Multi-body Self-Calibration
Andrea Porfiri Dal Cin (Politecnico di Milano),* Giacomo Boracchi (Politecnico di Milano), Luca Magri (Politecnico di Milano)
The 33rd British Machine Vision Conference
Abstract
One of the main assumptions behind Structure-from-Motion is that of a rigid scene, i.e., the scene is static or composed of a single moving object. The rigidity constraint -- typically encoded in the Kruppa equations -- is at the core of self-calibration and enables Euclidean upgrading from uncalibrated images. In this work, we show how it is possible to improve self-calibration by considering a dynamic scene composed of multiple moving rigid objects. The rationale of our solution is that each rigid motion provides a useful constraint that can be used to better estimate the intrinsics of the camera. Specifically, we introduce a self-calibration method for a single camera that exploits motion segmentation to identify rigid motions. Our solution capitalizes on all the available epipolar relations to robustly initialize the camera parameters, which are then optimized through nonlinear refinement. Experiments on real-world data show that our approach is comparable to state-of-the-art self-calibration methods when the scene is static and improves performance in the case of dynamic scenes. The code and a dataset with images of dynamic scenes and ground truth intrinsics are available at https://github.com/andreadalcin/MultiBodySelfCalibration.Video
Citation
@inproceedings{Cin_2022_BMVC,
author = {Andrea Porfiri Dal Cin and Giacomo Boracchi and Luca Magri},
title = {Multi-body Self-Calibration},
booktitle = {33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
publisher = {{BMVA} Press},
year = {2022},
url = {https://bmvc2022.mpi-inf.mpg.de/0471.pdf}
}
Copyright © 2022
The British Machine Vision Conference is organised by