Methods and systems for detecting crash events from live sensor data

A system comprising:

• one or more motion sensors configured to generate a stream of time-series data while the one or more motion sensors are disposed in a vehicle during a drive;
• and
• a computing device comprising:
• one or more processors;
• and
• a computer-readable medium comprising instructions that, when executed by the one or more processors, cause the one or more processors to:
• receive, in real-time, the stream of time-series data generated by the one or more motion sensors disposed in the vehicle during the drive;
• generate, for each contiguous interval within the stream of time-series data of a first predefined length, a crash prediction, wherein the crash prediction for a contiguous interval is indicative of whether the vehicle was involved in a crash during the contiguous interval, and generating the crash prediction for the contiguous interval comprises:
• executing a crash detection neural network on the time-series data generated by the one or more motion sensors during a time period that comprises the contiguous interval in response to receiving the time-series data generated during the time period, wherein the time period has a second predefined length that is greater than the first predefined length;
• and
• for each crash prediction:
• determine whether the crash prediction meets one or more predefined crash detection threshold criteria;
• and
• generate a crash analysis using the time-series data generated during the corresponding time period in response to determining that the crash prediction meets the one or more predefined crash detection threshold criteria.

A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of a computing device, cause the one or more processors to perform operations comprising:

• receiving, in real-time while the computing device is disposed in a vehicle during a drive, a stream of time-series data generated by one or more motion sensors disposed in the vehicle during the drive;
• generating, for each contiguous interval within the stream of time-series data of a first predefined length, a crash prediction, wherein the crash prediction for a contiguous interval is indicative of whether the vehicle was involved in a crash during the contiguous interval, and generating the crash prediction for the contiguous interval comprises:
• executing a crash detection neural network on the time-series data generated by the one or more motion sensors during a time period that comprises the contiguous interval in response to receiving the time-series data generated during the time period, wherein the time period has a second predefined length that is greater than the first predefined length;
• and
• for each crash prediction:
• determining whether the crash prediction meets one or more predefined crash detection threshold criteria;
• and
• generating a crash analysis using the time-series data generated during the corresponding time period in response to determining that the crash prediction meets the one or more predefined crash detection threshold criteria.