Technical University of Munich (TUM): Driver Assistance
“AutoVision” system-on-a-chip architecture achieves reconfigurability and high video throughput.
Video-based driver assistance systems demand real-time processing of complex algorithms in diverse situations. ASICs or off-the-shelf ASSPs can offer the required real-time processing, but they lack the necessary flexibility. In addition, the design times for ASICs and their development costs have risen over the past years. Because video algorithms for driver assistance are not standardized, design changes are quite frequent, which rarely makes an ASIC a suitable choice.
The Institute for Integrated Systems at the Technical University of Munich, Germany set out to develop a reconfigurable FPGA-based system-on-chip architecture for vision-based driver assistance that is capable of real-time processing and can rapidly adapt to any upcoming situation. A key to this project is the use of Xilinx FPGAs, which offer the unique capability of dynamic partial reconfiguration (DPR), and embedded design tools such as Xilinx ISE, EDK, and PlanAhead™.
The resulting AutoVision architecture has been successfully demonstrated, showing that the inter- or intra-video-frame reconfiguration (IntraVFR) can significantly lower resource utilization of the system, and achieve a throughput of bitstream data at a rate of 1.2GBytes/second. This real-time processing technique, therefore, offers great potential in maximizing the flexibility and reducing the overall costs in automotive and especially video-based driver assistance systems.
Additional Information
- Read the full story, "Dynamic Partial Reconfiguration of Xilinx FPGAs Lets Systems Adapt on the Fly" in Xcell Journal Issue 70