Systems and methods for near-crash determination

1. A method comprising:

capturing an external video with an external-facing camera of an on-board system mounted on a vehicle;

determining an obstacle parameter for an obstacle detected from the external video;

capturing an internal video with a driver-facing camera of the on-board system;

generating a risk score for a spatial position relative to the vehicle, the risk score being determined based on a parametric model, the obstacle parameter, and a driver parameter based on the internal video; and

detecting a near-collision event when the risk score exceeds a risk threshold.

2. The method of claim 1 further comprising:

identifying an independent parameter of the parametric model with a highest weighted value as a cause of the near-collision event.

3. The method of claim 1 further comprising:

transmitting information indicative of the near-collision event to a computing system that is remote from the vehicle.

4. The method of claim 1 further comprising:

determining multiple obstacle parameters for multiple obstacles detected from the external video, wherein a risk score is calculated for each obstacle.

5. The method of claim 1 , wherein the cause of the near-collision event is a driver cause as determined based on the driver parameter.

6. The method of claim 1 further comprising:

transmitting a portion of the external video and a portion of the internal video to a computing system at a location remote from the vehicle.

7. The method of claim 1 further comprising:

determining a driver response to the near-collision event based on the internal video.

8. The method of claim 1 , wherein the parametric model includes a Gaussian model.

9. A method comprising:

obtaining an indication of a sampled video captured by an external-facing camera of a vehicle, the indication of the sampled video including an obstacle parameter for an obstacle detected based on the sampled video;

obtaining an indication of a near-collision event labeled with a driver parameter, the near-collision event being detected based on a spatial risk map generated based on a parametric model, the obstacle parameter, and the driver parameter; and

determining a cause of the near-collision event by identifying an independent parameter of the parametric model with a highest weighted value.

10. The method of claim 9 , wherein the cause of the near-collision event is determined at a computing system remote from the vehicle.

11. The method of claim 9 , wherein a processing system of the vehicle detects and labels the near-collision event.

12. The method of claim 9 further comprising:

obtaining an indication of a secondary obstacle parameter for the obstacle detected based on the sampled video, the spatial risk map being further generated based on the secondary obstacle parameter.

13. The method of claim 9 , wherein the driver parameter includes a driver response determined based on a sensor signal obtained during the near-collision event.

14. The method of claim 13 , wherein the sensor signal includes an internal image sampled by an internal-facing camera of the vehicle.

15. The method of claim 9 further comprising:

aggregating driver responses to near-collision events for multiple drivers;

identifying good drivers associated with driver scores that exceed a threshold;

identifying the driver responses of the good drivers; and

training a vehicle control system based on the driver responses of the good drivers.

16. The method of claim 15 further comprising:

determining a respective driver score based on a history of near-collision events associated with each driver.

17. The method of claim 9 further comprising:

providing feedback to the driver based on a history of near-collision events associated with the driver.

18. The method of claim 9 , wherein a risk score is calculated using the parametric model, the driver parameter, and the obstacle parameter.

19. A method comprising:

capturing a first image with a first camera facing away from a driver of a vehicle;

determining an obstacle parameter based on the first image;

determining a driver parameter associated with the driver, the driver parameter being based on a second image captured with a second camera facing the driver of the vehicle;

determining a risk score based on a parametric model, the obstacle parameter, and the driver parameter;

detecting a near-collision event based on the risk score; and

determining a driver response to the near-collision event based on a third image captured with the second camera facing the driver of the vehicle.

20. The method of claim 19 further comprising:

determining a driver score based on a history of near-collision events associated with the driver of the vehicle.