Operational Traffic Monitoring Using Very High Resolution Satellite Imagery

In this paper a solution for the extraction of vehicle counts from very high resolution satellite imagery intended for use in an operational setting, in order to accommodate road traffic authorities with traffic statistics is presented. In the proposed system, separate, fully automatic methods for each of the necessary steps: road detection, cloud and cloud shadow detection, and finally, vehicle detection and classification are developed. Input to the system is the satellite image and road vectors, containing geographical coordinates, including height information. The proposed system for extracting vehicle counts has been successfully demonstrated on a collection of WorldView-2 and Quickbird images from Norwegian rural area roads. The true detection rate is on average 85.4%, and false detection rate is 8.6%, i.e., the reported number of vehicles is on average 94% of the correct number. A key parameter to the road authorities is the annual average daily traffic (AADT), which is estimated based on short-term counts of traffic. The average error in AADT is estimated to 25% for low traffic roads (AADT<20,000 vehicles) using only two satellite images a year. The system will now be evaluated and implemented at the Norwegian Road Authorities as a complementary system for extracting traffic statistics.