CLOSE
aigocy@gmail.com

CUP

TWITTER (X) DRIBBBLE LINKEDIN

Configuration

COLORS
CUSTOM CURSOR
Skip to main content

espialtech

Menu
Start a Project

SiteGuard

Home
/
Blog
espial
July 27, 2026
No Comments
SiteGuard
Safety Analytics

Boundary surveillance, PPE compliance, intrusion detection and digital SOP enforcement
through real-time video analytics — running entirely on-premise.

Project Details

The client operated in an environment where sending camera footage to a cloud service was simply not an option — regulatory, contractual and cultural constraints all pointed the same way. SiteGuard is a desktop application that ingests live RTSP feeds from existing CCTV and IP cameras with multi-threaded decoding, and runs the entire deep learning pipeline on-device.

YOLOv8n handles person detection at 320x320 with a 50% confidence threshold; ByteTrack maintains multi-person tracking with IoU matching and a 30-frame history, enabling reliable IN/OUT counting; InsightFace provides 512-dimensional embeddings for face recognition where identity matters. PPE checks (helmet, vest) and restricted-zone violations are evaluated per frame, and every event is logged to a local SQLite database. Daily PDF and Excel compliance reports are generated and emailed automatically.

DELIVERABLES
Open vision models On-premise deployment Automated compliance reporting
INDUSTRY
Industrial Safety & Compliance

Project Research

The engineering constraint drove the architecture: everything had to run on commodity hardware already in the building, with no server, no cloud and no internet connection.

That meant choosing the smallest model that could still hold accuracy, tuning input resolution and confidence thresholds against real site footage rather than a benchmark, and packaging the whole application — models, database, reporting — as a single executable via PyInstaller that a site manager could install without an IT team.

Project Results

SiteGuard tracks at 20 FPS in real time with 95% PPE detection accuracy, deployed 100% on-premise. Compliance reporting, previously a manual clipboard exercise, is now generated and distributed automatically every day. Because it ships as a single executable with no external dependencies, rolling it out to a new site is an installation rather than a project.

EdgeForge
Model Compression
View Project

EdgeForge
Model Compression

DESCRIPTION
A unified compression pipeline — quantisation, pruning, distillation, ONNX and TensorRT — that shrinks models 6× while holding 90% accuracy, with benchmarks to prove it.
DELIVERABLES
Compression pipeline, Runtime optimisation, Benchmarking platform
INDUSTRY
Edge AI, MLOps