Konferenzen und Rahmenprogramm
Parallel Architectures for Object-Based Sensor Fusion on Automotive Embedded Systems
Autonomous or highly automated driving is an emerging development in science and industry in the last decades. Sensor fusion, in which information from different sensors is combined to achieve higher measuring accuracy or to create new information, acts as a key for this technology. Although much research is done in developing and improving algorithms for advanced driver assistance systems (ADAS), the development of automotive embedded hardware that is capable of running those applications is still in its beginnings. Automotive embedded systems have to meet some special requirements compared to other embedded applications. Important factors are costs per piece, energy consumption and safety requirements (cf. ISO 26262). Since power consumption increases proportional to CPU frequency it seems reasonable to use multi-core architectures to meet the computational requirements, while keeping the power consumption low at the same time. This paper studies the benefits of using parallel architectures for object-based sensor fusion on automotive embedded systems. For the evaluation a simulation using Kalman filtering for the state estimation and an auction algorithm for the data association was implemented. All simulations were performed on a NVIDIA DRIVE PX 2 board, containing four ARM A57 cores and a Pascal GPGPU. The results show that multi-core processors can be used to efficiently speed up object-based sensor fusion in embedded systems, whereas a GPGPU based implementation largely suffers from high latency caused by memory accesses.
--- Datum: 28.02.2018 Uhrzeit: 11:30 Uhr - 12:00 Uhr Ort: Conference Counter NCC Ost