Online video stabilization using a novel MeshFlow motion model
The MeshFlow is a spatial smooth sparse motion field with motion vectors only at the mesh vertexes. The MeshFlow is produced by assigning each vertex an unique motion vector via two median filters. The path smoothing is conducted on the vertex profiles, which are motion vectors collected at the same vertex location in the MeshFlow over time. The profiles are smoothed adaptively by a novel smoothing technique, namely the Predicted Adaptive Path Smoothing (PAPS), which only uses motions from the past.
To stabilize a video execute the script src/Stabilization.py
python Stabilization.py <path_to_video>
To run experiments ipython-notebook is present in src/mesh_flow.ipynb
The stable output video is saved to home
directory of cloned repository
Required Packages:
opencv: pip install --user opencv-python
numpy: pip install --user numpy
scipy: pip install --user scipy
tqdm: pip install --user tqdm
Optional Packages:
pip install --user cvxpy
Mesh Flow only operate on a sparse regular grid of vertex profiles, such that the expensive optical flow can be replaced with cheap feature matches. For one thing, they are similar because they both encode strong spatial smoothness. For another, they are different as one is dense and the other is sparse. Moreover, the motion estimation methods are totally different. Next, we show estimatation of spacial coherent motions at mesh vertexes.
results/old_motion_vectors
results/new_motion_vectors
A vertex profile represents the motion of its neighboring image regions. MeshFlow can smooth all the vertex profiles for the smoothed motions. It begin by describing an offline filter, and then extend it for online smoothing.
results/paths