HumanSense AI - Face ID Solution

👤

High-Precision Face Recognition

512-dimensional embedding vectors with cosine similarity matching. Supports unlimited face registration with multi-condition profiles.

512D Embedding 106 Landmarks

🛡️

Liveness Challenge

딥페이크 대응을 위한 "라이브니스 챌린지 기능 도입"으로 인쇄, 동영상 대응에 안정적인 결과를 제공합니다.

Liveness Challenge feature for deepfake prevention, providing stable results against print and video attacks.

Deepfake Prevention Anti-Spoofing

😊

Emotion Recognition

Real-time detection of 7 emotions (Happy, Sad, Angry, Surprised, Fearful, Disgusted, Neutral) with attention-based neural networks.

7 Emotions Attention Model

📊

Age & Gender Estimation

Accurate demographic analysis with age group classification (child to senior) and gender prediction using multi-model ensemble.

6 Age Groups SSR-Net

🏃

Pose & Behavior Analysis

17-keypoint pose estimation with behavior pattern recognition. Detects pacing, fidgeting, and unusual postures for anxiety scoring.

17 Keypoints Anxiety Score

⚡

Edge-Optimized Performance

Optimized for NVIDIA Jetson platforms with CUDA acceleration. Achieves 30+ FPS with full analysis pipeline on embedded devices.

Jetson Ready 30+ FPS

Component	Technology	Specification
Face Detection Engine	FT-FaceDetect™	106-point facial landmark detection, multi-scale pyramid processing
Face Recognition	FT-FaceEmbed™	512-dimensional deep embedding, cosine similarity matching (threshold: 0.4)
Anti-Spoofing System	FT-LiveGuard™	11-layer verification pipeline, dual-network ensemble architecture
Person Detection	FT-PoseNet™	17-point skeletal keypoint detection, ~26M parameter neural network
Multi-Person Tracking	FT-TrackPro™	Appearance + motion hybrid tracking, up to 50 simultaneous subjects

Component

Technology

Specification

Face Detection Engine

FT-FaceDetect™

106-point facial landmark detection, multi-scale pyramid processing

Face Recognition

FT-FaceEmbed™

512-dimensional deep embedding, cosine similarity matching (threshold: 0.4)

Anti-Spoofing System

FT-LiveGuard™

11-layer verification pipeline, dual-network ensemble architecture

Person Detection

FT-PoseNet™

17-point skeletal keypoint detection, ~26M parameter neural network

Multi-Person Tracking

FT-TrackPro™

Appearance + motion hybrid tracking, up to 50 simultaneous subjects

Metric	Value
Processing Speed	30+ FPS (full pipeline)
Face Detection Accuracy	99.2% @ FAR 1e-6
Recognition Accuracy	99.5% (1:1 verification)
Liveness Detection Rate	98.7% (photo/video attack rejection)
Supported Resolution	Up to 4K (3840×2160)
Latency	< 50ms (detection to recognition)

Metric

Value

Processing Speed

30+ FPS (full pipeline)

Face Detection Accuracy

99.2% @ FAR 1e-6

Recognition Accuracy

99.5% (1:1 verification)

Liveness Detection Rate

98.7% (photo/video attack rejection)

Supported Resolution

Up to 4K (3840×2160)

Latency

< 50ms (detection to recognition)

Layer	Method	Attack Type Defended
1	Texture Analysis (LBP)	Printed photo
2	Frequency Domain Analysis (DCT)	Screen display
3	Color Distribution Analysis	Low-quality replay
4	Reflection Pattern Detection	Glossy print/screen
5	Skin Detail Analysis	Mask attack
6	Edge Sharpness Analysis	Digital manipulation
7	Moiré Pattern Detection	Screen capture
8	Challenge-Response (Head Direction)	Video replay
9	Temporal Consistency Check	Frame injection
10	Deep Neural Network Classifier (V1)	Advanced spoofing
11	Deep Neural Network Classifier (V2)	Deepfake attack

Layer

Method

Attack Type Defended

Texture Analysis (LBP)

Printed photo

Frequency Domain Analysis (DCT)

Screen display

Color Distribution Analysis

Low-quality replay

Reflection Pattern Detection

Glossy print/screen

Skin Detail Analysis

Mask attack

Edge Sharpness Analysis

Digital manipulation

Moiré Pattern Detection

Screen capture

Challenge-Response (Head Direction)

Video replay

Temporal Consistency Check

Frame injection

Deep Neural Network Classifier (V1)

Advanced spoofing

Deep Neural Network Classifier (V2)

Deepfake attack

Type	Protocol	Features
REST API	HTTP/HTTPS	Face registration, recognition, management
WebSocket	WS/WSS	Real-time video streaming, live analysis
Response Format	JSON	Structured metadata with confidence scores

Type

Protocol

Features

REST API

HTTP/HTTPS

Face registration, recognition, management

WebSocket

WS/WSS

Real-time video streaming, live analysis

Response Format

JSON

Structured metadata with confidence scores

Platform	Optimization	Performance
NVIDIA Jetson AGX Orin	TensorRT, CUDA 12.x	30+ FPS
NVIDIA Jetson Orin NX	TensorRT, CUDA 12.x	25+ FPS
x86 with NVIDIA GPU	CUDA acceleration	60+ FPS
CPU-only mode	ONNX Runtime	5-10 FPS

Platform

Optimization

Performance

NVIDIA Jetson AGX Orin

TensorRT, CUDA 12.x

30+ FPS

NVIDIA Jetson Orin NX

TensorRT, CUDA 12.x

25+ FPS

x86 with NVIDIA GPU

CUDA acceleration

60+ FPS

CPU-only mode

ONNX Runtime

5-10 FPS

HumanSense AI Platform

HumanSense AI Processing Pipeline

Real-time Analysis Flow

Solution Interface

Main Dashboard

Face Registration

Face Scanning

Liveness Challenge

Unknown Detection

Recognition Success

Key Features

High-Precision Face Recognition

Liveness Challenge

Emotion Recognition

Age & Gender Estimation

Pose & Behavior Analysis

Edge-Optimized Performance

FunnyTech Face ID System

Core Technology Stack

Performance Metrics

FT-LiveGuard™ Anti-Spoofing Layers

API Interface

Hardware Optimization

Security Features

Ready to Integrate HumanSense AI?