
See the small. Never lose it.
HANSUNG TS connects what the drone sees to what operations decide.
Finding people in wide aerial frames
Switch between the source footage and the detection results.
Detects a person in the water from footage over a river.
From camera to operations, one pipeline
The same pipeline deploys on NVIDIA, Intel and Qualcomm hardware.
- 01Capture
High-resolution drone video input (1080p·4K)
- 02Detect
Detects small objects
- 03Track
Continuous on-device tracking
Tracking continues without a connection
- 04Event
Structures detections into events
- 05Operations
Delivers events to the operations system
GPU handles decode to encode; INT8 model inference
Scene summaries arrive when a server link is available. Detection and tracking continue on the edge device.
Supported runtime environments
NVIDIA DeepStream / TensorRT (Jetson)
Intel OpenVINO Runtime (CPU / GPU / NPU)
Qualcomm AI Runtime (QAIRT/QNN, CPU / GPU / HTP)
Choose the runtime that fits your hardware.
Scenes turned into scene summaries
A video understanding model summarises the scene in one sentence.
Target, location, activity and status refresh every 2.5 seconds.
Scene summaries arrive when a server link is available. Detection and tracking continue on the edge device.
Summary
- Target
- Location
- Activity
- Status
- Timestamp
Scene summary refresh every 2.5 s
Applications
The wider the site, the smaller the target.

Defense and coastal perimeter
Along a long perimeter, approaching targets shrink to a few pixels.
hi-bee detects and tracks approaching targets and alerts the operations system.

Water and missing‑person search
On crowded water, one person in distress must be singled out.
hi-bee marks only the person in distress and keeps tracking them.

Port operations
At a large port, patrol footage piles up faster than teams can review it.
hi-bee surfaces only the event segments, each with a scene summary.
Evidence and sources
Every number here carries its measurement condition and its source. First, the size of the targets hi-bee actually works with.

The 1,920 × 1,080 frame is drawn at 1 : 6, as 320 × 180The 1,920 × 1,080 frame is drawn at 1 : 4, as 480 × 270The 1,920 × 1,080 frame is drawn at 1 : 2.5, as 768 × 432One tick = 192 px in the source
360px²
Median target area
0.017%
Share of the 1,920 × 1,080 frame area
9px
Minimum target height
MeasuredBased on 4,610 detections across three KISA certification test clips
- INT8 inference at the edge
- Real-time 1080p · 4K input
- Detection and tracking keep running offline
- Zero-copy GPU pipeline
- Multiple targets are tracked at once, each with its own ID.
Capabilities without a measured value are not typeset as numbers. Platform performance is cited from public benchmarks with source, device and precision.
The scale of the problem

Water and sea
104people
Water-activity deaths over five years (2021–2025)
Source: Ministry of the Interior and Safety, press release, 2026-07-27
26,394people
People rescued at sea in 2025 · 72 not rescued

Mountain
31,330responses
Mountain-accident rescue responses over three years (2022–2024)
These are published public statistics, not figures HANSUNG TS contributed to.
Certifications
- Certificate No.
- KISA-IC-2025-045
- Product version
- hi-bee V1.0
- Scope
- General Field / MissingPerson RGB detection
- Valid through
- 2025-09-25 – 2028-09-24
- Certificate No.
- KISA-IC-2025-046
- Product version
- hi-bee V1.1
- Scope
- General Field / DrowningPerson RGB detection
- Valid through
- 2025-09-25 – 2028-09-24
Partner and pilot institutions
Talk to HANSUNG TS
Tell us about your operating environment and video inputs. We will scope the deployment and hardware for your site.
Phone
Address
409–410, Gasan YPP Arsen Tower, 24 Gasan digital 2‑ro, Geumcheon‑gu, Seoul, Republic of Korea





