Image Reconstruction for Hyperspectral Pushbroom Cameras Using Visual-Inertial SLAM

Abstract

Pushbroom cameras are one of the most popular types of cameras when it comes to remote sensing hyperspectral imaging. However, since these cameras work by scanning a single line at a time, the image data can be distorted if there is no consistent relative motion between the camera and the scene to be captured. This constraint presents a substantial obstacle in modern battlefield deployments, where drones carrying these cameras may be subject to vigorous maneuvering. To address this problem, we propose a novel method in which we employ a visual-inertial SLAM system, rigidly attached to the hyperspectral camera, which captures the 3D geometry of the scene and the trajectory of the system. With this additional information, we project the scanned lines on the scene geometry and then reconstruct the images, effectively reversing the distortion caused by the camera movement. We illustrate the performance of our method on a data set that has been created by manually holding a hyperspectral pushbroom camera, causing significant distortions in the data due to natural human movement. When postprocessing the data with our method, we observe a clear reduction in distortion.