Using Computer Vision with Instantaneous Vehicle Emissions Modelling

Air pollution and in particular PM2.5 emissions are a major problem worldwide. Road transport is a significant contributor to PM2.5 emissions in urban areas and as such it is important to understand and be able to accurately model the effects of vehicles on PM2.5 emissions. In this paper a computer vision algorithm is introduced which is able to extract vehicle trajectories from video footage. The algorithm has a 100% accuracy for overall total vehicle counting. Comparing the speeds predicted by the computer vision script to manually following a single vehicle feature on the video file, the average relative speed accuracy is 2.7% at a 1 Hz time resolution. Using these vehicle trajectories in an instantaneous vehicle emissions model and also as input to COPERT v5, tailpipe PM2.5 emissions were estimated and compared to on-road measurements. It was shown that a local sensor is not sufficient to determine vehicle tailpipe emissions due to the influence of meteorological conditions and other emission sources. Combining computer vision with an instantaneous vehicle emissions model is a useful method to evaluate changes in emissions caused by transport policies.