Skip to Main content Skip to Navigation

Quality Evaluation for Stitched Panoramic Videos

Abstract : High quality panoramic videos for immersive VR content are commonly created using a rig with multiple cameras covering a target scene. Unfortunately, this setup introduces both spatial and temporal artifacts due to the difference in optical centers as well as the imperfect synchronization. Traditional image quality metrics cannot be used to assess the quality of such videos, due to their inability to capture geometric distortions. In this thesis, we propose methods for the objective assessment of panoramic videos based on optical flow and visual salience. We validate this metric with a human-centered study that combines human error annotation and eye-tracking.An important challenge in measuring quality for panoramic videos is the lack of ground truth. We have investigated the use of the original videos as a reference for the output panorama. We note that this approach is not directly applicable, because each pixel in the final panorama can have one to N sources corresponding to N input videos with overlapping regions. We show that this problem can be solved by calculating the standard deviation of displacements of all source pixels from the displacement of the panorama as a measure of distortion. This makes it possible to compare the difference in motion between two given frames in the original videos and motion in the final panorama. Salience maps based on human perception are used to weight the distortion map for more accurate filtering.This method was validated with a human-centered study using an empirical experiment. The experiment was designed to investigate whether humans and the evaluation metric detect and measure the same errors, and to explore which errors are more salient to humans when watching a panoramic video.The methods described have been tested and validated and they provide interesting findings regarding human-based perception for quality metrics. They also open the way to new methods for optimizing video stitching guided by those quality metrics.
Complete list of metadatas

Cited literature [76 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Tuesday, March 26, 2019 - 2:43:12 PM
Last modification on : Wednesday, September 16, 2020 - 3:35:45 AM
Long-term archiving on: : Thursday, June 27, 2019 - 4:54:06 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01979682, version 2



Sandra Nabil Mahrous Yacoub. Quality Evaluation for Stitched Panoramic Videos. Computer Vision and Pattern Recognition [cs.CV]. Université Grenoble Alpes, 2018. English. ⟨NNT : 2018GREAM067⟩. ⟨tel-01979682v2⟩



Record views


Files downloads