Minimizing fault handling time interval of camera sensor failure for fail-operational autonomous driving systems

To realize fully autonomous driving, autonomous driving systems must be able to operate even in the event of failure of certain components. As an essential component for autonomous driving, this study is focused on the camera sensor failure in an object detection system, in which redundant cameras are utilized to develop a fail-operational perception system. Although redundant sensors are widely applied, only a few studies have focused on the timing analysis and optimization of the fault handling time interval (FHTI). As defined in the second edition of the ISO 26262 standard, the FHTI is the time between the failure of a component and the instant at which the system recovers to a safe state through safety mechanisms. Our safety mechanism minimizes the system FHTI by reducing the two delay components. First, we reduce the fault reaction time by adopting a backup camera in the hot standby mode. Second, we implement an optimal failure check mechanism that tests the status of the primary camera with an optimal period that can minimize the worst-case fault detection time, considering both the camera frame rate and object detection frame rate. The proposed fail-operational system is implemented through the Darknet object detection framework. The experimental results demonstrate that the FHTI is reduced from 391.17 ms to 39.23 ms, corresponding to a reduction of by approximately 90% compared to that for a baseline fail-operational architecture. © ICROS 2021.