Digital Video Stabilization: Methods, Datasets, and Evaluation
Resumo
Video stabilization removes shaky camera motion from videos. In our thesis, we presented an extensive review, including a formal problem defini tion, meta-analysis, and other elements, resulting in two survey papers. We introduced new measures for stability assessment and studied the correlation between them and human perception. We also proposed a novel evaluation ap proach for 2D camera motion estimation. We then introduced NAFT, a semi-online DWS method with a neighborhood-aware mechanism to stabilize without an explicit stability definition. We supervised NAFT with SynthStab, our pro posed synthetic dataset. NAFT closed the quality gap with non-DWS methods while reducing the number of parameters and model size by 14×.
Referências
Liu, S., Yuan, L., Tan, P., and Sun, J. (2013). Bundled Camera Paths for Video Stabilization. ACM Transactions on Graphics.
Liu, S., Yuan, L., Tan, P., and Sun, J. (2014). Steadyflow: Spatially Smooth Optical Flow for Video Stabilization. In IEEE/CVF Conference on Computer Vision and Pattern Recognition.
Souza, M. R., Maia, H. A., and Pedrini, H. (2022). Survey on Digital Video Stabilization: Concepts, Methods, and Challenges. ACM Computing Surveys.
Souza, M. R., Maia, H. A., and Pedrini, H. (2023a). NAFT and SynthStab: A RAFT-based Network and a Synthetic Dataset for Digital Video Stabilization. Springer International Journal of Computer Vision (under review).
Souza, M. R., Maia, H. A., and Pedrini, H. (2023b). Rethinking Two-Dimensional Camera Motion Estimation Assessment for Digital Video Stabilization: A Camera Motion Field-based Metric. Elsevier Neurocomputing.
Souza, M. R., Maia, H. A., and Pedrini, H. (2023c). Survey on Digital Video Stabilization: Datasets and Evaluation. ACM Computing Surveys (under review).
Zhao, M. and Ling, Q. (2020). PWStableNet: Learning Pixel-Wise Warping Maps for Video Stabilization. IEEE Transactions on Image Processing.