Applying Neural Networks with a High-Resolution Automotive Radar for Lane Detection

Konferenz: AmE 2019 – Automotive meets Electronics - 10. GMM-Fachtagung
12.03.2019 - 13.03.2019 in Dortmund, Deutschland

Tagungsband: GMM-Fb. 93: AmE 2019

Seiten: 6Sprache: EnglischTyp: PDF

Persönliche VDE-Mitglieder erhalten auf diesen Artikel 10% Rabatt

Feng, Zhaofei; Zhang, Shuo; Kunert, Martin (Advanced Engineering Sensor Systems, Robert Bosch GmbH, 71226 Leonberg, Germany)
Wiesbeck, Werner (Institut für Hochfrequenztechnik und Elektronik, Karlsruher Institut für Technologie, 76131 Karlsruhe, Germany)

Neural Networks are widely used for object-level detection and classification or pixel-level segmentation with optical sensors like video cameras or laser scanners. Conversely, limited by its reflection point number, classification with neural network based on radar data gains much less attention. For state-of-the-art radar sensors on the market, functions like Adaptive Cruise Control, Forward Collision Warning, and Automatic Emergency Breaking are implemented. Since radar sensor poses advantages over camera or laser scanner under certain adverse environmental conditions, the system reliability for autonomous vehicle can be significantly improved by fusing the radar detections with the data coming from other optical devices. Currently, with the improvement in resolution for the next generation automotive radar systems, neural network also becomes a very promising tool in classification or segmentation. Since knowing the driving lane is a crucial task for autonomous driving, this paper shows such a potential for radar sensor to process driving lane segmentation together with a popular neural network that is mostly applied in computer vision domain.