HK Lab
EN
한국어English简体中文日本語
Sign in Sign up
02 / PROJECTS

Built with the field

Projects we have developed under contract with companies and institutions and deployed in the field. Most recent first.

2026
MANUFACTURING · FPC2026.05

FPC AI vision inspection · part tracking system

In collaboration with SAMKEE Corp. · CM Insight

We added a smartphone inspector and a BLE conveyor to an FPC inspection line, handling defect judgment, part-code reading and history logging in a single pass.

Highlights
  • Real-time YOLO26 detection on the smartphone NPU — part condition judged per model
  • FPC code recognition via QR · Data Matrix reading, then upload to the server
  • Conveyor slowdown · stop via ESP32 BLE, automatic alignment at the inspection position
  • Originals stored on the in-house NAS + cloud search web, assisting electrode microscope inspection
ONNX · QNNYOLO26ML KitESP32 BLENASBlazor
Patent applications
10-2026-0081586
MANUFACTURING · FPC

FPC AI vision inspection · part tracking system

SYSTEM System Configuration

FPC AI conveyor system diagram — smartphone, conveyor, ESP32 motor driver
Smartphone inspector · conveyor · ESP32 BLE motor control
AcuVis app — detection → code reading → server upload

FLOW Inspection Flow

  1. 01
    Feeding · decelerationConveyor · BLE control
  3. 02
    AI detectionYOLO26 · NPU
  5. 03
    Code readingQR · Data Matrix
  7. 04
    History storage · searchNAS originals · cloud web

DATA Data Structure

  • Heavy originals stay only on the in-house NAS, and only thumbnails and metadata go to the cloud — searches are fast, and originals load only when clicked.
  • If the internet goes down, the NAS queues uploads and resends them later.
  • Search by FPC code, inspection date/time and result, and download result images in bulk.
SMART FARM · WSN2026.05In operation

Poultry farm AIoT environment monitoring

In collaboration with DIT · CM Insight

We built a system that ties together temperature, humidity, CO₂, ammonia and light sensors and six cameras in the poultry houses, so the farm can be watched from anywhere and alerts go out as soon as something goes wrong.

Highlights
  • Analog, ammonia and wireless logger sensors — Modbus collection, InfluxDB storage
  • 6 IP cameras — WebRTC/HLS live view · 24-hour RAID recording
  • Floor-plan 2D live map · 3D view, sensor graphs · video/CSV download
  • Threshold alarms → HK Push app notifications, role-based external viewing permissions
ModbusInfluxDBMediaMTXBlazorDocker
Related technology
2D/3D monitoring
Patent applications
10-2026-0096389
SMART FARM · WSN

Poultry farm AIoT environment monitoring

SCREENS Monitoring screens

2D live map — cameras and sensor values on the poultry house floor plan
2D live map — camera and sensor positions and current values on the floor plan
Sensor graphs and camera video
Sensor trend graphs · camera video

SENSORS Data collection setup

EnvironmentCO₂ · Light · Temp/Humidity
GasNH₃ Ammonia
Wireless loggers6units
Cameras6units · 24h
  • ADAM modules (4–20mA) and ammonia sensors (RS485 → Modbus TCP gateway) are read every 5 seconds and stored in InfluxDB.
  • Wireless loggers in each zone add temperature and humidity readings, and cameras record 24/7 to RAID.
  • Runs in containers so it recovers on its own after a reboot. Remote users see different screens depending on their role.
RVOS monitoring — dashboard, live map, multilingual broadcasting (1/3)
RVOS monitoring — dashboard, live map, multilingual broadcasting (2/3)
RVOS monitoring — dashboard, live map, multilingual broadcasting (3/3)
SMART FIELD · SAFETY2026.09 ~In progress

RVOS 5G integrated monitoring platform

Geonam Co., Ltd.

We are building a monitoring platform that lets operators view, broadcast to and handle alarms from wireless broadcast and alarm devices (RVOS) spread over many construction sites, all on one screen.

Highlights
  • Dashboard · map-based live map · device status and output control
  • Real-time PTT · TTS broadcasting in 12 languages · sirens · scheduled broadcasts
  • Camera multi-view · alarm rules · inspection reports, 4 permission levels
  • Device controller based on Raspberry Pi CM5 · 5G/Wi-Fi failover design
.NET 9 BlazorLeafletMQTTTTSRaspberry Pi CM5
Anyang University — Dept. of Information, Electrical and Electronic Engineering home, track roadmap, AICL lab home (1/3)
Anyang University — Dept. of Information, Electrical and Electronic Engineering home, track roadmap, AICL lab home (2/3)
Anyang University — Dept. of Information, Electrical and Electronic Engineering home, track roadmap, AICL lab home (3/3)
WEB PLATFORM · AI2026.10In operation

Anyang University department · lab websites

Anyang University Dept. of Information, Electrical and Electronic Engineering · AI Communications Lab

We built the websites of the Department of Information, Electrical and Electronic Engineering and the AI Communications Lab (AICL), running on the university's server with a regulations AI chatbot and AI-driven site editing.

