Implementing active vision in embedded systems

Machine vision is a powerful tool for process automation and control. However, the memory requirements and interface overheads of video buffers found in typical video grabbers are not easy to interface to low cost embedded systems. This paper describes an active vision interface that enables a low end microcontroller to use a standard video camera as an input for process control. The active vision interface consists of a frame grabber with a small (4 kb) memory mapped buffer that can be zoomed out to sample the whole image at low resolution, or zoomed in and panned around the image at high resolution. This scheme takes advantage of the fact that most real time vision systems operate on low resolution images or small parts of a high resolution system in order to reduce the number of operations performed on each frame. The system integrates tightly with the embedded microcontroller by providing current video in the microcontroller's memory space. This minimises the overheads in interfacing and maximises the bandwidth to the vision processing algorithm. The interface has been designed to minimise latency, and to provide contention-free access to live video.