Stabilization of spherical videos based on feature uncertainty

A. Luchetti, M. Zanetti, D. Kalkofen, M. De Cecco

Research output: Contribution to journalArticlepeer-review

21 Downloads (Pure)


Nowadays the trend is to acquire and share information in an immersive and natural way with new technologies such as Virtual Reality (VR) and 360 video. However, the use of 360 video, even more the use of VR head-mounted display, can generate general discomfort (“cybersickness”) and one factor is the video shaking. In this work, we developed a method to make the viewing of 360 video smoother and more comfortable to watch. First, the rotations are obtained with an innovative technique using a Particle Swarm Optimization algorithm considering the uncertainty estimation among features. In addition, a modified Chauvenet criterion is used to find and suppress outliers features from the algorithm. Afterward, a time-weighted color filter is applied to each frame in order to handle also videos with small translational jitter, rolling shutter wobble, parallax, and lens deformation. Thanks to our complete offline stabilization process, we achieved good-quality results in terms of video stabilization. Achieving better robustness compared to other works. The method was validated using virtual and real 360 video data of a mine environment acquired by a drone. Finally, a user study based on a subjective and standard Simulator Sickness Questionnaire was submitted to quantify simulator sickness before and after the stabilization process. The questionnaire underlined alleviation of cybersickness using stabilized videos with our approach.

Original languageEnglish
Pages (from-to)4103-4116
Number of pages14
JournalVisual Computer
Issue number9
Early online date9 Jul 2022
Publication statusPublished - Sept 2023
Externally publishedYes


  • 360 video
  • Chauvenet’s criterion
  • Particle swarm optimization
  • Shaking
  • Uncertainty estimation
  • Video stabilization


Dive into the research topics of 'Stabilization of spherical videos based on feature uncertainty'. Together they form a unique fingerprint.

Cite this