Highlights
  • Department overview · faculty · courses · Information/Electrical/Electronic track roadmap (prerequisite arrows) · board
  • An AI chatbot that automatically collects the university's 239 regulations every week and answers with citations
  • Automatic editing · preview · deployment, where an AI agent makes the changes an admin asks for in plain words
  • Main-page slides · pop-up notices · visitor count · app install (PWA) · push notifications
  • AICL lab website — research · paper DB (automatic Crossref lookup) · board
BlazorMariaDBDockerRAGAI AgentPWA
Dept. of Information, Electrical and Electronic Engineering ↗AICL lab ↗
Integrated portal app prototype — home · notices · AI counseling, TF collaboration platform on desktop — shared calendar · notices, TF collaboration platform on phone — status board · files · AI assistant (1/3)
Integrated portal app prototype — home · notices · AI counseling, TF collaboration platform on desktop — shared calendar · notices, TF collaboration platform on phone — status board · files · AI assistant (2/3)
Integrated portal app prototype — home · notices · AI counseling, TF collaboration platform on desktop — shared calendar · notices, TF collaboration platform on phone — status board · files · AI assistant (3/3)
WEB · MOBILE · RESPONSIVE2026.09 ~ 10

HKNU integrated portal · AI-centered university TF collaboration platform

Hankyong National University

We built, as responsive web apps, an integrated portal prototype that brings student ID, timetable, notices and AI counseling into one app, and a collaboration platform where the AI-centered university project TF manages schedules, drafts and files together.

Highlights
  • Integrated portal prototype — mobile student ID · timetable · notices · academic affairs · AI counseling in one app
  • Status board — owner · draft · page count for each section of the business plan at a glance
  • Files — browse folders like a file explorer; uploads are filed in the right place automatically
  • AI assistant — answers from documents · draft status · page-count checks · document review · file tidying
  • Responsive web app (PWA) — the same screens adapt, with a left menu on PC and bottom tabs on phones
BlazorPWAResponsive WebAI Agent
Open the prototype ↗
Physical AI semantic communication concept, USRP B210 demo — 4-camera hub superposition, video to two terminals (original · semantic · digital) (1/3)
Physical AI semantic communication concept, USRP B210 demo — 4-camera hub superposition, video to two terminals (original · semantic · digital) (2/3)
Physical AI semantic communication concept, USRP B210 demo — 4-camera hub superposition, video to two terminals (original · semantic · digital) (3/3)
R&D SERVICE · SEMANTIC2026.08 ~In progress

Semantic communication contract research

Chung-Ang University Industry-Academic Cooperation Foundation

We design and validate AI-based multi-user semantic communication technology with the Intelligent Networks Lab at Chung-Ang University.

Highlights
  • Multi-user semantic multiplexing based on extended shared embeddings
  • Embedding Division Multiple Access based on semantic relevance (EDMA)
  • Performance validated with text · image embeddings, 2 patents filed jointly
AI-NativeSemantic Comm.Transformer
Related technology
Semantic Communication
Patent applications
10-2026-008793210-2026-0087933
2025
MANUFACTURING · BATTERY PARTS2025.04

AI vision inspector · AI vision dimension inspector

Yesun Tech Co., Ltd.

We developed and delivered a defect inspector (AI Detector) and a non-contact dimension gauge (AI Ruler) for a secondary-battery component line, operated on the HawkAI platform.

Highlights
  • AI Detector — judges missing holes and color differences, accuracy ≥ 99% · Takt ≤ 400ms
  • AI Ruler — straight length and curve radius ±0.15mm, dual 50/200MP modes
  • Two configurations — Jetson edge and lightweight smartphone — with pass/fail routing guidance
  • HawkAI — labeling · augmentation · training · OTA deployment · statistics
YOLOFaster R-CNNJetson Orin NanoGalaxy S25 Ultra
Patent applications
10-2026-008158610-2026-0082439
MANUFACTURING · BATTERY PARTS

AI vision inspector · AI vision dimension inspector

01 / DETECTOR AI Detector — real-time AI vision defect inspection

Type I — Jetson Orin Nano + touch monitor edge device
TYPE I · EDGE DEVICE Jetson Orin Nano + touch monitor Instant on-line inference on a 40 TOPS edge device. 5MP webcam + 15.8" touch UI.
Type II — smartphone + stand inspection setup
TYPE II · MOBILE Lightweight smartphone setup (app + stand) Deploy on any line instantly with just a stand. Models download automatically from the server.
Inspection result: pass — green box, move-right guidance
PASS Pass — green box → move right After automatic pass/fail judgment, the operator is immediately shown which way to route the part.
Inspection result: defective — red box, move-left guidance
FAIL Defective — red box → move left Instant alarm on misrouting. Accuracy ≥ 99%, Takt time ≤ 400ms.
Accuracy ≥ 99%
Takt Time ≤ 400ms
Criteria Holes & color
Pass/fail handling Route branching
TYPE I · EDGE DEVICE

Edge Device + Webcam

NVIDIA Jetson Orin Nano (40 TOPS) + 5MP webcam + 15.8" touch monitor. Ubuntu 22.04 LTS + JetPack 6.0. Enclosure W370 × D310 × H480 mm.

