Airborne monitoring of vehicle activity in urban areas

We describe possibilities to extract vehicles from data of differentairborne sensor systems. We focus on three frequency domains, namely(i) visual, (ii) thermal IR and (iii) radar. Corresponding to thedata different processing methods are demanded.(i)Vehicle detectionin aerial images relies upon local and global features. For modellinga vehicle on local level, a 3D-wireframe representation is used describingprominent geometric and radiometric features of cars including theirshadow region. A vehicle is extracted by a ‿top-down‿ matching ofthe model to the image. On global level, vehicle queues are modelledby ribbons that exhibit typical symmetries and spacing of vehiclesover a larger distance. Fusing local and global extractions makesthe result more complete. (ii) Particularly at night video sequencesfrom an infrared camera yields suitable data to assess the activityof vehicles. At the resolution of approximately one meter vehiclesappear as elongated spots. However, in urban areas many additionalother objects have the same property. Vehicles may be discriminatedfrom these objects either by their movement or by their temperatureand their appearance in groups. Additional knowledge from a vectormaps is helpful. The grouping of vehicles into rows along the marginsof the roads is performed by using of map knowledge as context. (iii)The increasing resolution of SAR data opens the possibility to detectvehicles even under bad weather conditions. Along-track interferometryallows to estimate vehicle movement. However, in urban areas SARspecific illumination phenomena like foreshortening, layover, shadow,and multipath-propagation burden the interpretation. Particularlythe visibility of the vehicles in inner city areas is in question.A high resolution LIDAR DEM is incorporated to determine the visibilityof the roads by a SAR measurement from a given sensor trajectoryand sensor orientation. Shadow and layover areas are detected byincoherent sampling of the DEM. In order to determine the optimalflight path a large number of simulations are carried out with varyingviewing and aspect angles.     «

We describe possibilities to extract vehicles from data of differentairborne sensor systems. We focus on three frequency domains, namely(i) visual, (ii) thermal IR and (iii) radar. Corresponding to thedata different processing methods are demanded.(i)Vehicle detectionin aerial images relies upon local and global features. For modellinga vehicle on local level, a 3D-wireframe representation is used describingprominent geometric and radiometric features of cars including theirshadow region. A vehi...     »