Geometric correction and stabilization of images collected by UASs in river monitoring

While the field conditions and the choice of an image velocimetry technique usually have the largest impact on the accuracy of unmanned aerial systems (UAS) velocimetry, significant errors can be associated with the UAS hardware. Three such sources of errors are usually considered and accounted for: (1) errors related to the imperfections of the camera optics and sensor, (2) errors introduced by the UAS motion relative to the water, and (3) projection errors related to the positioning of the camera relative to the region-of-interest. The first source produces geometric distortions whose intensities depend on the type of the camera and quality of its manufacturing. The second source of errors is related to the UAS platform itself—even in favorable weather conditions the platform and thus the camera as well usually exhibit unwanted motion as a result of its (electro)mechanics, which reduces the overall velocimetry accuracy. Once the camera distortions have been minimized and the images are stabilized, the mapping of the information from images to the real-world coordinates can begin. A specific procedure—orthorectification—aims to ensure that the image coordinate system is aligned with the water surface. This technique emulates nadir positioning of the camera (even in cases when camera is not in perfect nadir) and thus stabilizes the ratio between image and real-world distances. The accuracy of velocity estimation greatly benefits from the reduction of the geometric distortion, orthorectification, and the reduction of residual motion of the camera, which are commonly performed in the preprocessing stage of UAS velocimetry. This chapter explains various sources of errors in the acquired optical data along with the effects they can have on the final velocimetry results, and the corresponding methods for their reduction/removal. Finally, a list of tips and good practices is provided which can facilitate the reduction of UAS velocimetry errors.