Target modeling and dynamic classification for adaptive sensor data fusion

Abstract

Volkswagen research has developed a sensor fusion system for integrating different sensor data of the environment into a combined description for use with advanced driver assistance systems. Sensor models allow a detailed characterization of sensor properties. Due to the fused extensive information about the detected objects a classification into different dynamic classes, such as stationary or vehicle like moving objects, is reasonable. The classification switches between different dynamic models and is based on statistical tests that evaluate the fused data. The technique is tested with recorded sequences from stop and go scenes and shows robust and well-fitting classification results including reduced uncertainties compared to the usage of a universal model.