AI vision
Recognises five pedestrian groups on the crossing: children, pregnant women, the elderly, wheelchair users, and general pedestrians.
- Camera
- NVIDIA Jetson edge computer
JAPATI at a Glance
JAPATI is a crossing signal that can see. A camera and AI recognise children, pregnant women, the elderly, and wheelchair users, then give them enough time to cross. A radar watches oncoming traffic and sounds a warning when a vehicle is going too fast.
Developed by lecturers and students of Universitas Pendidikan Indonesia, JAPATI has completed field testing at Technology Readiness Level (TRL) 7 on the UPI Cibiru campus.
Background
The same crossing time is not enough for everyone.
Most pedestrian signals use a fixed duration. Slower pedestrians are forced to hurry, and vehicles speeding towards the crossing get no warning at all.
01Fixed timing
The walk phase is the same for everyone, regardless of road width or who is crossing.
02Vulnerable road users
Children, pregnant women, the elderly, and wheelchair users need more time and often cross anxiously.
03No early warning
Drivers who do not slow down at a red light are noticed only when it is too late.
How It Works
Four steps, from the button press to the other side of the road.
- Press the button
A pedestrian presses the touch button on the pole to request a crossing.
- AI recognises
The camera watches the zebra crossing and the AI recognises the pedestrian group in real time.
- Time adapts
Duration is calculated from road width and the walking speed of that group. The LED countdown can only increase, never get cut short.
- Traffic is watched
While the light is red, the radar monitors vehicles. If one exceeds the safe speed, a warning siren sounds.
Components
One pole, four subsystems working together.
Tap a number on the pole to see what it does.
Drag to rotate, pinch or scroll to zoom.
The solar panel in the illustration and 3D model is a planned upgrade.
Road sensors
The radar tracks the speed of approaching vehicles. The laser measures road width during installation as the basis for timing.
- TSR20 radar
- LDS30DA laser
Signals
The light stops traffic, the LED matrix shows the countdown, and the speaker gives voice guidance and the warning siren.
- Traffic light
- LED matrix
- TOA speaker
Interaction
Pedestrians request a crossing. The camera and AI run only on request, which saves power.
- Touch button
Test Results
Figures from the TRL 7 Field Test Report. Each figure is labelled by how it was tested.
- Model validation85.19%Detection accuracy across 5 pedestrian groupsmAP@0.5
- Lab test18 msDetection time per frame50.86 FPS on average
- Field test0.375%Road width measurement error8 m, daytime · 0.625% at night
- Field test< 0.5 sSiren response after a violationtarget < 1 second
- Field test> 40 km/hSpeeding vehicles detected by radarat 50 m distance
- Field test+6 sAutomatic extra crossing timesimulated elderly pedestrian
Field test on 28 August 2026, UPI Cibiru campus.
The JAPATI Journey
From the lab to the open road.
- 2023–2024Design
Technology architecture, sensor integration, and adaptive AI model development.
- 2025–2026Validation
Vulnerable road user recognition and the early warning system tested on real roads.
- 28 Aug 2026TRL 7 field test
UPI Cibiru campus, Bandung.
View location - 2027Commercialisation target
Road-worthiness certification, production, and partnership with government.
Next development
Energy independenceA planned ±300 Wp solar panel with battery storage, so the pole can run without grid power.