HOME > Ambient Intelligence:Knowledge Structuring Group

Ambient Intelligence:Knowledge Structuring Group

The Knowledge Structuring Group is concerned with ambient intelligence systems, which is a topic at the very core of basic infrastructural technology for ambient intelligence. It will consider such issues as environmental recognition and understanding, as well as user conditions recognition and understanding, and intelligent interaction.

Researchers of NAIST

- Naokazu Yokoya (Professor: Vision and Media Computing Lab.)
- Yoshitsugu Manabe (Professor: Vision and Media Computing Lab.)


- Takeo Kanade (Professor: Carnegie Mellon University)

Visiting Researchers

- Kentaro Takemura (Assistant Professor: Robotics Lab.) 2010.12.14-2012.03.02
- Hideyuki Kume (Ph.D candidates: Vision and Media Computing Lab.) 2011.10.01-2012.10.01
- Jun Takamatsu (Associate Professor: Robotics Lab.) 2012.02.20-2013.03.30

Research Reports

Research report of Takemura Kentaro Takemura: [PDF]
Head-mounted camera systems such as First-Person Vision and Google Project Glass have been studied actively, and these systems require eye tracking function as a intuitive method. A typical head-mounted eye tracker consists of a scene and eye cameras, thus a calibration procedure is required prior to use. Nevertheless, if calibration is conducted completely in advance, the calibration record might be broken during a long period of use. Therefore, we studied the runtime calibration based on Saliency map for head-mounted eye tracker.

Research report of Kume Hideyuki Kume: [PDF]
In most autonomous driving applications, such as parking and commuting, a vehicle follows a previously taken route, or almost the same route. In this paper, we propose a method to localize a vehicle along a previously driven route by using images. In the method, first, we estimate rough position by using topometric localization which identifies the most similar image from the previously captured images by considering topological and metric information. Next, precise position and posture are estimated from 3D positions of feature points which have been reconstructed beforehand. In the experiment, we estimate vehicle poses by using images captured in indoor parking lot.

Research report of Takamatsu Jun Takamatsu: [PDF]
Estimating human intention is one of the key components to realize service robots providing preferable service. Estimating human poses from images helps us understand intention of non-verbal actions, such as gestures. In this research, we consider a method to achieve this. Especially we consider to use region-based image features, such as colors. First, we define the sketch model to minimally represent the color model of an individual. Next we describe the estimation method if we know the sketch model. Then, we describe a method to simultaneously estimate pose and sketch model from roughly estimated pose. Finally we describe lazy decision of human poses by using multiple local minima of the estimation as candidates.


International conference (Peer-Reviewed)
  1. Hideyuki Kume, Arne Suppe, Takeo Kanade: "Vehicle Localization along a Previously Driven Route Using Image Database", Proc. IAPR Conf. on Machine Vision Applications (MVA2013), pp. 177-180, May 2013
  2. K. Takemura, A. Ito, J.Takamatsu, and T. Ogasawara: gActive Bone-Conducted Sound Sensing for Wearable Interfaces,h Proc. of the 24th ACM Symposium on User Interface Software and Technology(UIST2011), pp.53-54, 2011.

Domestic (Japanese) conference
  1. ŒH Gs, Arne Suppe, ‹ΰo •—Y: "Šω‘–sŒo˜H‚Ι‚¨‚―‚ι‰ζ‘œƒf[ƒ^ƒx[ƒX‚π—p‚’‚½Ž©ŽΤˆΚ’uEŽp¨„’θ", ξ•ρˆ—Šw‰ο Œ€‹†•ρ, Vol. 2013-CVIM-186, No. 13, Mar 2013

  1. Jun Takamatsu: "Human Body Alignment", People Image Analysis Workshop 2012, Carnegie Mellon University, Dec. 13 2012.