Individualized situation recognition using approximate case-based reasoning
Arezoo Sarkheyli-Hägele · DuEPublico (University of Duisburg-Essen) · 2018
Situation recognition is a significant part of humans perception as well as in the process of supervising human operators decision making in unplanned, imprecise, and uncertain environments. It is a process for identification of actual situation as the result of the occurring events within the environment. With outstanding performance, different situation recognition approaches for various applications have been developed. However, far too little attention has been paid to individualization of situation recognition. Situation recognition process could be individualized for supervision of human operators by learning and considering exclusive behaviors, preferences, and priorities of individual human operators. The event-discrete situations which are generated with a sequence of triggered actions could express individual behaviors of human operators. Accordingly, representation and identification of event-discrete situations are considered in this contribution. The purpose of this thesis is to propose a new framework for individualized situation recognition by applying novel knowledge representation and reasoning approaches. The most major challenges to be solved through the proposed framework for individualized situation recognition are modeling and representation of experienced knowledge as well as learning the new unknown situations. Those challenges could be stated as two questions as follows: How to model and represent the event-discrete situations to a knowledge base? How to reuse the knowledge for further situation recognition? In this work, Case-Based Reasoning (CBR) approach is applied to realize individualized situation recognition for supervision of human operators. The classical CBR is improved with a new learning process to recognize known occurring situations and generate new knowledge from unknown occurring situations. To deal with the noted challenges and realize the situation recognition, the classical CBR is also improved with application of a knowledge representation approach based on Situation-Operator Modeling (SOM) and fuzzy logic (FL). This work details the proposed CBR approach as a part of approximate reasoning. An integrated knowledge representation approach based on SOM and FL is introduced for representation of knowledge in the CBR. The SOM approach models the knowledge and describes the relations between discrete-event situations in a dynamic environment. The presented SOM approach supports the learning process by defining a sequence of situations and actions for each situation pattern. The FL approach structures the knowledge modeled by SOM for approximate knowledge inference. Additional processes need to be carried out in the proposed fuzzy SOM-based CBR to support online learning, data reduction, and knowledge indexing. The presented framework is applied for realization of an individualized lane-change situation recognition to supervise human drivers. The goal is to recognize the suitable driving situations for changing the lane for individual human drivers. The framework is evaluated using various data acquired by a driving simulator. This evaluation is done using different test drivers to highlight the effectiveness of the proposed approach for individualized situation recognition. The results demonstrate that the proposed framework can realize a successful driving situation recognition in terms of accuracy, detection rate, false alarm rate, and recognition elapsed time. It is shown that individualized situation recognition can significantly improve the recognition accuracy.