Automatic Detection of Vehicles from Low-altitude UAV Remote Sensing Imagery

摘要

Vehicle information is very useful for transportation management, security surveillance, and military applications. Very high resolution remote sensing images allow automated monitoring of road traffic conditions and detection of vehicles. Spaceborne and airborne surveillance has several obvious advantages over current methods, which consist of expensive single-point measurements made from pressure sensors, video surveillance, etc.. The main limitation of using conventional satellite or aircraft surveillance is the time resolution; the continuously changing traffic situation must be deduced from a snapshot image. In this paper, a new method is presented to detect vehicles from low-altitude UAV (Unmanned Aircraft Vehicle)remote sensing imagery. As an alternative to conventional airborne and spaceborne remote sensing platforms, UAV provides an economical, flexible, real-time and mobile way to gather very high spatial resolution and high time resolution images. In the presented technique, color high resolution low-altitude UAV images (about 8 cm resolution on the ground)with standard metadata like calibration and orientation parameters as well as the image capture time are used. Vehicles were detected with an object-oriented method consisting of image multi-resolution segmentation, feature extraction and classification. The experimental results show that object-oriented classification techniques enhance quantitative analysis of traditional pixel-based change detection applied to very high resolution image data and facilitate the interpretation of ground objects. Experiments demonstrate that it is feasible to use this method to detect vehicles from low-altitude UAV remote sensing imagery. However, that also shows the deficiencies which guide our research in the future.