GPI Seminar Series: Guillem Palou

Guillem Palou, Dense Depth Recovery using Optical flow
Tuesday January 21st, at 11:00, Seminar room D5-007

Abstract:
This talk explains the necessary steps to recover the 3D structure using only optical flow in monocular video sequences. The recorded scenes are assumed to be static, with no independently moving objects. From the camera rotations and translations, the algorithm is able to track points appearing in different frames and triangulate its position in a three dimensional space. Once the position for the points are known, a self calibration step is performed, allowing a metric reconstruction of the scene. Results are refined by minimizing non-linear error functions, a process known as bundle adjustment. After some post-processing of the depth maps, results are qualitatively comparable with some state of art algorithms.