Systems and methods for traffic light detection
Techniques for detecting and labeling traffic lights and/or traffic signs and the traffic lanes with which such lights and signs are associated in an environment are disclosed. Images may be evaluated to identify pixels that may be associated with a light, sign, or lane. Associations between lights and/or signs and lanes in the environment may be determined along with probabilities for the individual pixels that the pixels may be associated with one of the light/sign and lane associations. Those pixels having a sufficient probability of be associated with a light/sign and lane association may be assigned a corresponding label. An output image with such labels can be provided for vehicle control and for other operations, such as top-down segmentation and trajectory determination.
1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing instructions executable by the one or more processors, wherein the instructions, when executed, cause the system to perform operations comprising:
receiving vision data associated with a vehicle traversing an environment;
determining, based at least in part on the vision data, a two-dimensional sensor perspective image representing a portion of the environment;
determining, by a first machine-learned (ML) model, based at least in part on the two-dimensional sensor perspective image, a first image comprising a traffic light and traffic lane association label for a first pixel and a second pixel of the two-dimensional sensor perspective image, wherein:
the first pixel represents at least a portion of a traffic light in the environment;
the second pixel represents at least a portion of a traffic lane in the environment; and
the traffic light and traffic lane association label indicates that the first pixel is associated with the second pixel;
determining, by a second ML model, based at least in part on mapping data associated with the environment, traffic lane features in the environment;
determining, based at least in part on determining that a first confidence score associated with the first ML model is greater than a second confidence score associated with the second ML model, that the first ML model is to be used for the environment; and
controlling the vehicle based at least in part on the traffic light and traffic lane association label.
2. The system of claim 1 , wherein determining that the first ML model is to be used for the environment comprises determining that the mapping data is inconsistent with the environment.
3. The system of claim 1 , wherein the operations further comprise:
performing a fusion operation using first output of the first ML model and second output of the second ML model to determine a fused output data structure; and
controlling the vehicle further based at least in part on the fused output data structure.
4. The system of claim 1 , wherein controlling the vehicle based at least in part on the traffic light and traffic lane association label comprises:
determining that the vehicle is in traveling in the traffic lane associated with the traffic light based at least in part on the traffic light and traffic lane association label;
determining a condition of the traffic light; and
controlling the vehicle based at least in part on determining that the vehicle is in traveling in the traffic lane associated with the traffic light and the condition of the traffic light.
5. The system of claim 1 , wherein first ML model is trained to determine that the traffic light is associated with the traffic lane based at least in part on:
determining a probability that the first pixel is associated with the second pixel;
determining that the probability meets or exceeds a probability threshold; and
determining, based at least in part on the determining that the probability meets or exceeds the probability threshold, the traffic light and traffic lane association label for the first pixel and the second pixel.
6. One or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving sensor data associated with a sensor configured at a vehicle traversing an environment;
determining, based at least in part on the sensor data, a data structure representing the sensor data associated with at least a portion of the environment;
executing a first machine-learned (ML) model configured to determine, based at least in part on the data structure, an output data structure comprising a traffic control object and drivable region association label for a first data element and a second data element of the data structure, wherein:
the first data element represents a traffic control object in the environment;
the second data element represents a drivable region in the environment; and
the traffic control object and drivable region association label indicates that the first data element is associated with the second data element;
executing a second ML model configured to determine, based at least in part on mapping data associated with the environment, traffic lane features in the environment;
determining, based at least in part on determining that a first confidence score associated with the first ML model is greater than a second confidence score associated with the second ML model, that the first ML model is to be used to control the vehicle in the environment; and
controlling the vehicle based at least in part on the output data structure.
7. The one or more non-transitory computer-readable media of claim 6 , wherein the first ML model is trained to perform traffic control object and drivable region association detection using labeled training data comprising one or more data elements labeled with the traffic control object and drivable region association label.
