Production Case Study 06 • Verified Enterprise System

Plug & Play — Retail POS-Integrated Video & NVR Platform

Custom Linux NVR Java / RTSP C# Desktop Client POS & Scale Overlays PostgreSQL Loss Prevention

Engineered an end-to-end retail surveillance platform from the ground up: a headless Linux Java NVR capturing RTSP camera streams, synchronized POS/scale data overlays, and a C# forensic investigation desktop client.

Role End-to-End System Architect
Timeline 2018 – 2026 (7-Year Run)
Scale & Throughput 478+ Cameras • 31+ Supermarkets
Client / Organization Plug & Play (SPAR, Pick n Pay)
Executive Brief

Operational Snapshot & Impact

High-stakes systems integration demands real-world reliability, sub-second latency, and deterministic execution under peak production load. Here is the operational profile:

Role End-to-End System Architect
Timeline 2018 – 2026 (7-Year Run)
Scale & Throughput 478+ Cameras • 31+ Supermarkets
Client / Organization Plug & Play (SPAR, Pick n Pay)

The Challenge: Rigid Proprietary Surveillance

Retail loss-prevention teams were struggling with off-the-shelf commercial NVRs that treated video recording as an isolated silo disconnected from cash register transactions. When investigating suspected shrinkage, sweethearting, or till fraud, auditors had to manually scrub through hours of unsynchronized footage matching timestamps against paper receipts.

The client required a bespoke, maintainable ecosystem capable of ingesting arbitrary IP camera RTSP streams (Hikvision, Dahua), tapping live checkout scanner transactions, interfacing butchery and deli weighing scales, and synchronizing video with financial transaction data in real time.

Headless Java NVR Architecture

Rather than packaging an existing commercial VMS, Warren architected and coded a custom Network Video Recorder (NVR) in Java deployed on Linux servers:

IP Cameras (Hikvision & Dahua) → Multi-Channel RTSP Video Streams
↓ Continuous Ingestion & Chunking
Custom Java Linux NVR Service → Segmented Local Video Storage
↓ Parallel Transaction Feed
POS Tills & Retail Scales (RS-232 / TCP-IP) → Ingestion Daemon
↓ Millisecond Timestamp Synchronization
PostgreSQL Relational Storage → Till/Camera Spatial Mapping
↓ On-Demand Forensic Retrieval
C# Desktop Client via HTTP Stream & Embedded LibVLC Player

The Java NVR connected directly to RTSP camera feeds, handled network reconnects automatically, wrote rolling segmented video files to disk, and exposed a lightweight HTTP streaming server for client playback.

POS & Butchery Scale Overlays

The system extended deep into retail floor hardware, tapping both front-of-store checkouts and service departments (butchery, deli, bakery, produce):

  • Hardware Protocol Interfacing: Captured raw data from legacy RS-232 serial scales via serial-to-Ethernet converters, as well as modern network scales over native TCP/IP.
  • Live Heads-Up Video Overlays: Burned and synchronized dynamic metadata directly into video playback: displaying PLU codes, item descriptions, gross weight, tare deduction, calculated price, till ID, and cashier number.
  • Instant Visual Verification: Allowed auditors to visually verify whether the cut of meat on the scale matched the PLU item code entered by the operator.

C# Forensic Investigation Client

Warren developed a responsive Windows desktop workstation in C# utilizing the VLC media engine for fluid timeline playback:

  • Instant Transaction Lookup: Auditors could query by slip number, barcode, cashier, or date range and instantly launch video at the exact second the barcode was scanned.
  • Multi-Camera Synchronization: Linked overhead till cameras with wide-angle entrance cameras to trace suspect movements across the store.
  • Live Grid Monitoring: Real-time multi-channel matrix monitoring alerting operators to offline cameras or network dropouts.

Flagged Item Loss-Prevention Alerts

The platform transformed surveillance from passive recording into proactive loss prevention. Store managers could configure "red-list" items (such as high-value electronics, premium meats, or baby formula).

When a flagged barcode was scanned at any till, the backend immediately dispatched an asynchronous message over the queue to the security control room, popping up the live camera feed for immediate operator observation.

Production Scale & 7-Year Track Record

The platform achieved exceptional commercial longevity and operational scale:

  • Massive Footprint: One deployment managed 279 IP cameras across 21 retail clients; another supported 199 cameras across 10 flagship SPAR and Pick n Pay supermarkets.
  • 7-Year Production Run: First deployed in 2019, the platform operated continuously into 2026. Two full years passed after the initial rollout before the first client support request occurred—which proved to be a physically full hard drive rather than a software bug.
  • Measurable Shrink Reduction: Provided irrefutable video and metadata evidence for hundreds of internal theft, sweethearting, and vendor fraud audits.

Technology Stack

Backend & NVR Java, Linux Headless Daemon, Multithreaded RTSP Capture
Client Workstation C#, .NET Windows Forms, LibVLC Video Engine Component
Database PostgreSQL (High-throughput transaction and camera spatial mapping)
Hardware Integration Hikvision & Dahua IP Cameras, Retail Weighing Scales, POS Terminals
Protocols & I/O RTSP, HTTP Streaming, RS-232 Serial, TCP/IP Sockets, Messaging Queues
Retail Deployments Supermarkets (SPAR, Pick n Pay), Multi-Site Franchise Groups
Enterprise Architecture Consulting

Facing Similar Systems Integration Bottlenecks?

Get a 48-Hour High-Stakes Architecture Audit ($8,500) to diagnose bottlenecks and design a zero-downtime execution roadmap with Warren Basterfield.

Book 48-Hour Audit ($8,500) ➔ Consult Principal Architect