Learning Computer Vision Through the Development of a Camera-Trackable Game Controller

Introduction

The laboratory is a tool of paramount importance in science education for several reasons. Experimentation and hands-on activities allow students to connect theoretical facts with their effects. Moreover, challenges and intellectual rewards coming from such activities help in providing motivation for further studying. However, the design of good laboratory experiences is not an easy task (Hofstein, A., & Lunetta, V., 2004). The quandary between the conceptual complexity of the activities to be offered and the limited scope of an experimental session is even more noticeable when dealing with advanced subjects such as Computer Vision (Bebis, G., Egbert, D., & Shah, M., 2003; Maxwell, B., 2001) or even Image Processing (Roman, D., Fisher, M., & Cubillo, J., 1998; Greenberg, R., 1998). In fact, these topics tend to be taught in graduated courses (Pridmore, T., & Hales, W., 1995) still, given the increasing importance of vision algorithms in the industry, they begin to be introduced also at undergraduate level (Sarkar, S., & Goldgof, D., 1998; Egbert, D., Bebis, G., McIntosh, M., LaTouttette, N., & Mitra, A., 2003; Hoover, A., 2003) or even during the high school (Greenberg, R., Raphael, J., Keller, J., & Tobias, S., 1998). In this paper we describe a laboratory setup that has been designed to be useful in teaching some basic Computer Vision concepts to high school students that are keen to join the undergraduate Computer Science course at the University of Venice. These prospective students exhibited a wide range of different knowledge backgrounds, since some of them have attended humanities studies, whilst others were more at ease with mathematics and physics.

Figure 1.

Shots of the vision-based controller and of the main panel of the educational application

For this reason it has been necessary to assume no previous programming skills and a very limited set of mathematical tools. To this end, we designed a laboratory session that includes about an hour of theoretical foundations and four hours of lightweight programming activity. The theoretical session introduces two concepts that will be central in the following experience: the YUV colorspace and the Hough Transform. They are both introduced with a minimal level of technicality and using very simple math notations. The programming session is performed within a custom environment that allows to write the body of the required functions directly without the need to know details such as how the images are acquired or how the methods are called. The functions have to be written with a subset of the Java syntax, that has proved to be simple to understand and suitable for this kind of laboratory lectures (Moscariello, S., Kasturi, R., & Camps, O., 1997). Before starting the programming activity itself, each student was supplied with a one-sheet handbook of the needed Java syntax and with some general instructions about the use of the integrated editor. The overall lab experience, which will be explained in detail in the following sections, imply the use of a simple game controller cut-out (see Figure 1) made out of a PVC sheet that must be recognized from an image with the help of the color of the band printed on it. Once the recognition happens, the orientation of the band must be determined and used to control a little game.