At the Electronica 2008 Conference, Xilinx Inc. released their Xilinx Automotive (XA) Optical Flow solution for vision-based driver assistance systems.
This Field Programmable Gate Array (FPGA) implementation integrates high- performance image processing capabilities from Digital Design Corporation (DDC) into a cost-effective, programmable platform that can be customized to support a variety of advanced applications, including pedestrian detection, collision warning, sign recognition, blind spot detection, back-up and parking aid.
The XA Optical Flow system delivers the scalable processing power to implement pixel-level image processing functions that extract relative object motion information from a sequence of video frames.
XA Optical Flow is built on a Xilinx Spartan(R)-3 device and DDC Optical Flow IP core with its innovative motion estimation algorithms. The Optical Flow IP core extracts relevant information about the roadway environment from the video images provided by vehicle-based camera systems. The extracted motion images are segmented into separate objects (pixel blocks) and processed using a block matching technique. The output is a two- dimensional vector map that can be utilized to assess collision threats.
The block matching function requires a very high number of operations per second to perform sum of absolute difference (SAD) calculations across image frames. The Xilinx FPGA executes these operations in parallel, so in a single clock cycle multiple operations can be carried out. This is a higher performance, more efficient optical flow solution than is possible with traditional digital signal processing (DSP) CPUs, which generally operate serially and require very high clock rates.