Automated Attitude Determination for Pushbroom Sensors Based on Robust Image Matching

Abstract

Accurate attitude information from a satellite image sensor is essential for accurate map projection and reducing computational cost for post-processing of image registration, which enhance image usability, such as change detection. We propose a robust attitude-determination method for pushbroom sensors onboard spacecraft by matching land features in well registered base-map images and in observed images, which extends the current method that derives satellite attitude using an image taken with 2-D image sensors. Unlike 2-D image sensors, a pushbroom sensor observes the ground by changing its position and attitude according to the trajectory of a satellite. To address pushbroom-sensor observation, the proposed method can trace the temporal variation in the sensor attitude by combining the robust matching technique for a 2-D image sensor and a non-linear least squares approach, which can express gradual time evolution of the sensor attitude. Experimental results using images taken from a visible and near infrared pushbroom sensor of the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) onboard Terra as test image and Landsat-8/OLI images as a base map show that the proposed method can determine satellite attitude with an accuracy of 0.003° (corresponding to the 2-pixel scale of ASTER) in roll and pitch angles even for a scene in which there are many cloud patches, whereas the determination accuracy remains 0.05° in the yaw angle that does not affect accuracy of image registration compared with the other two axes. In addition to the achieved attitude accuracy that was better than that using star trackers (0.01°) regarding roll and pitch angles, the proposed method does not require any attitude information from onboard sensors. Therefore, the proposed method may contribute to validating and calibrating attitude sensors in space, at the same time better accuracy will contribute to reducing computational cost in post-processing for image registration.

Ultra-sensitive absolute angular position sensing using pixelated metasurface

“…Ultra-sensitive angle sensing is critical for determining angular positions at the micro and nano scales, with wide-ranging applications in fields such as precision machines, chip assembly, robotics control, satellite attitude determination and biomedical devices. [1][2][3][4][5] Utilizing a laser beam to measure the rotation or deviation of the object, laser angle sensors offer advantages of high precision, high speed, non-contact operation, and strong resistance to interference. 6 Common optical systems used for precisely obtain angular information include autocollimators and angle interferometers.…”

Local Deep Descriptor for Remote Sensing Image Feature Matching

“…Feature matching is the basis of various remote sensing processing tasks, such as image retrieval [1,2], object recognition [3,4] and image registration [5][6][7][8]. eature matching can also contribute to calibrate attitude sensor performance [9,10]. Taking remote sensing registration as an example, feature based methods are also mainstream methods in the practical application.…”

High-Resolution Optical Remote Sensing Image Registration via Reweighted Random Walk Based Hyper-Graph Matching

“…RANdom Sample Consensus (RANSAC) [21] is the mostly used methods to remove outliers in the matching points, and researchers have developed several RANSAC-like algorithms such as maximum likelihood estimation sample consensus [22], and maximum distance sample consensus [23]. Kouyama et al and Sugimoto et al utilized observation geometry as a constraint in their RANSAC steps and achieved robust performance of outlier rejection even under the cloudy condition [24,25], but the discrimination ability of this kind of algorithms largely depends on the geometric transformation model. The interpolation methods of bilinear interpolation or cubic convolution interpolation are frequently used to warp the input image to the reference image by using the estimated parameters of the image transformation model [26].…”