Vehicle trajectory-based road type and congestion recognition using wavelet analysis

In many intelligent transportation system applications, understanding vehicle activity patterns of probe vehicles is becoming increasingly important for determining congestion levels on different roadways. Vehicle activity patterns can be characterized at a microscale level as vehicle velocity trajectories. Many ITS applications now use probe vehicles that are constantly monitoring and recording these velocity data as a component of traffic monitoring and management. Rather than transmitting probe vehicle velocity data to a centralized management center for processing (requiring high bandwidth), it is proposed to perform real-time on-board analysis of velocity trajectories to estimate roadway type and congestion level. These estimates can then be used for a variety of purposes including traffic information systems, intelligent real-time vehicle control systems, and energy consumption/emissions estimation. A method is presented to estimate roadway type and congestion level using a wavelet analysis technique combined with principal components analysis. This technique is applied to 128-second snippets of real-time vehicle velocity trajectories. The complexity of this combined wavelet/PCA approach is o(l³), where l Lt N. Training and recognition has taken place on a small data set of approximately 300 trajectories collected in the Held. Results thus far indicate an approximate 90 % correct estimation rate