Real-Time Imaging – IGIC
November 2, 2011, International Games Innovation Conference, Orange, CA–Wen-Chung Kao from National Taiwan Normal University presented information on real-time imaging and recognition techniques for game applications. The move to body control of on-screen activities is driving work on alternate controllers.
The issues for imaging are exposure, color, tone, and dynamic range. Exposure control is only realistic with raw data, and is made more difficult by the camera non-linearity. Exposure control is impacted by overall scene brightness and contrast range, so most algorithms quantize the brightness levels and analyze the values in each bin.
Problems for color fidelity are related to the reproduction and to the algorithms used to render the colors. Highlights and saturation cause errors and the averaging used in most algorithms drives all of the colors and intensities towards a middle grey. Large, uniform color blocks create distortions as does high dynamic range.
One way to get better imaging is to calibrate the sensor linearity and use a calibrated light source to establish a baseline setting. Techniques to reduce dynamic range without adding additional distortions would help. Ideally, the systems would detect images in multiple light bands and merge the information into a single image. Then the algorithms could work to equalize the histograms.
A local contrast map evaluation enables greater photographic tome reproduction by flattening the response curves and reducing dynamic range. All of these advances would improve facial detection and recognition and improve resolution. Ideally, the imaging system should recognize finger motions rather than large body movements. These developments will improve the human-machine interface and make the unattached game experience much better.
