Car Crashes

Crash Detection Solution

Our high-impact crash model is built to detect crashes where we expect to see considerable damage to the car and potentially require emergency assistance. We capture crashes that happen at speeds above 35-40kph results with a peak acceleration of 5g or above.

In Crash Detection, we use an on-device ML model and context checks to automatically detect a crash during transport. The Crash Detection collects data of the crash like:

  • the exact time of the crash

  • the exact location of the crash

  • magnitude of the impact

  • speed before impact

  • delta-v of the impact

This crash data is then sent to the Sentiance cloud platform for further processing.

Our Crash Detection solution has a higher than 70% recall rate. Crashes are reported near real-time (maximum one minute after the collision).

Detailed Crash Reporting

In the Cloud, we calculate all of the features mentioned in the transports section (driving events before, during, and after the crash) and augment it with weather and traffic conditions at the time of impact for more extensive crash forensics. Crash reports can be used to foster faster claims processing, and reduce fraudulent claims and loss ratios.

Frequently Asked Questions (FAQs)

I experienced a bump while parking or while I was standing still near the intersection, why was it not detected?

Currently, we focus only on capturing crashes while you were moving at speeds above 35-40 kph and then coming to a complete stop.

How can I fake the crash on the real road to test the integration?

It is currently not possible to fake a crash. With every model improvement, it gets harder to trick it (simulate a crash). To fast-track the QA process and simulate the crashes during testing we expose a debug print flag. This functionality reveals the specific reasons why we did not report a crash event despite detecting a high-impact signal in the accelerometer data. For example: if you bump your phone on the table you should see that the transport mode is not a vehicle and the speed is too low.

I have got a false crash identification, why?

An average user has less than 1 out of 5 chances of getting an alert in 5 years. As with any alarm, it might go off once in a while, but this is a price you pay for having a high detection rate. Usually, false positives are easy to explain, for example, you dropped your phone after stopping, bumped it into the door while exiting the car, etc. We recommend first sending a notification to an end user and asking if they are OK and only if they do not respond escalate it further.

Last updated