A spatiotemporal deep learning framework for detecting unusual human activity in surveillance videos
Manoj Kumar, Mantosh Biswas, Anoop Kumar Patel, Abhay Kumar, Kumar Abhishek, Ahamed Shafeeq B. M. · 2026
Read A spatiotemporal deep learning framework for detecting unusual human activity in surveillance videos by Manoj Kumar, Mantosh Biswas, Anoop Kumar Patel, Abhay Kumar, Kumar Abhishek, Ahamed Shafeeq B. M. (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0355690.txt.
Description
Manoj Kumar leads a team of six building software to spot unusual human activity in real surveillance footage, in PLOS ONE in 2026. Many anomaly detectors are trained on staged or synthetic clips and stumble in messy real scenes. The authors instead combined transfer learning with a two part network. A proven image model reads each frame for spatial content. A bidirectional memory network then reads the sequence forward and backward for timing.
Fused together, the two views judge whether behavior in a clip turns abnormal. The team evaluated the system on real world crime surveillance recordings rather than simulations, and report performance ahead of established comparison methods. Applications range from station platforms to hospital corridors, anywhere a human watcher cannot stare at every feed. The authors stress that training on genuine footage, not synthetic stand ins, is what makes the detector plausible for deployment.
Catalog record
| CREATOR | Manoj Kumar, Mantosh Biswas, Anoop Kumar Patel, Abhay Kumar, Kumar Abhishek, Ahamed Shafeeq B. M. |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51309 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |