GENERATING ROAD SURFACE ORTHOPHOTO FROM IMAGES OF MOBILE MAPPING VEHICLES

摘要

Orthophoto of road surface is a valuable data source for extracting road features and 3D road modeling. Road surface orthophoto were traditionally generated using aerial photographs or satellite imagery. However, road surface maybe obscured or sheltered by vegetation or manmade structures from aerial perspective. Images acquired from perspective of a ground vehicle would provide a higher resolution and free of obscuration photograph, and higher data update rate. For maintaining the flexibility of using consumer cameras or camcorders as image acquisition sensors, the interior and exterior orientation parameters are assumed unknown, only rough GPS coordinates are used in this proposed procedure. The lower half of the image within a front facing camera mounted on a vehicle is the area where road surface is present, these areas are use to perform this procedure. The transformation between two images of a flat surface in the object space can be represented by homography matrices. By utilizing this property, two consecutive image of a set of image sequence can stitch their road surface together using the homography matrix between the two images. All images along the same road segment can be stitched and forming a trapezoid shaped image. Pavement markings such as dashed lines and pedestrian crossing can be used to detect the two set of perpendicular vanishing lines, and these two vanishing points correspond to the vanishing lines can be used to rectify the stitched road surface image. After rectification, this road surface image is without the correct scale. The scale can be estimated using the actual dimensions of the pavement markings since the dimensions are fixed with regulations. Based on our preliminary results, the generated road surface orthophoto is useful for presentation purpose after removing the discolorations caused by stitching multiple images.