Cloud-Induced Uncertainty for Visual Navigation
Abstract
This research addresses the numerical distortion of features due to the presence of clouds in an image. The research aims to quantify the probability of a mismatch between two features in a single image, which will describe the likelihood that a visual navigation system incorrectly tracks a feature throughout an image sequence, leading to position miscalculations. First, an algorithm is developed for calculating transparency of clouds in images at the pixel level. The algorithm determines transparency based on the distance between each pixel color and the average pixel color of the clouds. The algorithm is used to create a dataset of cloudy aerial images. Matching features are then detected between the original and cloudy images, which allows a direct comparison between features with and without clouds. The transparency values are used to segment the detected features into three categories, based on whether the features are located in the regions without clouds, along edges of clouds, or with clouds. The error between features on the cloudy and cloud-free images is determined, and used as a basis for generating a synthetic dataset with statistically similar properties. Lastly, Monte Carlo techniques are used to nd the probability of mismatching.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 26, 2014
- Accession Number
- ADA615951
Entities
People
- Alyssa N. Gutierrez
Organizations
- Air Force Institute of Technology