A hard brake is not a risk signal until you know what caused it.
Accelerometer telematics counts what happened to the vehicle. It cannot separate a driver who tailgates from a driver who reacted well to someone else's mistake — and those two carry opposite risk. wahanx recovers the cause from the forward camera.
Every event rate you have been quoted is a statement about a threshold.
Below is real driving from the comma1M dataset, scored live by the same detector that runs in our pipeline — ported to JavaScript and verified to reproduce the Python implementation to the digit. Drag the threshold and watch the rates move.
Move the threshold across the range every commercial programme would defend, and watch how far the same driving can be made to move.
The accelerometer records the reaction. The camera records the cause.
What an IMU sees
A deceleration of 4.2 m/s² lasting 1.3 seconds at 62 km/h. That is the entire record. Every mainstream usage-based programme is built on this signal, which is why it is commodity — a phone or a $30 dongle produces it.
The same trace comes from a driver following two car-lengths back at motorway speed, and from a driver with a safe gap reacting correctly to a child stepping into the road.
What the forward camera adds
Time headway at the moment of the event. Lane discipline in the seconds before it. Whether the lead vehicle's brake lights were already lit.
One driver is generating risk; the other is absorbing someone else's. Pricing them identically is the structural weakness in accelerometer-only scoring.
Kinematic event layer
Harsh braking, acceleration and cornering, debounced and exposure-normalised. Run end to end on real driving data.
See the measurements →Time headway
Continuous following distance from forward video — the signal an accelerometer provably cannot produce.
What we are building →Calibration
Fitting event weights to real collision outcomes. The one dependency engineering cannot remove.
Partner with us →