Security & Vision Intelligence Platform

An on-premise computer-vision platform integrated with existing cameras for security monitoring, attendance, identification and operational intelligence.

Business Problem

Facility security and attendance depended on manual monitoring of camera feeds and entry points — creating missed incidents, slow detection, and a heavy manual workload.

Users

Security and facility operations teams monitoring entry points and camera feeds across the site.

Constraints

Real-time processing Existing camera infrastructure Mixed CPU/GPU environments Concurrent detection tasks

Solution

I designed and deployed a vision platform combining fire and violence detection, license-plate recognition, driver-ID scanning, face-recognition attendance and people counting, with real-time alerting across CPU and GPU environments.

Architecture

Camera Feeds

→

Detection Models (Fire · Violence · Plates · Faces)

→

CPU/GPU Inference

→

Real-Time Alerting & Attendance

Your Contribution

Designed

the system architecture across detection and recognition models.

Integrated

multiple vision models onto existing camera feeds.

Built

real-time processing and alerting workflows.

Optimized

inference across CPU and GPU environments, and containerized components.

Technology Stack

Python OpenCV YOLO Face Recognition OCR Docker

Results

~95%

fire & violence detection accuracy

~90%

license-plate recognition accuracy

Automated

attendance & real-time alerting

Challenges & Trade-Offs

GPU vs. CPU execution

GPU inference gives more headroom for running multiple detection models concurrently in real time. CPU deployment trades some throughput for lower infrastructure cost on sites without GPU hardware.

Confidentiality

Enterprise project. This page uses architecture diagrams rather than real footage or facility interfaces.