Systems and methods for identifying high-risk driving situations from driving data

Patent metadata

Field	Value
Publication number	US-12097845-B2
Application number	US-202016950500-A
Country	US
Kind code	B2
Filing date	Nov 17, 2020
Priority date	Oct 28, 2020
Publication date	Sep 24, 2024
Grant date	Sep 24, 2024

Abstract

A method for identifying high-risk driving situations in driving data may include receiving driving data, identifying trigger events in the driving data, and using the trigger events in combination with first portions of the driving data corresponding to periods of time preceding the trigger events, training a machine learning model to identify features, in the driving data, associated with the trigger events. The method may further include using second portions of the driving data where no trigger events are identified, further training the machine learning model to ignore features common to both the first portions and the second portions. The method further includes detecting a high risk driving situation in real time driving data by using the machine learning model to identify the features associated with the trigger events, and performing a defensive driving maneuver in response to the detection of the high risk driving situation.

First claim

What is claimed is:

1. A method for identifying high-risk driving situations in driving data, the method comprising:
identifying trigger events in driving data;
using the trigger events in combination with first portions of the driving data corresponding to periods of time preceding the trigger events, training a machine learning model to identify features, in the driving data, associated with the trigger events;
when a high risk driving situation occurs:
detecting the high risk driving situation in real time driving data by using the machine learning model to identify the features associated with the trigger events;
performing a defensive driving maneuver in response to the detection of the high risk driving situation;
further training the machine learning model, using the identified features associated with the trigger events, to identify a secondary trigger event;
wherein secondary trigger event comprises features of sensor data which follow the high risk driving situation and were not initially identified as the trigger events, and the secondary trigger event is predicted by the features that are predictive of the trigger events; and
using the secondary trigger event as training data to further train the machine learning model to refine the identification of the features, in the driving data, associated with the trigger events; and

generating a prediction of the high risk driving situation in real time driving data by using the machine learning model to identify the features associated with the secondary trigger event, wherein at least one trigger event includes hard braking and the secondary trigger event includes turning on hazard lights.

2. The method of claim 1 , further comprising:
using second portions of the driving data where no trigger events are identified, further training the machine learning model to ignore features common to both the first portions and the second portions.

3. The method of claim 1 , wherein the driving data comprises the sensor data, and wherein the identifying of the trigger events comprises analyzing the sensor data to identify one or more of air bag deployment, the hard braking, fast acceleration, swerving, close proximity to obstacles, or collision.

4. The method of claim 1 , wherein the driving data comprises one or more of vehicle position information, vehicle speed information, or vehicle trajectory information of one or more vehicles.

5. The method of claim 1 , wherein the defensive maneuver comprises a maneuver that reduces risk of a current driving situation,
wherein the risk comprises a probability that a driving situation will result in the trigger event.

6. The method of claim 1 , wherein the identifying of the trigger events in the driving data comprises:
receiving a trigger event definition of the trigger event, defining one or more thresholds of the driving data; and
identifying the trigger event based on the trigger event definition including the sensor data exceeding the one or more thresholds.

7. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform the steps of:
identifying trigger events in driving data;
using the trigger events in combination with first portions of the driving data corresponding to periods of time preceding the trigger events, training a machine learning model to identify features, in the driving data, associated with the trigger events;
when a high risk driving situation occurs:
detecting the high risk driving situation in real time driving data by using the machine learning model to identify the features associated with the trigger events;
performing a defensive driving maneuver in response to the detection of the high risk driving situation;
further training the machine learning model, using the identified features associated with the trigger events, to identify a secondary trigger event;
wherein secondary trigger event comprises features of sensor data which follow the high risk driving situation and were not initially identified as the trigger events, and the secondary trigger event is predicted by the features that are predictive of the trigger events; and
using the secondary trigger event as training data to further train the machine learning model to refine the identification of the features, in the driving data, associated with the trigger events; and

generating a prediction of the high risk driving situation in real time driving data by using the machine learning model to identify the features associated with the secondary trigger event, wherein at least one trigger event includes hard braking and the secondary trigger event includes turning on hazard lights.

8. The non-transitory computer-readable medium of claim 7 , further comprising:
using second portions of the driving data where no trigger events are identified, further training the machine learning model to ignore features common to both the first portions and the second portions.

9. The non-transitory computer-readable medium of claim 7 , wherein the driving data comprises the sensor data, and wherein the identifying of the trigger events comprises analyzing the sensor data to identify one or more of air bag deployment, the hard braking, fast acceleration, swerving, close proximity to obstacles, or collision.

10. The non-transitory computer-readable medium of claim 7 , wherein the driving data comprises one or more of vehicle position information, vehicle speed information, or vehicle trajectory information of one or more vehicles.

11. The non-transitory computer-readable medium of claim 7 , wherein the defensive maneuver comprises a maneuver that reduces risk of a current driving situation, and
wherein the risk comprises a probability that a driving situation will result in the trigger event.

12. The non-transitory computer-readable medium of claim 7 , wherein the identifying of the trigger events in the driving data comprises:
receiving a trigger event definition of the trigger event, defining one or more thresholds of the driving data; and
identifying the trigger event based on the trigger event definition including the sensor data exceeding the one or more thresholds.

13. A system for detecting high risk driving situations, the system comprising:
one or more sensors configured to generate driving data;
one or more processors and a memory storing a machine learning model,
wherein the machine learning model is trained to identify features associated with trigger events in first portions of the driving data corresponding to periods of time preceding the trigger events;
wherein the machine learning model is further configured to detect a high risk driving situation in real-time driving data when the high risk driving situation occurs by identifying the features associated with the trigger events in the real-time driving data;
wherein the machine learning model is further trained to identify a secondary trigger event that comprises features of sensor data which follow the high risk driving situation and were not initially identified as the trigger events, and the secondary trigger event is predicted by the features that are predictive of the trigger events;
wherein the secondary trigger event is used as training data to further train the machine learning model to refine the identification of the features, in the driving data, associated with the trigger events;
wherein the secondary trigger event is further used to generate a prediction of the high risk driving situation in real time driving data by using the machine learning model to identify the features associated with the secondary trigger event;

Assignees

• Toyota Res Inst Inc

Inventors

• Marcotte Ryan
• Hammond Marcus Mccoy

Classifications

• B60W2540/30

  Driving style · CPC title

• B60W30/095

  Predicting travel path or likelihood of collision · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• B60W40/09

  Driving style or behaviour · CPC title

• B60W30/09 Primary

  Taking automatic action to avoid collision, e.g. braking and steering · CPC title