RPCAsurv

Robust PCA for use in surveillance footage

Contains:

• security.py
  • contains call to rpca via svt and creates data matrices from video
• securitypcp.py
  • same as security except solves nuclear norm min via PCP and not svt per alternating optimization
• pcp.py
  • algorithm for pricipal component pursuit from https://github.com/dfm/pcp
• playsparse.py
  • take result of rpca, either m, l or s (from m = l + s) and output to video in WD

Dependencies:

• Python version >3.X
• OpenCv
• Numpy
• Sufficent storage ~ 1.5 GB
• VLC video player to view output with mp4v codec

How to use:

• First put .py and mp4 into your working directory
• Run security.py to save M.csv, L.csv, and S.csv
• Note M = L + S per Robust PCA
• Num frames = 100 and other dimension choices are hardcoded, you can change this as you see fit
  • num iters and $\epsilon$ thresh are set to 10 and 1.7e-10 * l2(data) respectivley
• Run playsparse.py to output and save the video containing the sparse noise at 20 fps to your WD
• Reccomended to open output video in VLC or other 3rd party video player, out of box windows/mac players may not support relevant codec

References

• https://arxiv.org/pdf/1212.0467.pdf
• https://github.com/dfm/pcp
• https://arxiv.org/pdf/0912.3599.pdf
• https://www.hindawi.com/journals/mpe/2020/8812701/#EEq1
• https://github.com/facebookarchive/fbpca