Multi-view Geometry for Rolling Shutter Camera

Organizers:     Yuchao Dai, 
The Australian National UniversityAustralia
                         Hongdong Li, The Australian National UniversityAustralia
                         Laurent Kneip, The Australian National UniversityAustralia
Time: 1:30pm-5:00pm, November 24 Coffee break: 3:00-3:30pm

Location: 101A 

Abstract: Rolling-Shutter CMOS cameras are getting more and more popularly used in real-world computer vision applications due to their low cost and simplicity in design. To use these cameras in 3D geometric computer vision tasks (such as 3D reconstruction, object pose, visual SLAM), the rolling shutter effect must be carefully accounted for. Simply ignoring this effect and relying on a global shutter method may lead to erroneous, undesirable and distorted results as reported in previous work. Recently, many classic 3D vision algorithms have been adapted to the rolling shutter case (such as absolute pose, bundle adjustment, relative pose, and stereo rectification).

In this tutorial, we will give a clear classification of all the multi-view geometry problems for rolling shutter camera, namely absolute pose, relative pose, stereo, triangulation, bundle adjustment, degenerate cases and etc. We will further relate the rolling shutter camera models with other non-conventional cameras such as push-broom camera, multi-camera system, X-slit camera, axis-camera, Lidar and etc. Build upon this pipeline, we will demonstrate how the rolling shutter multi-view geometry contributes to real world visual odometry, SLAM and large scale structure-from-motion.

Tutorial Program: