Cutting-in vehicle recognition for ACC systems towards feasible situation analysis methodologies

Abstract

Models and methodologies for situation analysis, situation prediction and situation assessment have recently been proposed that undoubtedly base on fundamental theories. On the other hand, little effort has been taken to assess the feasibility of these approaches in the context of sensor systems currently available. This paper outlines a step-by-step prototype realization of a cutting-in vehicle recognition functionality for ACC-System (adaptive cruise control), that utilizes a probabilistic model for situation analysis and prediction. Cutbacks in the face of low sensor data quality are discussed and thereby a consistent methodology is presented to cope with uncertainty in both the developed models and the sensor data. The illustrated approach consistently combines sensor data filtering with Kalman filters and situation analysis with probabilistic networks in order to facilitate decision making under uncertainty. Statistics from test drives in traffic presents the capabilities and also the shortcomings of the approach taken, depicting the achievable enhancements and of course illustrating fail-operations of the system and their consequences. Moreover, the collected statistics is evaluated to come to a qualitative conclusion about what performance can be achieved also in the view of other applications.