel gets much better performance than the classifiers with some conventional kernel techniques.

CiNii Books

researchmap

Scopus

researchmap

DOI： 10.1109/ICSMC.2004.1399855

Scopus

researchmap

J-GLOBAL

researchmap

DOI： 10.1109/AFGR.2004.1301629

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人電子情報通信学会  

This paper proposes an algorithm of knowledge discovery for remarkable motion features in daily life action recognition based 6n SVM. The main characteristics of the proposed method are 1) basic scheme of the algorithm is based on Support Vector Learning and its generalization error, 2) detection of remarkable motion features is done in response to kernel parameters optimization via minimization of generalization error. Experimental result shows that the proposed algorithm makes the accurate rate of the recognition system to be high and enables us to detect remarkable motion features intuitively.

CiNii Books

researchmap

Language：Japanese   Publisher：一般社団法人 日本機械学会  

本研究では, 人の動作認識において「正しい認識」は人による認識であるという仮定の元に, 人に近い動作認識システムを実現した。ある既存パターンとの類似度を側るのではなく, 人の経験的な抽出特徴量をそのまま利用し, さらに, 複数動作の同時認識, 一定時間の動作の要約という人の動作認識の特徴を取り入れたシステムである。既存のHMMやCDPを利用した動作認識手法との比較を含めた実験によって, 本システムの認識性能を示す。

CiNii Books

researchmap

DOI： 10.1117/12.444165

Scopus

researchmap