The system described here is a real-time edge detector for use in image processing. The edge-detection algorithm is a 2 ¿¿ 2 Roberts product, thresholded with a function of the local average brightness. The hardware is designed to work at 10 MHz pixel rate, and will accommodate images of 512 ¿¿ 512 pixels at 25 frames/s with noninterlaced video, or of up to 512 ¿¿ 290 pixels with interlaced scanning. The hardware is pipelined and paralleled to achieve the required speed. The output from the system is a one-bit-per-pixel edge picture. This is transferred directly into the memory of the host computer for further processing.