Assertive vehicle detection model generation

Patent metadata

Field	Value
Publication number	US-11433892-B2
Application number	US-201916700825-A
Country	US
Kind code	B2
Filing date	Dec 2, 2019
Priority date	Dec 2, 2019
Publication date	Sep 6, 2022
Grant date	Sep 6, 2022

Abstract

There is disclosed herein examples of system and procedure for generating a model to be implemented within autonomous vehicles for identifying assertive vehicles at an intersection having stop indicators. The model can be generated via data-driven procedure where captures of the movement of vehicles through an intersection are analyzed and utilized for generating the model. The model may be provided to the autonomous vehicles and may be implemented by the autonomous vehicles for identifying assertive vehicles.

First claim

The invention claimed is:

1. A computer-implemented method, comprising:
collecting, by a plurality of autonomous vehicles, data representing movement of vehicles within an interest zone, wherein the interest zone corresponds to an intersection having stop indicators in at least two directions;
identifying, by a processing device of an analysis system, a first portion of the data associated with one or more identified assertive vehicles and a second portion of the data associated with one or more identified non-assertive vehicles;
generating a plurality of different models, wherein each model discriminates between assertive vehicles and non-assertive vehicles based on the first portion of the data and the second portion of the data, and the different models include at least two of the following: a logistic regression model, a support vector machine model, a decision tree model, a random forest model, and a feedforward neural network model;
selecting one of the different models based on precision-recall scores assigned to each model;
implementing the selected model on an autonomous vehicle to discriminate assertive and non-assertive vehicles; and
taking an action, by the autonomous vehicle, based on results of the selected model.

2. The method of claim 1 , wherein the data includes one or more of the following: a distance to a corresponding stop line, a distance to an interior zone corresponding to the intersection, a speed, an acceleration, an indication of a corresponding lead vehicle, an indication of a corresponding right of way order, and a difference in arrival time from a corresponding vehicle having a right of way for each of the vehicles.

3. The method of claim 1 , further comprising identifying the one or more identified assertive vehicles and the one or more identified non-assertive vehicles based on a distance to an interior zone corresponding to the intersection and a right of way order corresponding to the vehicles for each of the vehicles.

4. The method of claim 3 , further comprising:
identifying, based on the data, a portion of the vehicles that entered the interest zone prior to or within a threshold time of corresponding autonomous vehicles that captured the data; and
filtering the data representing the movement of the vehicles to remove a portion of the data associated with the portion of the vehicles, wherein the filtered data is utilized for identifying the one or more identified assertive vehicles and the one or more identified non-assertive vehicles.

5. The method of claim 3 , wherein identifying the one or more identified assertive vehicles and the one or more identified non-assertive vehicles includes:
determining right of way values for each of the vehicles;
determining that each of the vehicles has crossed a corresponding threshold position;
identifying the one or more identified assertive vehicles from the vehicles based on the one or more identified assertive vehicles crossing the corresponding threshold position prior to times corresponding to the right of way values for each of the one or more identified assertive vehicles; and
identifying the one or more identified non-assertive vehicles from the vehicles based on the one or more identified non-assertive vehicles crossing the corresponding threshold position at a time corresponding to the right of way values for each of the one or more identified non-assertive vehicles.

6. The method of claim 5 , wherein the threshold position for each of the vehicles comprises a position where a path of a vehicle of the vehicles crosses with a path of another vehicle of the vehicles.

7. The method of claim 1 , wherein the interest zone is a first interest zone, wherein the intersection is a first intersection, and wherein the method further comprises:
analyzing a plurality of scenarios of vehicle movement within a second interest zone using each of the plurality of different models, wherein the second interest zone includes a second intersection having stop indicators in at least two directions, and wherein each of the plurality of different models attempt to identify assertive vehicles and non-assertive vehicles within the plurality of scenarios; and
assigning the precision-recall scores to the different models respectively based on whether each of the plurality of different models correctly identifies the assertive vehicles and the non-assertive vehicles within the plurality of scenarios.

8. The method of claim 1 , wherein the precision-recall scores are each generated based on a combination of a weighted precision score and a weighted recall score.

9. The method of claim 1 , wherein the interest zone includes an intersection zone and a buffer zone, wherein the intersection zone extends among stop locations of the intersection and covers the intersection, and wherein the buffer zone extends outward from the intersection zone by a distance.

10. One or more computer-readable media having instructions stored thereon, wherein the instructions, in response to execution by one or more devices, cause the one or more devices to:
retrieve data representing movement of vehicles within an interest zone, wherein the interest zone corresponds to an intersection having stop indicators in at least two directions, and the data is collected by a plurality of autonomous vehicles that encountered the interest zone;
determine that each of the vehicles has crossed a corresponding threshold position based on the data, wherein the threshold position for each of the vehicles comprises a position where a path of a vehicle of the vehicles crosses with a path of another vehicle of the vehicles;
determine a right of way order of the vehicles based on the data;
determine whether each of the vehicles that crossed the corresponding threshold position corresponds to the right of way order of the vehicles;
label each of the vehicles as either an assertive vehicle or a non-assertive vehicle, wherein the assertive vehicle does not correspond to the right of way order and the non- assertive vehicle corresponds to the right of way order;
identify a first portion of the data associated with one or more labeled assertive vehicles and a second portion of the data associated with one or more labeled non-assertive vehicles;
generate a plurality of different models, wherein each model discriminates between assertive vehicles and non-assertive vehicles based on the first portion of the data and the second portion of the data;
select one of the different models based on performance scores assigned to each model;
cause the selected model to be implemented on an autonomous vehicle to discriminate assertive and non-assertive vehicles; and
cause the autonomous vehicle to take an action based on results of the selected model.

11. The one or more computer-readable media of claim 10 , wherein to label each of the vehicles as either an assertive vehicle or a non-assertive vehicle includes to:
determine, for each of the vehicles that are labeled as an assertive vehicle, a first time when each of the vehicles had a vehicle speed below a threshold speed or a distance to an intersection zone of the interest zone was greater than a threshold distance, wherein the first time is prior to a second time when each of the vehicles is determined to have crossed the corresponding threshold position; and
label a time interval from the first time to the second time as an asserting time interval for each of the vehicles.

12. The one or more computer-readable media of claim 10 , wherein the instructions further cause the one or more devices to:
identify a third portion of the data associated with irrelevant vehicles; and
filter the third portion of th

Assignees

• Gm Cruise Holdings Llc

Inventors

• Ghafouri Amin
• Palli Prasanth
• Allan Jeremy

Classifications

• G08G1/166 Primary

  for active traffic, e.g. moving vehicles, pedestrians, bikes · CPC title

• G06V20/58

  Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads · CPC title

• G06V20/56

  exterior to a vehicle by using sensors mounted on the vehicle · CPC title

• B60W30/18154 Primary

  Approaching an intersection · CPC title

• G06F18/25

  Fusion techniques · CPC title