Reconstructing fixed time traffic light cycles by camera data analytics

Abstract

Traffic light signals are a high impact factor on inner city traffic flows. Consequently, the prediction of traffic light properties can have a positive impact on environmental factors like CO 2 reduction or the creation of red-light warning assistants. In this paper, an approach for camera data based reconstruction of fixed time traffic light signal program properties is introduced. Adapted variants of this approach are presented to determine the color phase timings and phase durations of traffic lights, in relation to provided input data. This approach is based on camera data and the knowledge of the traffic light system's cycle time. It can utilize further additional traffic light properties as well as legal-or environmental-constraints, to create predictions on even fewer data. The base target of the approach is to predict parameters of traffic light programs. These can be used to predict traffic light phases and enable at later stages to reconstruct traffic light programs. Its performance is measured by achieving the required coverage and accuracy, based on as little information as possible. The provided approach is considered suitable to predict green start times and green phase duration accurately after several intersection crossings.