This document describes an FPGA-based system for real-time computation of disparity maps from stereo images and segmentation of scenes into foreground and background. The system generates four disparity maps in parallel using two similarity metrics (SAD and Census) and sweeping directions. It then merges the maps into two bitmaps indicating foreground versus background pixels. The key contributions are a custom hardware architecture for high-speed disparity computation and an optional post-processing stage. A prototype implemented on an FPGA was able to process 640x480 images at up to 40 fps with a maximum detectable disparity of 135 pixels.