Visual Predictions of Traffic Conditions

Abstract

The proliferation of internet connected cameras means that drivers can easily access camera images to find out the current traffic conditions as part of their daily commute planning. This paper describes our novel system that will enhance the commute planning process by presenting the current traffic camera image and forecast the expected traffic conditions. The forecast traffic conditions are presented as database reference images. Image processing is applied to each camera image to extract quantitative attributes related to weather, lighting conditions, and traffic activity. The resulting data set of images is then clustered to identify different categories of traffic conditions based on the extracted features. Our prediction system uses a time-series of image attributes to forecast the traffic condition category in the near future (up to 9 hours). The commuter can then not only look at the current camera image, but also view a traffic condition forecast image.