Evaluating Perception Systems for Autonomous Vehicles Using Quality Temporal Logic

For reliable situation awareness in autonomous vehicle applications, we need to develop robust and reliable image processing and machine learning algorithms. Currently, there is no general framework for reasoning about the performance of perception systems. This paper introduces Timed Quality Temporal Logic (TQTL) as a formal language for monitoring and testing the performance of object detection and situation awareness algorithms for autonomous vehicle applications. We demonstrate that it is possible to describe interesting properties as TQTL formulas and detect cases where the properties are violated.