Sentinel

Posted by:

Sentinel_2026

|

On:

|

UWI Mona — Department of Computing — May 2026

Meet Sentinel

An intelligent vehicle tracking system that follows targets across CCTV networks — no GPU, no cloud, and no expensive APIs required.

~70%
Next-camera prediction accuracy
6
Search pipeline phases
5
Architectural layers
The problem

Manual vehicle tracking across CCTV networks is slow and error-prone. By the time an operator spots a vehicle on one camera, it has already vanished from the next three. Sentinel automates this process — from detection to route reconstruction to probabilistic prediction of where the vehicle will appear next.

System interface
Sentinel dashboard
Live investigation dashboard showing CCTV feeds, reconstructed vehicle routes, and probabilistic next-camera predictions.
Six-phase search pipeline
01
Metadata filtering
Narrow the candidate vehicle pool before deeper matching.
02
Identity resolution
Compare embeddings using cosine similarity to match vehicles across cameras.
03
FAISS indexing
Accelerated nearest-neighbour search for cross-camera re-identification.
04
Trajectory reconstruction
Rebuild complete vehicle movement routes across the camera network.
05
Probabilistic prediction
Markov chain modelling predicts the most likely next camera location.
06
Ranked output
Prediction results are sorted and prioritised for investigators.
Key features
👁
Licence-plate-free tracking
Visual fingerprints maintain tracking even when licence plates are unreadable.
📊
Live dashboard
React + Leaflet.js dashboard with stream views, maps, and investigations.
⚙
No GPU required
Optimised for constrained environments without cloud infrastructure.
🔍
Natural language search
Extension-ready support for searches like “white SUV with roof damage”.
System architecture
User interface layer
React · Vite · Leaflet.js
Search layer
Six-phase pipeline · incremental streaming
Database layer
PostgreSQL · pgvector
Identity engine
YOLOv8 · OSNet · FAISS · PaddleOCR
Video processing pipeline
OpenCV · Python · FastAPI
Core technologies
Python FastAPI React PostgreSQL pgvector FAISS OpenCV YOLOv8 OSNet PaddleOCR
The team
Rohan Brown
UWI Mona
Steven-Kyle James
UWI Mona
Amoya Hamilton
UWI Mona
Delmika Johnson
UWI Mona
Department of Computing · UWI Mona · May 2026

Posted by

Sentinel_2026

in