Developing of Objective Similarity Measures for Real-World Driving Behaviors

概要

論文の詳細を見る
Recent advances in ITS allow us to collect large amount of real-world multi-modal driving data for research study and analysis. In order to effectively utilize database, ability to automatically mine driving situations of interest is one of the essential steps. In this paper, we propose a probabilistic technique to measure similarity of driving behaviors based on posterior probability of driving modes in a driving space. The similarity distance is then obtained from correlation coefficient between two feature matrices. In addition, the framework allows adaptation scheme of driver model to better fit individual driving characteristics. Experimental results on car-following situations have showed that the proposed framework achieved 40% accuracy, while human achieved 49%, in the similarity-ranking task
2010-06-11