HOME INTRODUCTION ALGORITHM DEMO MEMBER REFERENCE
Video Stablization with CUDA Implementation

INTRODUCTION

A video captured with a hand-held device (e.g., a cell-phone or a portable camcorder) often appears remarkably shaky and undi-rected. Digital videostabilization improves the video quality by re-moving unwanted camera motion. We implement a video stabilization method [Liu et al. 2013] which models cam-era motion with a bundle of (multiple) camera paths. The model is based on a mesh-based, spatially-variant motion representation and an adaptive, space-time path optimization. Also, the motion representation allows users to fundamentally handle parallax and rolling shutter effects while it does not require long feature trajectories or sparse 3D reconstruction. Furthermore, we speed up this method by parallelizing some parts with CUDA support.

A Single Global Path


Photo of Me

Bundled Paths


Photo of Me

ALGORITHM

Overall Pipeline
Model Estimation
Robust Estimation
Bundled Camera Paths
Path Optimization
Result Synthesis

DEMO

original video

bundled path

our result

SPEED COMPARASION

Under 640 x 360 resolution can achieve 25.54 fps

Under 1280 x 720 resolution is 7 times faster than paper

paper ours
extracting features 300ms 33ms
estimating motion 50ms 0.1ms
rendering the final result 30ms 21ms
total time 400ms 57ms
fps 2.5fps 17.5fps

MEMBER

陳學儀


Photo of Me

陳卓唅


Photo of Me

楊騏瑄


Photo of Me