Data fusion techniques for object space classification using airborne laser data and airborne digital photographs.

摘要

The objective of this research is to investigate possible strategies for the fusion of airborne laser data with passive optical data for object space classification. A significant contribution of our work is the development and implementation of a data-level fusion technique, direct digital image georeferencing (DDIG). In DDIG, we use navigation data from an integrated system (composed of global positioning system (GPS) and inertial measurement unit (IMU)) to project three-dimensional data points measured with the University of Florida's airborne laser swath mapping (ALSM) system onto digital aerial photographs. As an underlying math model, we use the familiar collinearity condition equations. After matching the ALSM object space points to their corresponding image space pixels, we resample the digital photographs using cubic convolution techniques. We call the resulting images pseudo-ortho-rectified images (PORI) because they are orthographic at the ground surface but still exhibit some relief displacement for elevated objects; and because they have been resampled using a interpolation technique. Our accuracy tests on these PORI images show that they are planimetrically correct to about 0.4 meters. This accuracy is sufficient to remove most of the effects of the central perspective projection and enable a meaningful fusion of the RGB data with the height and intensity data produced by the laser. PORI images may also be sufficiently accurate for many other mapping applications, and may in some applications be an attractive alternative to traditional photogrammetric techniques.; A second contribution of our research is the development of several strategies for the fusion of data from airborne laser and camera systems. We have conducted our work within the sensor fusion paradigm formalized in the optical engineering community. Our work explores the fusion of these two types of data for precision mapping applications.; Specifically, we combine three different types of data: the high resolution color images, the lower resolution near infrared (NIR) intensity images, and digital elevation model (DEM). We then investigate the use of a supervised statistical pattern recognition technique to combine these data for land-cover classification. We also investigate two decision-level data fusion algorithms: an expert system and an approach based on Dempster-Shafer evidential theory. (Abstract shortened by UMI.)