[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
A method of making a lecture video with multiple active cameras in a university classroom is proposed. This research focuses on how to select one image from several ones offered by the multiple cameras.
Lecture is one of the most important services in university. Students are not able to attend all lecture. Recording lecture on video tape could help them.
The students in the classroom turn their attention to multiple objects, but one can see only one object in the classroom at a moment. I propose a method of making a lecture video by selecting images from the multiple active cameras located at multiple viewpoints.
The lecture video includes ``lecture information''. The ``lecture information'' indicates all the objects and events related to lecture, which consists of three elements, ``teacher'', ``blackboard'' and ``student''. These elements are called ``visualized object''.
The process of making the lecture video is divided into three steps, ``camera-work'' step, ``camera-work distribution'' step and ``image selection'' step. In the ``camera-work'' step, %I describe a method that one or two cameras image ``visualized object''. In the ``camera-work distribution'' step, the problem is how to offer appropriate viewpoint for each camera, and in the ``image selection'' step, I describe how to select one of the images from the multiple cameras for students.
To start with the camera-work step, there are three functions named ``Teacher-imaging'', ``Blackboard-imaging'' and ``Student-imaging''. Each of the functions corresponds to a visualized object. Teacher-imaging function gets two cameras to offer teacher's image for triangulation. Blackboard-imaging function has a camera to offer an image of a part of the blackboard when a teacher uses it. The part is determined by the teacher's position. Student-imaging function has a camera to offer an image of students, for the purpose of taking a view of their behavior.
Next in the camera-work distribution step, each camera-work function is assigned to a certain camera whose viewpoint is appropriate to image its visualized object. It is ideal for each camera-work function to use the most appropriate camera. Since there is a limit of number of cameras, the preference order for camera-work functions is introduced to assign a camera according to the orders. In the first place, the Teacher imaging function is dynamically assigned to the two cameras that are the most appropriate to triangulate. In the second place, the Blackboard imaging function is assigned to one of the remaining cameras, only when the teacher stands near the blackboard. Furthermore, the Student imaging function is assigned to all the cameras which are not assinged any other camera-work functions.
Finally in the image selection step, a method of selecting images is proposed for the purpose of making the lecture video for students. A process which selects one of the images as substitute for students is called ``Director''.
The Director refers ``confidence-value'' to evaluate the images. ``Confidence-modify-value'' represents the requests of students. The confidence-value, which indicates an image desirability, is calculated in each camera-work. In Teacher imaging function, it is determined by the distance between the teacher and the camera. Similarly in Blackboard imaging function, it is determined by the distance between the teacher and the blackboard, and by the scattering of students in Student imaging function. The confidence-modify-value represents the change of student's interest to visualized object.
Based on the proposed method, I construct a system which makes the lecture video for experiments in the classroom where four active cameras are equipped. In the experiment,it has shown that the appropriate images were obtained from the cameras, each camera-work function is assigned to appropriate cameras, and appropriate images were selected during the lecture.