Spatial Intelligence Engine (SIE)

From pixels to spatial intelligence.

Welcome to the official documentation for the Spatial Intelligence Engine (SIE), a modular, OpenCV 5-based spatial intelligence layer for real-world vision applications. Instead of exposing only pixels, labels, or bounding boxes, SIE converts visual observations into a structured spatial representation — coordinates, zones, objects, movement, trajectories, and events — that other applications can consume through a clean API.

SIE bridges raw camera feeds and spatial imagery with precise metric coordinate systems, enabling autonomous agents, industrial robots, and human operators to share a common operational map in real time. Perception and spatial representation are deliberately kept separate, so object detection, point grounding, tracking, or depth models can be replaced without redesigning the spatial API.

5-Minute Quickstart

Get your first spatial stream ingested, calibrated, and analyzed in minutes using our Python SDK or Docker instance.

pip install spatial-intelligence-engine

from sie import SpatialEngine, GridConfig

# Initialize engine with warehouse feed
engine = SpatialEngine.connect("rtsp://warehouse-cam-01.local:554/stream")

# Configure 8x12 metric grid
grid = GridConfig(rows=8, columns=12, line_thickness=1)
calibrated_scene = engine.calibrate(grid)

print(f"Grid active with {calibrated_scene.total_cells} total cells.")
calibrated_scene.run_analysis(model="nanodet", tracking=True)

System Requirements

Sources & Ingestion Module

SIE supports multi-modal media ingestion. Whether you are uploading static high-res JPEGs of facility layouts, connecting live USB sensors, or streaming RTSP security feeds, the pipeline normalizes frames into synchronous spatial matrices.

Grid Calibration & Setup

Calibrating your physical environment turns raw pixels into actionable metric grid coordinates `(x, y)`. Use the interactive UI or configure programmatically.

from sie.calibration import GridCalibrator

calibrator = GridCalibrator(image="warehouse.jpg")
grid_matrix = calibrator.generate(
    rows=8,
    cols=12,
    line_thickness=1,
    color="#3B82F6"
)
grid_matrix.export_config("config/warehouse_grid.json")

REST API & Webhooks Reference

Interact directly with the SIE daemon using standard HTTP REST endpoints or configure outbound webhooks for real-time event triggers.

Docker Compose Deployment

Deploy SIE inside production environments with our official containerized docker-compose stack.

version: '3.8'
services:
  sie-engine:
    image: spatialintelligence/engine:v0.2.0
    ports:
      - "8080:8080"
    environment:
      - SIE_ENV=production
      - ACCELERATION=cuda
    volumes:
      - ./data:/app/data

Object Detection & Analysis

Run real-time object detection, tracking, and movement prediction with calibrated spatial context.

from sie import SpatialEngine

engine = SpatialEngine.connect("rtsp://cam.local/stream")
engine.calibrate(rows=8, cols=12)
engine.run_analysis(model="nanodet", classes=["person","forklift","box"], tracking=True, prediction=True)

Predictive Analytics & Reporting

Aggregate telemetry, compute zone occupancy, and export operational reports.

report = engine.generate_report()
print(report.summary())
report.export_csv("reports/hourly.csv")

Python SDK Client

Programmatically control SIE with the Python client library.

from sie.client import SieClient

client = SieClient("http://localhost:8080")
client.upload_image("layout.jpg")
client.calibrate(rows=8, cols=12)
client.analyze()

Webhooks & Real-time Feeds

Receive push notifications for detection events and configure callbacks.

{
  "url": "https://your-app.example.com/webhook/sie",
  "events": ["detection.created", "zone.occupied"],
  "secret": "your_webhook_secret"
}

Edge Device Deployment

Deploy to NVIDIA Jetson, Intel NUC, or ARM64 devices with minimal footprint.

docker run --gpus all -p 8080:8080 spatialintelligence/engine:edge

Security & Compliance

Best practices for securing deployments, managing secrets, and network hardening.

• Use environment variables for secrets; never commit credentials
• Enable TLS for all external endpoints
• Restrict network access to API and webhook endpoints
• Apply regular updates and image scanning

System Architecture

SIE is a modular middleware layer that sits between computer vision and real-world spatial applications. The defining architectural principle is that perception and spatial representation are separate: detection, grounding, tracking, and depth models are interchangeable components behind a stable spatial API.

A representative structured output from the engine looks like this:

{
  "object_id": "object_01",
  "type": "person",
  "image_position": [642, 381],
  "spatial_position": [2.41, 1.73],
  "zone": "B4",
  "timestamp": 12.4
}

Applications no longer need to know that an object is at image coordinate (642, 381) — they receive where it is in the environment, which zone it occupies, how it is moving, and what spatial event has occurred.

Five-Phase Development Roadmap

The engine is built incrementally, starting from a deterministic spatial foundation and progressively adding perception, motion, and reasoning.

AWS Architecture and Services

AWS accelerates development and provides infrastructure for larger experiments, model inference, performance benchmarking, and the optional agentic layer. It is an enabling platform rather than a hard dependency — the core engine runs locally with the same architecture.

COOL & Agentic Vision Paths

Evaluation Method

Evaluation is staged so that each capability has measurable criteria rather than relying on a single subjective demonstration.

Target Users & Applications

The long-term objective is to make spatial information a reusable interface rather than something every computer-vision application has to build independently.