Game Engine, The New Interactive System Based On Computer Vision

It was recognized that the new realtime human-computer interaction based on web cameras is a much better interactive means than traditional one, in which keyboard and mouse are often used. Espacially, when two cameras are used, it can be fully simulated that how human eyes sense the objective world to obtain the 3D description of the scenes through accurate positioning and analysis. It also allows the operator who virtually simulates the real scenes to realize human-computer interaction in a more natural way, such as space location, gesture control, etc.It was mainly done by useing computer image analysis technology in this new interactive way based on two web cameras to implement the analysis process to every frame captured by the cameras, and abstracts from images the foreground objects and the background. Then, the foreground objects are used to be located or traced in the calculation of the process of stereo computer vision.The research described in this thesis is based on the platform of Game Engine which helps to realize the whole system as the functional extension to the engine. It was mainly focused on the modular feature and reusability in the design of the system. As a result, it is actually a complete system with its own structure, which, independent of the game engine, offers strong transplantablity and maintainability and can be used as a separate component in the applications and systems requiring to use computer vision.As a system of computer vision, it was provided a detailed description on how the computer vision was realized and the important arithmetic was used. And it also deals with the steps applied in the processing of these arthmetics. For those which has been proven, it was directly implemented after been studied; and for those ones that could not meet the requirements of the performance or functions of our system, some combinations on a detail scale, or some new methods was made to meet the system requirements. And then it was optimised as a whole. Since some steps, which is essential without appropriate arithmetics, new applicable arithmetics was developed to implemente them.Our main works involved are described as follows: A detailed design of the arithmetics process for the whole system and system structure implementation was made and it was proved that how the different arithmetics were applied realized step by step. The realtime performance required by the system is guaranteed in the seamless combination with the game engine.When dealing with the catching of interest points as one of the key processes, A new way to combine the edge message and the corner points of the two characteristics of the image was suggested. The numbers of the usable points are increased and the amount of information requiring processing are reduced as well as making up the loss in filtering section, which helps to improve the efficiency and robustness of the arthmetics.As for the other key process of matching the interest points, an innovative arithmetic was designed by using the epipolar geometry restriction and homography restriction to limit the matching scope and then match by means of diffusive search which substantially improves the system performance. And the use of feedback correction parameter also helps to meet the system requirement on reliablity.