Method and system for vehicle speed estimation

1. A method comprising:

receiving, from a mobile device disposed within a vehicle, a set of sensor measurements collected from an accelerometer of the mobile device during a first time period;

converting the set of sensor measurements into a frequency domain using a Fourier transform;

filtering, using a digital bandpass filter implemented in hardware or software of the mobile device or a server, the set of sensor measurements to eliminate high frequency sensor measurements from the set of sensor measurements;

defining a set of contiguous windows based on a remaining sensor measurements in the set of sensor measurements, each contiguous window of the set of contiguous windows representing a contiguous portion of the remaining sensor measurements in the set of sensor measurements;

generating, for each contiguous window of the set of contiguous windows, a set of features by resampling the remaining sensor measurements of the contiguous window at one or more predefined frequencies;

generating an estimated speed of the vehicle during the first time period by executing a trained neural network using the set of features;

generating a vehicle crash prediction using the estimated speed; and

transmitting, to a remote device, the vehicle crash prediction.

2. The method of claim 1 , wherein generating the vehicle crash prediction using the estimated speed includes:

determining, based on the estimated speed, whether the vehicle stopped after a crash event.

3. The method of claim 1 , wherein at least two contiguous windows of the set of contiguous windows represent a different quantity of the remaining sensor measurements in the set of sensor measurements.

4. The method of claim 1 , wherein the trained neural network is a convolutional neural network with one or more dilated convolutions.

5. The method of claim 1 , wherein defining the set of contiguous windows includes:

deriving, for each acceleration signal in the set of sensor measurements, a time difference between when each pair of consecutive sensor measurements were collected; and

defining a first contiguous window of the set of contiguous windows as starting at a first time in which a first time difference between a first pair of consecutive sensor measurements exceeds a threshold and ending at a second time after which a second time difference between a second pair of consecutive sensor measurements exceeds the threshold, wherein the second time occurs after the first time.

6. The method of claim 5 , wherein the threshold is equal to one or more standard deviations from a mean time difference between each pair of consecutive sensor measurements.

7. The method of claim 1 , wherein generating the vehicle crash prediction using the estimated speed includes:

executing a crash prediction machine-learning model using the set of sensor measurements to generate a first vehicle crash prediction, the first vehicle crash prediction indicating a crash event occurred at a first time;

determining, at a time after the vehicle crash prediction, that the estimated speed is greater than zero;

determining, based on determining that the estimated speed of the vehicle is greater than zero, that the crash event did not occur; and

defining the vehicle crash prediction to indicate that the crash event did not occur.

8. The method of claim 1 , wherein generating the vehicle crash prediction using the estimated speed includes:

executing a crash prediction machine-learning model using the set of sensor measurements to generate a first vehicle crash prediction, the first vehicle crash prediction indicating a crash event occurred at a first time;

determining, at a time after the first vehicle crash prediction, that the vehicle stopped after the first time;

generating, using the estimated speed, a set of distance features, each distance feature of the set of distance features corresponding to a position of the vehicle relative to the crash event; and

defining the vehicle crash prediction to be an output from a second execution of the crash prediction machine-learning model using the set of sensor measurements and the set of distance features.

9. The method of claim 1 , wherein the set of sensor measurements include activity data.

10. A method comprising:

receiving, from a mobile device disposed within a vehicle, a set of sensor measurements collected from an accelerometer of the mobile device;

converting the set of sensor measurements into a frequency domain using a Fourier transform;

filtering, using a digital bandpass filter implemented in hardware or software of the mobile device or a server, the set of sensor measurements to eliminate high frequency sensor measurements from the set of sensor measurements;

generating a set of contiguous windows using remaining sensor measurements in the set of sensor measurements, wherein each contiguous window of the set of contiguous windows represent a subset of the remaining sensor measurements in the set of sensor measurements;

generating, for each contiguous window of the set of contiguous windows, an estimated speed of the vehicle during each contiguous window by executing a trained neural network using the subset of the remaining sensor measurements in the set of sensor measurements represented by each contiguous window of the set of contiguous windows;

receiving an indication that the vehicle is involved in a crash event;

authenticating the crash event using the estimated speed;

determining, based on authenticating the crash event, that the crash event did not occur; and

purging, in response to determining that the crash event did not occur, the crash event from memory.

