SDK & Runtime Engine

The perception engine.
You build the product.

AuraSense ships as a C++ SDK with Python bindings. Embed neuromorphic crack detection into your drone inspection platform in days, not months. P95 <5ms on CPU-only hardware.

C++17 / Python 3.8+ | Linux & Linux/ARM | No cloud dependency
<5ms
P95 Latency
30fps
Sustained Throughput
0
GPU Required
6
STDP Lanes
Integration Guide

From frames in to events out

A four-stage pipeline from raw drone frames to structured crack detection events. All on-device, all real-time.

1

Feed frames

Push frames via RTSP, RTMP, or the C++ push API. The engine ingests at up to 30fps. Pass a FrameBuffer pointer directly from your camera driver for zero-copy ingestion.

2

Configure lanes & thresholds

Load a JSON config to select active STDP lanes (1–6), set crack score thresholds, enable degraded-mode policy, and configure output destinations (callback, WebSocket, or file log).

3

Read FRAME_LOG events

Every processed frame emits a structured event: crack_score, risk_score, frame_id, roi, and per-lane breakdown. Subscribe via C++ callback or read from the Python binding.

4

Visualise or forward

Pipe events to your inspection dashboard, a WebSocket ground station stream, or the bundled Streamlit dashboard. Trigger alerts at custom risk thresholds. Log to JSON for audit trails.

Output Format

Every frame. A structured event.

Every processed frame emits a FRAME_LOG event your application can consume via C++ callback, Python binding, or WebSocket stream.

Field Type Description
frame_id uint64 Monotonic frame counter from engine start
crack_score float [0,1] Crack detection confidence for this frame
risk_score float [0,1] Fused risk scalar — threshold this to trigger alerts
roi BoundingBox Pixel-space region of interest (x, y, w, h)
alarm bool True when risk_score exceeds your configured threshold
Hardware Targets

CPU-only. By design.

The AuraSense SDK runs on standard x86-64 and ARM64 Linux hardware. No CUDA, no NPU, no special silicon required.

x86-64 (Intel / AMD)

Intel NUC, mini-PC, or any Linux laptop. Requires AVX2. Tested on Core i5/i7 and Ryzen 5/7. P95 <5ms on 4-core systems at 30fps.

ARM64 (Jetson / RPi 5)

NVIDIA Jetson Orin Nano (CPU only, no CUDA), Raspberry Pi 5, or any ARM64 SBC running Ubuntu 20.04+. Ideal for tight payload budgets.

Requirements

Linux (Ubuntu 20.04+ or Debian 11+). 4 GB RAM minimum. 8 GB recommended for concurrent inference + logging. GCC 10+ or Clang 12+.

Use Cases

Built for the hardest environments

SFSVC deploys wherever conventional connectivity fails and inspection stakes are highest.

Building Facades

Detect hairline cracks, spalling, and water damage on high-rise exteriors without scaffolding. Inspect 30 floors in under an hour.

Bridge & Infrastructure

Continuous monitoring of bridge decks, piers, and expansion joints. SFSVC identifies sub-millimeter cracks invisible to the human eye.

Runway & Airfield

Automated surface inspection for airports. Detect FOD, cracks, and pavement degradation at taxiway speeds without halting operations.

Ready to add neuromorphic crack detection to your platform?

We're onboarding pilot integrators now. Tell us your hardware stack and inspection use case — we'll spec the integration for you.

Apply for Pilot SDK Back to Home