In this paper, we built the haptic display function into a multi robot and 3D VR environment simulator "Gazebo" which enables to deal with force feedback information and established the environment which supports the development of the robot control programming. Gazebo can have robot hardware-independent features for the control program. However, the functions of the control are limited to fundamentals such as motor voltage control, sensor feedback as a camera and so on. The proposed environment makes it possible to facilitate development of robot manipulation interface and realizes the safe and intuitive operation for users.
This paper describes a method and experiment on recognition of online sign language operation. The sign language data is acquired from motion capture. The recognition process is based on the Procrustes matching, which is a generalization of linear regression model. The Procrustes matchingt is a permissive one for rotation changes. The proposed method is corroborated by a computer simulation under several condition , which correctly classified 81% of 110 patterns (11category) for 3D Procrustes matching and correctly classified 67% for 2D Procrustes matching.
Fractal models have been widely applied to image processings and related fields. However, fractal application to image analysis have not been successful adequately. Because, conventional fractal feature --- fractal dimension--- is scalar valued, and it is difficult to discriminate various image information using them. In this paper, fractal matrix model, which is proposed to overcome this drawback, is applied to texture analysis. The fractal matrix is a Hermite matrix, and abundant information compared with fractal dimension. The matrix elements are estimated by the least square method. The texture image analyzing experiments showed the effectiveness of the fractal matrix as an image model and image feature.
We propose a tracking method via 1D flow. This method calculates 1D flow as distribution. The characteristics of this method are avoidance of ill-posedness by 1D flow, and expansion of the traceable motion by using the multidirectional distribution. The experimental results show the possibility of tracking by distribution of 1D flow.
In image analysis, it is important to detect the lines which represent the edges of the objects in the image. The most popular technique to detect the lines is Hough Transform. In the transform, resulting peaks in the accumulator array which are gotten by a voting procedure in the parameter space represent strong evidence that a corresponding line exists in the image. In the voting procedure, a large number of votes which are far from the peaks are unnecessary and it makes the transform slow. In this article, we propose a speed-up method by introducing a procedure (using a inter-frame operation) in image space which removes lines except for limited directional lines. Owing to the procedure, we can reduce voting range and remove the unnecessary vote. From the experimental results, we confirm that the proposed method has good performance.
We have investigated a method that can represent 3D CG of room layouts in housing magazines, which are drawn in 2D. There are some researches that can construct 3D model data from 2D layout images; however, most of the images are drawn with CAD or graph paper and there is little research that can treat the layout images drawn by hand with unknown size. Therefore, we have tried to represent 3D CG of the room layouts, which size is unknown, by recognition technique. 3D representation of 2D room layout helps us to recognize the layout of the room and search new rooms effectively.
A virtual coupling is a conventional method to link a haptic device and a virtual object in VR environment. Because the virtual coupling method is designed mainly for stability, it may have low fidelity. Therefore, the application developer had to select the coefficient of coupling of each application by the trial-and-error method. In this paper, we describe a method by which the coefficient can be automatically selected by using machine learning.