Comparison of Vehicle Detection Using Very High-Resolution Satellite Images

Abstract

Traffic can be monitored using data obtained from mobile or permanent sensors such as induction loops, bridge sensors or cameras. This is an opportunity to obtain traffic data on main roads, but data from large parts of the road network is not available. Today´s optical sensors on satellites provide images covering large areas with resolution better than 1 meter and with frequency better 1 week, which can provide us with various information. Such information is important for urban and transport planning, intelligent transport systems, emergency control etc. Panchromatic imagery from WorldView3 was processed. The pilot area for WorldView3 is located in Prague, close to the Old Town Square. Panchromatic images were processed in two software. First software was ENVI and second was CATALYST Pro. Object detection was performed, then training data were created and finally classification methods were used. ENVI offers three classification methods (SVM, PCA, KNN) and CATALYST Pro offers two classification methods (SVM, RT). The detection of vehicles was relatively successful, especially in open public places without shade or vegetation. The detection of dark vehicles had the best results. The detection of vehicles in shadow had the worst results.