Applied Vision AI Gröbenzell · Munich

Vision AI that thinks on the device.

We build object detection, body tracking and mapping AI that runs in real time on phones, wearables and IoT devices. No cloud. Proven in DeepSwing, the launch monitor in your iPhone.

real-time object detection
240fps
carry deviation from radar
0–2%
golf courses captured by our course AI
40,057
Live · on-device

Fig. 01 — Swing plane and club head trail, detected live on the iPhone. Real footage from the range.

01 — Thesis

An iPhone carries more computing power in your pocket than a 600-dollar radar launch monitor. We turn it into a measuring instrument.

02 — Live from the pipeline

From the first frame to ball flight.

  1. 01 / 240 fps

    Every frame is a measurement.

    The camera films at 240 frames per second with 1/8000 s exposure. Sharp enough to freeze a golf ball in flight.

  2. 02 / Armed

    Ball detected. Armed.

    The model finds the ball on the mat, locks it and waits for the strike. Full precision is spent only when it counts.

  3. 03 / 145 km/h

    Club head acquired.

    Our own YOLO model detects the club head frame by frame, milliseconds before impact, and measures its speed.

  4. 04 / 186 km/h · 13.3°

    Ball flight computed.

    A few frames after impact give ball speed, launch angle and direction, and from those the entire flight.

Fig. 02 — Real footage with the DeepSwing vision HUD: high-speed camera looking down at the mat.240 FPS · 1/8000 s

04 — DeepSwing Launch Monitor

The launch monitor is now an app.

No radar. No sensor. An iPhone on a tripod measures every shot and shows it on the phone, the iPad, the Mac or big on a projector.

Fig. 04 — Screen recording of the app, 7-iron on the range.
Carry
123m
Ball speed
159km/h
Launch angle
15.8°
Direction
2.6° L
  • Ball speed
  • Club speed
  • Launch
  • Direction
  • Carry
  • Total
  • 12,000+ courses
  • Daily challenge
  • Up to 4 players
  • Live duels
  • Projector & TV
  • GSPro
  • Putting

0–2%

carry deviation from a radar launch monitor (Rapsodo MLM2 PRO), depending on the club.We estimate spin. And we say so.

05 — Our course AI

We taught an AI to read golf courses.

Our own segmentation model draws fairways, greens, tees, bunkers and water onto aerial images. A second one assembles the holes. Here it runs on a course it never saw in training.

  • Fairway
  • Green
  • Tee
  • Bunker
  • Water
of greens found
99%
precision on greens
96%
green centre error (median)
0.6m
of bunkers found
94%

Fig. 05 — Golf-Club Bad Pyrmont, validation course. Aerial image: Geobasis NRW (dl-de/zero-2-0). Segmentation: Deep Platforms course AI.

40,057

golf courses captured worldwide

10,229

of them GPS-ready

512

regions processed

06 — Golf course GPS

12,000+ courses. Free on iPhone.

What the course AI maps becomes GPS in DeepSwing: distances to green and hazards, plays-like with elevation, green heatmaps and a scorecard with handicap. On Apple Watch too, even without iPhone.

Hole 1, par 4: 307 m to the centre, plays like 303 m
Fig. 06.1 — Hole 1, par 4: 307 m to the centre, plays like 303 m
Green heatmap: slope and fall line
Fig. 06.2 — Green heatmap: slope and fall line
Set the flag on the green, get the distance instantly
Fig. 06.3 — Set the flag on the green, get the distance instantly
  • 01Front, centre and back
  • 02Hazards with reach and carry
  • 03Plays-like with club suggestion
  • 04Green heatmaps
  • 05Scorecard and handicap
  • 06Apple Watch and Dynamic Island

07 — Cartography

Terrain in centimetres.

Laser-scan elevation models, 1-metre grid, 0.01 metres off the official model. From it come contour lines, green slopes and the distance a shot really plays.

Golf Oberstaufen-Steibis, 18 holes between 860 and 1,041 m. Contours every 4 m, computed from DGM1 laser-scan data.
Fig. 07 — Golf Oberstaufen-Steibis, 18 holes between 860 and 1,041 m. Contours every 4 m, computed from DGM1 laser-scan data.

Hole 4 · Par 5

Golf Oberstaufen-Steibis

on the map
458m
plays like
419m

26 m downhill, 926 m above sea level: thinner air, longer flight.

08 — Smart data fusion

One shot. Seven data sources.

The connection is what makes the number valuable. We combine what camera, map, terrain and physics know.

  • 01DistanceGPS position to the front, centre and back of the green.
  • 02ElevationHeight profile every 5 m along the hole from LiDAR.
  • 03Air densityAltitude, about 1% more distance per 290 m.
  • 04WindDirection and strength in the ball flight model.
  • 05Ball flightSpeed, launch and direction from the launch monitor.
  • 06SwingPhases and checks from body tracking.
  • 07GreenSlope and fall line on a 1-m grid.

09 — Ball flight in 3D

Fast camera moves. Real terrain.

Every shot is flown again in 3D. The camera chases the ball across courses whose terrain comes from real elevation data, from the Swiss Alps to the Sonoran Desert.

11 — Why Deep Platforms

Measure more. Carry less.

FeatureDeep PlatformsRadar hardwareTypical golf apps
Works without an extra device Yes No, device from about $600 Yes
Launch monitor and swing analysis in one app Yes No Rarely
Own object detection for club and ball Yes Radar, not camera Rarely
Body tracking with 30 checks Yes No Partly
Courses on real LiDAR terrain Yes Sometimes, paid extra Rarely
Own course AI from aerial imagery Yes No No
On-device analysis, no upload Yes Partly Rarely

Comparison based on publicly available product information, as of October 2026.

12 — For partners

Your vision AI. Built by us.

What we built for golf works wherever cameras need to recognise something in real time: sports, health, industry, agriculture, smart home.

  • 01Sports & fitness
  • 02Movement & physio
  • 03Industrial inspection
  • 04Agriculture & geodata
  • 05Smart home & IoT
  • 06Wearables

Contact

What should your camera recognise?

Object detection, body tracking, segmentation or cartography, on your device, in real time. Let's talk about it.