For many years, the Digital Safety and Security Department of the Austrian Institute of Technology has been developing stereo vision algorithms for various application purposes. Recently, these algorithms have been adapted for use in a collision avoidance system for tramways. The safety validation of such a system is a specific challenge as - like in the automotive domain - the rate of false positives cannot be lowered to zero. While automotive suppliers typically tackle with this problem by reducing the sensitivity of the system and validating it in hundreds of thousands of test kilometres, this paper presents an approach how it is possible to demonstrate safety with a carefully chosen functionality and less field test kilometres.
Gruber, T., & Zinner, C. (2015). Approach for demonstrating safety for a collision avoidance system. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9338, pp. 167–177). Springer Verlag. https://doi.org/10.1007/978-3-319-24249-1_15