[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
When a user manipulates objects in a virtual space using 3D position input devices and stereoscopic devices, there occur problems that the user cannot manipulate the objects as he/she does in the real world. One of the reasons of this problem is the gap between the virtual space and the 3D space that the user perceives with the stereopsis. In this paper, we propose a method to correct the gap between these two spaces, so that the user can succeed in manipulating objects as he/she intends.
In the real world, we perceive the depth of a scene from various features including binocular disparity. On the other hand, in a virtual space, its depth is presented only by stereo images with binocular disparity. Thus, the 3D space that the user perceives from the stereo images does not coincide with the virtual space even if the distance between the centers of projection to generate the stereo images of the virtual space is set to be the distance between the pupils. As the result, when a user tries to point at the surface of an object with a ``pointer'' in the virtual space, the pointer leaves from the object or sinks into the object.
In this paper, we propose to correct the gap between the two spaces with the information when a user manipulates virtual objects. First, when a user tries to manipulate a virtual object, it is assumed that the user judges whether the pointer and the virtual object are in contact based on a certain threshold, (which we call the ``threshold of error''), to evaluate the distance between the pointer and the object. The threshold of error is the maximum distance between two objects considered that they are in contact with each other by the user. Different values could be taken for different users, and the same value is employed by the same user.
We consider the manipulation of objects by the user in virtual space is failed when the gap between the two spaces exceeds the threshold of error for a user. Therefore, if the gap can be narrowed within the threshold of error, the manipulations of objects in the virtual space by the user become successful.
To detect the case in which the gap between the two spaces exceeds the threshold of error, an assumption about the situation, which should be satisfied when a user manipulates objects, is employed. That is, the pointer and the object under manipulation must be recognized by the user that they are in contact with each other in the distance below the threshold of error. In this paper, this is called spatial constraints in object manipulation. In order to apply these constraints, the point called a manipulating point, at which a user recognize him/her to touch with a pointer in virtual space, is searched when the user begin to manipulate. If the distance between the pointer and the manipulating point is longer than the threshold of error, the spatial constraints are not filled, and the gap between the two spaces are corrected that the constraints are satisfied at the manipulating point.
If the virtual space itself is corrected to narrow the gap, different stereo images would be presented to the user, and the space percepted by the user would also be changed as the result. To avoid it, another space that is initially the same as the virtual space, which is called a model space, was synthesized and modified to represent the user's perception.
In the model space, the user's perception from the stereo images is represented by a 3D space recovered by stereopsis with two cameras. The 3D position obtained by stereopsis from the stereo images is controlled by the two of parameters for stereopsis, the distance between two cameras and the focal length of the cameras. Correction of the model space is realized by changing these parameters for at each position of the space.
The parameters for a position are changed based on the user's perception at the position while the parameters for neighboring positions do not greatly different from each other. The threshold of error is changed as well as the parameters. It is expected that their variables finally converge and the success rate of the user's manipulation increases through the modification.
It was confirmed through some experiments that the model space was corrected forward the personal perception in several trials, and that the success rate increased.