Multi-sensor based high-precision direct georeferencing of medium-altitude unmanned aerial vehicle images

A high-precision direct georeferencing method based on multi-sensor information for unmanned aerial vehicle (UAV) remote sensing is proposed in this research. The method makes contributions in three areas. First, the direct georeferencing model for the most popular system structure of fixed wing UAV loading of a camera with two degrees of freedom is established by using multi-sensor parameters. Second, based on this model, it performs image geographical reconstruction under complex motion and ultimate imaging conditions, which constructs the resulting image from the image frame, geographical frame and image pixels, and provides real geometric and geographic information. Third, in order to improve the accuracy of the georeferencing of the corrected image, a method of error prediction and compensation is presented. It converts the six factors affecting the accuracy of image georeferencing, to three factors for reducing the complexity. Then, the error prediction model is constructed, the parameters of which could be determined by supervised machine learning method. A medium-altitude UAV is employed to collect large amount of images and other sensor data. The experiments show quite encouraging results.