8. The one or more non-transitory computer-readable media of claim 6 , wherein controlling the vehicle based at least in part on the output data structure comprises:
executing the second ML model to determine, based on a second data structure associated with the environment, a second output data structure;
performing a fusion operation using the output data structure and the second output data structure to determine a fused output data structure; and
controlling the vehicle based at least in part on the fused output data structure.
9. The one or more non-transitory computer-readable media of claim 8 , wherein the second ML model is configured to determine the second output data structure based at least in part on the mapping data associated with the environment.
10. The one or more non-transitory computer-readable media of claim 6 , wherein the first ML model determines the output data structure comprising the traffic control object and drivable region association label for the first data element and the second data element based at least in part on a probability that the first data element is associated with the second data element.
11. The one or more non-transitory computer-readable media of claim 6 , wherein:
the traffic control object and drivable region association label is a first traffic control object and drivable region association label;
the traffic control object is a first traffic control object;
the output data structure comprises a second traffic control object and drivable region association label for a third data element and a fourth data element of the data structure;
the third data element represents a second traffic control object in the environment that is distinct from the first traffic control object;
the fourth data element represents a second drivable region in the environment that is distinct from the drivable region;
the second traffic control object and drivable region association label indicates that the third data element is associated with the fourth data element; and
the second traffic control object and drivable region association label is distinct from the first traffic control object and drivable region association label.
12. The one or more non-transitory computer-readable media of claim 11 , wherein controlling the vehicle comprises:
determining that the vehicle is traveling in one of the drivable region or the second drivable region; and
determining one of the first traffic control object or the second traffic control object based at least in part on determining that the vehicle is traveling in one of the drivable region or the second drivable region,
wherein controlling the vehicle is further based at least in part on one of the first traffic control object or the second traffic control object.
13. The one or more non-transitory computer-readable media of claim 6 , wherein the operations further comprise executing the first ML model to determine, based at least in part on the data structure, a second label for the first data element, wherein the output data structure further comprises the second label, wherein the second label is indicative of a classification of the first data element.
14. The one or more non-transitory computer-readable media of claim 6 , wherein controlling the vehicle comprises providing the output data structure to a trajectory determination system configured to determine a vehicle trajectory based at least in part on the output data structure.
15. A method comprising:
receiving sensor data associated with a sensor configured at a vehicle traversing an environment;
determining, based at least in part on the sensor data, a data structure representing the sensor data associated with a portion of the environment;
determining, by a first machine-learned (ML) model, based at least in part on the data structure, an output data structure comprising a traffic control object and drivable region association label for a first data element and a second data element of the data structure, wherein:
the first data element represents a traffic control object in the environment;
the second data element represents a drivable region in the environment; and
the traffic control object and drivable region association label indicates that the first data element is associated with the second data element;
determining, by a second ML model, based at least in part on mapping data associated with the environment, traffic lane features in the environment;
determining, based at least in part on determining that a first confidence score associated with the first ML model is greater than a second confidence score associated with the second ML model, that the first ML model is to be used to control the vehicle in the environment; and
providing the output data structure to a vehicle control component.
16. The method of claim 15 , wherein the first ML model determines the output data structure further based at least in part on a first label for the first data element that is distinct from the traffic control object and drivable region association label and a second label for the second data element that is distinct from the traffic control object and drivable region association label.
17. The method of claim 16 , wherein determining the traffic control object and drivable region association label for the first data element is further based at least in part on determining that a probability that the first data element is associated with the second data element meets or exceeds a threshold.
18. The method of claim 15 , wherein providing the output data structure to the vehicle control component comprises:
using the output data structure to determine a top-down representation of the environment; and
providing the top-down representation of the environment to the vehicle control component.
19. The method of claim 15 , wherein the data structure comprises one or more of:
a lidar point cloud;
a two-dimensional image; or
a three-dimensional representation of the environment.
20. The method of claim 15 , wherein the traffic control object is one of:
a traffic sign;
a traffic light; or
a human traffic controller.