Real-Time Recognition of Signaling Lights in Road Traffic

A method for recognizing signal lights in traffic scenes in real time has been developed. It consists of a set of different feature extraction operations acting as sequential filters. In order to be classified as part of the image of an active signaling light, a set of pixels must pass each one of these filters. This multif ilter approach is model-based, and it comprises much of the available knowledge about signaling lights. In addition to color, this includes characteristics like the locations of the lights relative to the vehicle, their size, symmetry, and blinking frequencies. The individual filters, and the way how they c ooperate, have been designed in such a way that the system recognizes active signaling lights reliably, while false alarms are largely avoided. We have implemented the approach on a real-time vision system and tested it both in simulations under approximately real conditions and in real highway scenes by using a robot-car.