Cities require data on pedestrian movement to evaluate the use of public spaces. We propose a system using thermal cameras and Computer Vision (CV) combined with Geographical Information Systems (GIS) to track and assess pedestrian dynamics and behaviors in urban plazas. Thermal cameras operate independent of light and the technique is non-intrusive and preserves privacy. The approach extends the analysis to the GIS domain by capturing georeferenced tracks. We present a pilot study conducted in Copenhagen in 2013. The tracks retrieved by CV are compared to manually annotated ground truth tracks, and an example of pedestrian behavior is analyzed.