Fault-tolerant scheme for vision-based lane keeping system

Abstract

In this paper, we propose a fault tolerant method for lane recognition in a vision-based lane keeping system (LKS). In the presence of vision sensor failure due to malfunction and/or environmental conditions, LKS immediately releases its control. For vehicle safety, it is required to provide available virtual lane information to LKS until the driver becomes aware of the situation and takes over control of the steering system. Thus, we cope with vision system failure by proposing a new lane estimation scheme that uses a vehicle dynamic lateral motion model and clothoidal road constraints. In addition, we introduce on-road vehicle localization with a global positioning system (GPS) even under long term failure of a vision sensor. The fault tolerant system using GPS can successfully restore lane information for a few seconds so that virtual lane information becomes available to the lateral control system and the driver can take over steering control without panicking. The developed algorithm was validated via computational simulation results with MATLAB/Simulink and we included experimental results with a test vehicle equipped with a Micro AutoBox from dSPACE. ? ICROS 2017.