Eight-Layer Pipeline

How F-GRAF Works

An eight-layer intelligence pipeline that transforms raw visual data into prioritized, explainable forensic insights with threat assessment.

🔍
LAYER 01

Evidence Detection

See What's There

AI-powered object detection identifies physical evidence items in images and video frames with high accuracy and speed.

●Lightweight YOLOv8-based detection model
●Multiple objects detected per frame
●Bounding boxes with confidence scores
●Works with cameras, drones, and video feeds
🔗
LAYER 02

Scene Understanding

Build the Map

Detected objects become nodes in a spatial graph. Distances, overlaps, and directional relationships are computed automatically.

●Objects converted to graph nodes
●Spatial relationships computed
●Proximity, overlap, and adjacency analysis
●Directional awareness (N, S, E, W)
🧠
LAYER 03

Spatial & Temporal Reasoning

Understand the Context

Rule-based reasoning engine analyzes evidence relationships. Every conclusion is traceable to a specific rule — no black box.

●7+ deterministic reasoning rules
●Co-occurrence detection
●Cluster identification
●Multi-frame temporal persistence
⚡
LAYER 04

Evidence Priority & Scoring

Focus What Matters

Composite scores rank evidence by proximity, connectivity, persistence, and confidence. Each item gets a HIGH, MEDIUM, or LOW priority.

●Weighted composite scoring
●Proximity & connectivity factors
●Temporal persistence bonus
●Explicit reason for every score
📐
LAYER 05

Spatial Analysis

Heatmaps & Anomalies

Advanced density heatmaps, anomaly detection using z-scores, and region-of-interest identification for deep spatial understanding.

●Evidence density heatmap grid
●Statistical anomaly detection
●ROI zone identification
●Quadrant analysis & clustering
📊
LAYER 06

Graph Analytics

Network Intelligence

PageRank, centrality metrics, and community detection reveal the most important evidence nodes and hidden relationships.

●Degree & betweenness centrality
●PageRank importance scoring
●Community detection algorithms
●Critical node identification
🎯
LAYER 07

Threat Assessment

Risk Evaluation

Multi-factor threat scoring synthesizes all layers into actionable risk levels with automated scene narrative generation.

●5-level threat classification
●Multi-factor risk analysis
●Automated scene narrative
●Actionable intelligence items
📋
LAYER 08

Investigative Insights

Guide the Investigator

Human-readable reports highlight observations, relationships, priority items, and suggested focus areas. Not judgment — just intelligence.

●Structured JSON outputs
●Observations & relationships
●Priority assessment
●Suggested investigator focus areas

Design Principles

🔒

Forensic-Safe

Read-only processing. No data modification. Evidence integrity preserved.

📖

Fully Explainable

Every score and finding is tied to a traceable rule. No black box decisions.

⚖️

Legally Neutral

No inference of intent, guilt, or crime type. Observations only.

🎯

Deterministic

Same input always produces same output. Reproducible and auditable.

See real-world applications

Explore Use Cases