Real-Time Image Processing

Every digital consumer camera today can read images from a sensor chip and (optionally) display them in some form on a screen. However, what we want to do is implement an embedded vision system, so reading and maybe displaying image data is only the necessary first step. We want to extract information from an image in order to steer a robot, for example following a colored object. Since both the robot and the object may be moving, we have to be fast. Ideally, we want to achieve a frame rate of 10 fps (frames per second) for the whole perception–action cycle. Of course, given the limited processing power of an embedded controller, this restricts us in the choice of both the image resolution and the complexity of the image processing operations.