Algorithm Fusion for Windscreen Obstruction Detection in Autonomous Driving

• Published in: 2017 19th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC) pp. 385 - 388
• Main Authors: Radu, Petru, Regep, Cezar Valeriu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2017
• Subjects: algorithm fusion; Automotive components; Automotive engineering; autonomous vehicle; Autonomous vehicles; camera blockage; Cameras; Detection algorithms; Detectors; Fuzzy logic; self driving car
• DOI: 10.1109/SYNASC.2017.00069

In recent years, Advanced Driver Assistance Systems (ADAS) for automotive industry has been increasingly using camera-based algorithms to detect various events that occur whilst driving. Usually, several computer vision algorithms run concurrently by processing frames grabbed from the windscreen camera. For a normal operation of camera-based ADAS algorithms, is mandatory that the field of view of the camera is not obstructed by noise factors. In this paper, an algorithmic fusion approach is explored to observe how two different methods for obstruction detection of camera’s field of view improve the detection rate when fused. The first obstruction detection algorithm relies on classical image processing operations, while the second obstruction detection algorithm employs a machine learning approach during its functioning. The present work investigates how the stability of the windscreen blockage detection is enhanced by employing trainable and non-trainable fusion techniques. The results are validated on numerous recordings from a range of vehicles and driving conditions.