  • 6-core ARM CPU + 1024-core Ampere GPU
  • 8GB / 4GB LPDDR5, MicroSD 256GB
  • QHD 5MP auto exposure / white balance
TYPE II · MOBILE

Smartphone + stand

The same inspection flow, slimmed down to a smartphone app. With just a stand it can be deployed on any line immediately.

  • Same inspection flow as Type I
  • Model download / automatic updates from the server
  • Temporary inspection station set up beside the conveyor line

Inspection result UX

PASS Pass Green box → move right
FAIL Defective Red box → move left
ALARM Misrouting Instant alarm on wrong direction

02 / RULER AI Ruler — non-contact high-precision dimension measurement

Jig + Backlight + Vacuum Chuck + monitor
JIG Backlight + Vacuum Chuck Work surface 470×250 mm², LED 15–25W adjustable illumination, vacuum-suction flat fixing.
Monitor + smartphone alignment screen
PREVIEW Monitor + smartphone alignment Automatic outline (Blob) alignment + fine adjustment within ±1.5° gets the part ready for measurement.
Dimension measurement result — 50MP / 200MP re-measurement
RESULT Dimension measurement result (PASS / FAIL) 50MP mode for speed, 200MP mode for precision. Re-measure instantly on one screen.
Measurement error ± 0.15mm
FOV 420×200mm²
Inspection cycle ≤ 4s
Resolution 50/200MP
JIG

Backlight + Vacuum Chuck

Work surface 470 × 250 mm², LED 15–25W adjustable illumination, vacuum surface for flat fixing. Enclosure W555 × D355 × H355 mm.

  • Automatic outline (Blob) alignment of the measured part
  • Precise alignment within ±1.5°
  • Galaxy S25 Ultra 50MP / 200MP rear camera
FLOW

Measure → judge → save

Select model → align preview → visual inspection → automatic external-camera capture → dimension measurement → save / upload results. Stops automatically after 60 seconds of inactivity.

  • 50MP mode: speed / stability first
  • 200MP mode: high-resolution precision measurement
  • Automatic EXIF + JSON history logging
2024
ROBOT · IMMERSIVE CONTROL2024.08

Immersive robot monitoring based on semantic communication

ELUON Corp.

Multiple cameras and edge devices collaborate to estimate human 3D pose in real time, replayed in 3D on the web for robot monitoring.

Highlights
  • Multi-view camera footage generated in a Unity virtual test environment
  • End-device 2D pose estimation → 3D reconstruction by edge-server triangulation
  • Real-time 3D playback in a WebGL client, with latency · error (MPJPE) display
UnityYOLOv8-PoseWebGL.NET
Paper arXiv:2504.03052 ↗GitHub ↗
2023
ICT Robotics Engineering and AIoT Lab websites (1/2)
ICT Robotics Engineering and AIoT Lab websites (2/2)
WEB PLATFORM2023 ~In operation

Department · lab web systems

Hankyong National University

We develop and operate the ICT Robotics Engineering website with its capstone design management system, and the AIoT Lab website with its paper management system.

Highlights
  • Capstone design project · graduation project management, membership permission requests
  • Management of papers · patents · research · development history, photos · board
  • Automatic website editing · deployment, where an AI agent makes the changes requested in the admin screen (2026)
  • An AI chatbot that answers on university rules · campus guides · graduation projects with sources (2026)
  • Lab camera · temperature · humidity · light monitoring
BlazorMariaDBDockerNginx
ictrobot.hknu.ac.kr ↗aiot.hknu.ac.kr ↗
2022
AI CCTV safety monitoring — multi-device edge inference screen
SAFETY · AI CCTV2022.03

Joint development of an AI CCTV platform

VTouch Inc.

We deploy AI models to CCTV cameras and compute boxes to detect safety issues such as fire, smoke, falls and headcount right on site.

Highlights
  • Fire · smoke detection — SL factory
  • Fall detection — geriatric hospital, Daejeon
  • Face · headcount detection — O-World, Daejeon
  • Crowd object detection — Navy headquarters, Pyeongtaek
Real-timeEdge AIYOLOAlarm automation
2021
Training data management — datasets and label lists per model (1/2)
Training data management — datasets and label lists per model (2/2)
DATA · TRAINING SET2021.05

Building AI deep learning training data

SmartLabs Inc.

We collected, cleaned and labeled video data for training vision AI models and built it into a training dataset.

LabelingDatasetData Cleansing

Preparing a similar project?
Tell us about your site conditions and we will propose an approach.

Project inquiry
An unexpected error occurred. Reload 🗙
Lost connection to the server. Reconnecting…