Usage and Improvement of Bundle Adjustment Method for Determining Image Orientation

Please enable javascript in your browser.

Mousavi Khoshdel, Hossein | 2017

769 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 50246 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Manzouri, Mohammad Taghi
Abstract:
Pose and motion estimation using a set of images is a common application in photogrammetry and one of the most popular methods in doing so is Bundle adjustment. This method later got adapted to the field of computer vision and is mainly used to reconstruct a 3d scene while also estimating pose and motion of the camera. There are a variety of other methods used to estimate pose and motion. Most non-visual methods like using inertial navigation system and GPS suffer from shortcomings like accumulative error, availability problems or low accuracy. While most vision based one are computationally expensive. In this thesis two methods are proposed to increase accuracy and efficiency of one variant of Bundle Adjustment which avoids estimating scene structure and just estimates pose and motion. One method is to take advantage of available data from an inertial sensor to reduce outliers in data extracted from images and the other is running bundle adjustment inside two sliding windows to increase speed while maintaining good accuracy
Keywords:
Sensor Fusion ; Optimization ; Navigation ; Bundle Adjustment ; Three Dimensional Reconstruction