Real-time motion estimation using spatiotemporal filtering in FPGA

Reliable visual motion estimation is typically regarded as a difficult problem. Noise sensitivity and computational requirements often prohibit effective real-time application on mobile platforms. Despite these difficulties, biological systems reliably estimate visual motion in real-time and heavily rely on it. Here we present an FPGA implementation of a biologically inspired spatiotemporal energy model for motion estimation. The model realises 720 motion sensitive units per pixel for video of resolution 128×128 pixels at 30FPS, thus providing a computational tool for further investigation of spatiotemporal energy models.