Hironobu Fukai

Apparent age estimation system based on age perception

The age is one of important information in our living. If the age estimation that uses face image by computer becomes possible, it is thought that the age estimation assumes an important role in various scenes. In this paper, we propose an age estimation every age by using the supervised SOM. Furthermore, the important features for the age estimation are selected by the GA.

An age estimation system on the AIBO

In this paper, we propose an age estimation system on the AIBO. AIBO is an entertainment robot produced by SONY co., Ltd.. AIBO has many sensors to get information around itself and moves according to its instinct. This autonomous action is considerably natural. However, it is inadequate to communicate with people. If AIBO can estimate the human age from a face image, it becomes more excellent entertainment robot. Then, we propose the age estimation method on the AIBO by using face image. In this paper, the apparent age feature is extracted by the fast Fourier transform (FFT), and it is selected by the GA. Moreover, the age is estimated by the 1-dimensional SOM. In order to show the effectiveness of the proposed method, we show the simulation examples.

Appearance-age feature extraction from facial image based on age perception

An age is relatively important information in the various features which are shown human from face. It is possible to apply it to an amusement and an aesthetic surgery by analyzing the appearance-age feature which is used for age perception in a facial region. In this paper, we analyze facial feature of appearance-age based on age perception. In order to construct the appearance-age database, age estimation experiment is performed to the HOIP facial database by 15 subjects. The appearance-age feature used potentially when human performs age estimation is decided by using the genetic algorithms and the neural network. By using the proposed method, facial features that influenced on the appearance-age in each gender was confirmed. Moreover, it is suggested that that appearance-age feature which is used for age estimation is different in each generation when human performs age estimation.

Facial expression estimation using simplified head model based on particle filtering

We propose a method for real time estimation of head poses and facial expressions without using a complex 3D head model. Most conventional methods for head pose and facial expression estimation required preliminary estimation using a complex 3D head model. In the proposed method, however, a large number of feature points and a whole facial image are not required in order to estimate facial expressions. The proposed method estimates the head pose and facial expression based on a simplified 3D deformable facial shape model approximated by a cylindrical model. In order to show the effectiveness of the proposed method, we performed facial expression estimation experiments.

Age and gender estimation by using facial image

In this paper, we propose age and gender estimation system by vairous features. Age and gender has a lot of characteristics. These characteristics are one of the difficult cognitive process in human interaction. If we can extract the important feature of this cognitive process, it is considered that the age and gender estimation by the machine becomes possible. Therefore, we propose a method of age and gender feature extraction and estimation using the face image. In this paper, the age of the face means apparent-age that is based on the human perception of age. Moreover, persons aging and gender difference appear in the faces. For example, the pigmented spot, the wrinkle, sagging skin, shape, color of skin, and so on. Thus, we extract these several features for age and gender estimation. Furthermore, we estimate a continuous age and gender using a neural network (NN).

Interpolative intra prediction by adapting processing order in block-based image coding

This paper proposes a novel intra prediction method which can be used with block-based image coding algorithms typically using DCT. In this method, a non-directional and several directional prediction modes are adaptively selected on a block-by-block basis. The non-directional prediction mode produces smooth image values from the reference pels based on mean value coordinates. For the directional modes, the angular prediction used in HEVC is extended to enable interpolative prediction as far as possible. In both cases, the proposed method has no restrictions on positions of the reference pels. This feature allows adaptation of processing order for the blocks. Simulation results indicate that the proposed method outperforms the angular prediction in terms of PSNR of predicted images when almost the same amount of side-information is assigned.

Fast and robust registration of multiple 3D point clouds

In this paper, we propose a method for the fast and robust registration of multiple 3D point clouds. This method contributes the simultaneous localization and mapping (SLAM) robot. The problem of point cloud registration can be roughly classified into two. The problem of estimation of the relation from unknown geometry position is coarse registration. Moreover, in case of roughly position is known, the problem of estimation of the relation from known initial position is fine registration. In this study, we solve the coarse registration problem by exhaustive search, and prepare the initial positions for fine registration. To employ exhaustive search, we realize robust matching by using no feature point. The fine registration problem is solved by the ICP algorithm. In this case, reduction of initial positions for the ICP algorithm by non-extremum suppression that uses distance between two range data solves the problem of calculation cost. The distance evaluation function that robust for measurement error tackles the outlier problem. The problem of calculation cost is solved by registering the distance from the last point cloud data beforehand by the distance field. The effectiveness of the proposed method is shown by a actual data.

An apparent age estimation system using the evolutionary algorighm

The age is one of important information in our living. If the age estimation that uses face image by computer becomes possible, it is thought that the age estimation assumes an important role in various scenes. In this paper, we propose an age estimation by using the supervised SOM. Further, the important features for the age estimation are selected by the GA.

The calibration between range sensor and mobile robot, and construction of a obstacle avoidance robot

Recently, robot that assists a human is desired, because the aging society will come. For example, if there is a robot which avoid a obstacle automatically, it is possible to assist humans. Therefore, we construct a obstacle avoidance robot. In order to achieve a autonomous robot, obstacle avoidance, path planing, and mobile robot control are required. In this study, TOF type range sensor that is a possible to measure the distance is used. Moreover, an obstacle is detected using the distance information acquired by TOF camera. Optimal path is decided by the potential method. When TOF camera is put on a mobile robot, it is a problem that a robot coordinate system differs from the coordinate system of TOF camera. Then, the calibration of a robot and TOF camera is performed by odometry information and estimating the pose between TOF camera and actual environment. In order to verify the practicality of this proposal system, we experiment at room environment with flat floor.

Robust and fast 3D object recognition using exhaustive search and ICP algorithm

This paper proposes a robust and fast alignment method of 3D point cloud. The problem of point cloud alignment can be roughly classified into two. The problem of estimation of the relation from unknown geometry position is coarse registration. Moreover, in case of roughly position is known, the problem of estimation of the relation from known initial position is fine registration. In this study, we solve the coarse registration problem by exhaustive search, and prepare the initial positions for fine registration. To use exhaustive search, we realize a robust. The fine registration problem is solved by the ICP algorithm. In this case, reduction of initial positions for the ICP algorithm by non-extremum suppression that uses distance between range data and model solves the problem of calculation cost. The distance evaluation function that robust for measurement error tackles the outlier problem. The problem of calculation cost is solved by registering the distance from the model beforehand by the distance field. The effectiveness of the proposed method is shown by a actual data.