Motion Estimation from Sensor Images

Abstract

Recently, with the advances in computer vision algorithm, an alternative mechanism to navigate the lander to its precise location on interplanetary mission to moon and mars has been studied. Keeping up with the current trend, ISRO has planned to use vision based navigation in its next mission to moon. This report presents horizontal velocity, angular velocity and altitude changes for accurate landing of lander based on the extraction and tracking of Scale Invariant Feature Trans- form features through image sequences. Implementation of SIFT operator is analyzed and adapted to be used in a feature-based motion estimation algorithm. Nearest Neigh- bor approach is implemented to match SIFT features and RANSAC is applied to remove matched outlier features. After that Transformation matrix is used to find rotation and scale information. The algorithm was tested on images acquired by Moon Impact Probe onboard ISRO's Chandrayan-1. For its robustness and accuracy, the results obtained were found to be comparable with simulated dataset.