AI CCTVs are being developed by a lot of companies throughout the world. Most of them follow one of the two structures below.

Our Approach

some other important points regarding why the others can not be used by us

• the 1 approach can not be presented by us as it needs a specialized camera. the 2 can’t be done as it seems immature of us to dump everything together in one thing.
• we also can’t make a full-fledged CCTV-connected system, as there will be no way for us to showcase it at the actual hackathon
• the 3 approach is the closest we can get to the actual CCTV implementation while letting us showcase the system.

AI model

• ActionXPose model from the Newcastle University
  • Angelini F 2020.pdf (ncl.ac.uk)
  • [Pose‐driven human activity anomaly detection in a CCTV‐like environment - Yang - 2023 - IET Image Processing - Wiley Online Library](https://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/ipr2.12664#:~:text=Therefore%2C a human pose-drivenand object-detector-based deep learning architecture,RGB based counterpartlimitations in different human activities classification.)
  • 1810.12126v1.pdf (arxiv.org)
• Continual learning - this will allow the AI to become more and more familiar with the place it is stationed at with time

Dashboard

• Statistical analysis of the anomalies detected in the following periods
  • 1 day
  • 1 week
  • 1 month
  • 6 months