Heterogeneous Framework Architecture of Specialized Accelerators for Vehicle Sensors

Abstract:
The development of Advanced Driver Assistance Systems (ADAS) requires the processing hardware to be aligned with the demands of sensor data processing. Automotive grade radar and LIDAR sensors are essential for autonomous driving and produce a significant amount of raw data which needs to be processed in real-time. This work proposes a software framework which decouples sensors, processing algorithms, and processing hardware to be independent to the greatest possible extent. A modular based approach is the key to quickly derive conclusive insights about the impact and suitability of individual components on the overall system. This framework enables interchanging and testing of multiple constellations with minimized work efforts for the user. The benefit and simplicity of this framework is exemplified by a radar processing use case.