Detection of traffic signs using posterior classifier combination

Mobile mapping of environment information from a moving platform plays an important role in the automatic acquisition of GIS (Geographic Information Systems). The extraction of railway infrastructure from video frames captured on a driving train requires a robust visual object detection system that provides both high localization accuracy and the capability to cope with uncertain information. This work presents a radial basis functions (RBF) neural network that models appearance based object recognition of traffic lights and railway signs within a probabilistic framework. A comparison of different classifier combination strategies demonstrates that a classifier prioritization scheme based on an information theoretic selection criterion provides the best recognition performance.