Infynix Vision AI Case Study | Infynix Solutions
Case Study: See how we helped a healthcare business unify systems and scale faster.
Development

AI Face Recognition & Surveillance System

INFYNIX AI SURVEILLANCE SYSTEMS

Facilities relying on manual attendance and access checks needed a way to identify and track individuals across multiple camera feeds in real time. Infynix built a containerized face-recognition surveillance platform combining a high-accuracy detection engine with live video streaming and a mobile enrollment app.

Book Discovery Session ↗
Infynix Vision AI Interface
OPERATIONAL TRANSFORMATION

Before & After Comparison

BEFORE
  • Identity verification and monitoring done manually across camera feeds
  • No way to track a person consistently as they moved between cameras
  • Face enrollment required manual, in-person setup with no mobile option
  • No safeguard against false positives when a person briefly left frame
AFTER
  • Real-time face recognition running on InsightFace (ArcFace/SCRFD) and YOLOv8
  • Cross-camera de-duplication so the same person isn’t double-counted across feeds
  • OUT confirmation buffer logic to prevent false exits from momentary occlusion
  • Flutter mobile app for on-the-go face enrollment
  • Geofencing to trigger zone-based alerts automatically

"The system reliably tracks people across every camera without losing them at the handoff points, and the mobile enrollment app means we're not tied to a desk to add someone new."

Facility Operations Lead
These outcomes reflect how the Infynix Vision AI team uses the system day to day.

Objectives

Build a containerized, real-time face recognition system with reliable cross-camera tracking, low-latency inference, and mobile-based enrollment.

Strategy

Combined InsightFace and YOLOv8 for detection/recognition with DeepSort for tracking and FAISS for fast similarity search, wrapped in a FastAPI engine with ONNX Runtime optimization and a TTL-based recognition cache.

Execution

Built the Docker-based FastAPI face engine, integrated MediaMTX for RTSP/WebRTC streaming, implemented OUT confirmation buffering and geofencing logic, shipped a Flutter enrollment app, and delivered handover documentation for the engineering team.

PythonFastAPIInsightFace (ArcFace/SCRFD)YOLOv8DeepSortFAISSONNX RuntimeDockerNode.jsReactMediaMTXFlutter

MEASURABLE OUTCOME

Reliable real-time recognition across multiple camera feeds with no duplicate tracking, faster enrollment via mobile, and a fully documented handover for ongoing maintenance.

Book Discovery Session ↗