Projects

Engineering Progress

Where the system has actually been built and tested so far — no live customer installations yet.

Every installation starts with a working core system — detection, matching, and alerting validated end-to-end in controlled testing — before any live rollout. The list below reflects where that process actually stands today, not a roadmap.

Ongoing

  1. No ongoing installations listed yet.

Completed

  1. Completed

    Phase 1 — Core Detection & Matching Pipeline

    GPU-accelerated face detection and matching pipeline, tested end-to-end with true-positive and true-negative verification against enrolled faces.

    Internal development · Verified working

Engineering facts

AccelerationCUDA (GPU)
GPU vs CPU benchmark~340x faster (measured, RTX 4060)
Confidence tiers3 — authorized / borderline / unknown
Lighting correctionCLAHE preprocessing + CNN detection

Case Study

Completed

Phase 1 — Core Detection & Matching Pipeline

Challenge

Build a face-recognition access-monitoring pipeline fast enough for live video, accurate enough for real-world lighting and angles, and honest about its own confidence rather than a blunt match/no-match.

Approach

CNN-based detection instead of the faster-but-weaker HOG detector, CLAHE lighting preprocessing before every detection pass, GPU acceleration via CUDA/cuDNN, and a three-tier confidence system (authorized / borderline / unknown) instead of a binary decision.

Result

GPU-accelerated detection measured at ~0.12s/frame versus ~40.6s/frame CPU-only (~340x faster) on an RTX 4060 — verified end-to-end with true-positive and true-negative testing against enrolled faces.

See Demo Get in Touch