Rule-Based High-Level Situation Recognition from Incomplete Tracking Data

Fuzzy metric temporal logic (FMTL) and situation graph trees (SGTs) have been shown to be promising tools in high-level situation recognition. They generate semantic descriptions from numeric perceptual data. FMTL and SGTs allow for sophisticated and universally applicable rule-based expert systems. Dealing with incomplete data is still a challenging task for rule-based systems. The FMTL/SGT system is extended by interpolation and hallucination to become capable of incomplete data. Therefore, one analysis to the robustness of the FMTL/SGT system in situation recognition is removing parts of the ground truth input tracks. The recognition results are compared to ground truth for situations such as “load object into car”. The results show that the presented approach is robust against incomplete data. The contribution of this work is, first, an extension to the FMTL/SGT system to handle incomplete data via interpolation and hallucination, second, a knowledge base for recognizing vehicle-centered situations.