A vision-based statistical methodology for automatically modeling continuous urban traffic flows

We introduce an online video-based virtual sensor allowing to automatically estimate and forecast the number of vehicles passing through a road section over a continuous time interval. The strategy consists, in the first place, in defining a Motion Intensity Index (MII) whose role is to quantify the visual activity in a traffic video. A wavelet-based cause-and-effect statistical model is then used to match the actual number of vehicles to their respective motion scores. This leads to an efficient estimator of the urban traffic flow. The implementation is well optimized in such a way that the local sampling rate is directly proportional to the amount of visual activity in localized sub-shot units of the video. The procedure allows designing an autonomous sensor giving every moment a measure of the flow on a road section and an expectation of its future levels. The device can be very useful for optimizing transportation management, facilitating strategic decision-making, and analyzing networks with the purpose of optimizing transportation equipment efficiency.