11. The method of claim 10 , wherein at least two contiguous windows of the set of contiguous windows represent a different quantity of sensor measurements in the set of sensor measurements.

12. The method of claim 10 , wherein the trained neural network is a convolutional neural network with one or more dilated convolutions.

13. The method of claim 10 , wherein authenticating the crash event using the estimated speed includes:

determining whether the vehicle stopped after the crash event, wherein the crash event is determined not to have occurred upon determining that the vehicle did not stop.

14. The method of claim 10 , wherein authenticating the crash event using the estimated speed includes:

determining that the vehicle stopped at a distance from the crash event;

generating one or more distance features that represent the distance from the crash event;

executing a crash prediction machine-learning model using the set of sensor measurements and the distance features to generate a vehicle crash prediction; and

determining that the vehicle crash prediction is less than a threshold, wherein determining that the vehicle crash prediction is less than the threshold indicates that the crash event did not occur.

15. A mobile device comprising:

one or more processors;

an accelerometer; and

a non-transitory computer-readable medium storing instructions, that when executed by the one or more processors, cause the one or more processors to perform operations including

receiving, while the mobile device is disposed within a vehicle, a set of sensor measurements collected from the accelerometer of the mobile device during a first time period;

converting the set of sensor measurements into a frequency domain using a Fourier transform;

filtering, using a digital bandpass filter implemented in hardware or software of the mobile device, the set of sensor measurements to eliminate high frequency sensor measurements from the set of sensor measurements;

defining a set of contiguous windows based on a remaining sensor measurements in the set of sensor measurements, each contiguous window of the set of contiguous windows representing a contiguous portion of the remaining sensor measurements in the set of sensor measurements;

generating, for each contiguous window of the set of contiguous windows, a set of features by resampling the remaining sensor measurements of the contiguous window at one or more predefined frequencies;

generating an estimated speed of the vehicle during the first time period by executing a trained neural network using the set of features;

generating a vehicle crash prediction using the estimated speed; and

transmitting, to a remote device, the vehicle crash prediction.

16. The mobile device of claim 15 , wherein at least two contiguous windows of the set of contiguous windows represent a different quantity of the remaining sensor measurements in the set of sensor measurements.

17. The mobile device of claim 15 , wherein the trained neural network is a convolutional neural network with one or more dilated convolutions.

18. The mobile device of claim 15 , wherein defining the set of contiguous windows includes:

deriving, for each acceleration signal in the set of sensor measurements, a time difference between when each pair of consecutive sensor measurements were collected; and

defining a first contiguous window of the set of contiguous windows as starting at a first time in which a first time difference between a first pair of consecutive sensor measurements exceeds a threshold and ending at a second time after which a second time difference between a second pair of consecutive sensor measurements exceeds the threshold, wherein the second time occurs after the first time.

19. The mobile device of claim 15 , wherein generating the vehicle crash prediction using the estimated speed includes:

executing a crash prediction machine-learning model using the set of sensor measurements to generate a first vehicle crash prediction, the first vehicle crash prediction indicating a crash event occurred at a first time;

determining, at a time after the vehicle crash prediction, that the estimated speed is greater than zero;

determining, based on determining that the estimated speed of the vehicle is greater than zero, that the crash event did not occur; and

defining the vehicle crash prediction to indicate that the crash event did not occur.

20. The mobile device of claim 15 , wherein generating the vehicle crash prediction using the estimated speed includes:

executing a crash prediction machine-learning model using the set of sensor measurements to generate a first vehicle crash prediction, the first vehicle crash prediction indicating a crash event occurred at a first time;

determining, at a time after the first vehicle crash prediction, that the vehicle stopped after the first time;

generating, using the estimated speed, a set of distance features, each distance feature of the set of distance features corresponding to a position of the vehicle relative to the crash event; and

defining the vehicle crash prediction to be an output from a second execution of the crash prediction machine-learning model using the set of sensor measurements and the set of distance features.