Metrics for modeling lane closure zones
Systems and methods for determining metrics for lane closure zone detection and evaluating the performance of lane closure zone detection. A lane closure zone can include any area where a lane is closed off. The changes to the typical travel paths for vehicles on the roadway can indicate that vehicles are to travel along a different path than that indicated on the maps the AVs use to operate. Thus, it is important for accurate driving operations of the AVs for the AVs to be able to interpret the changes to the typical travel paths indicated by the lane closure. Navigating through lane closure zones generally includes mapping (by the AV) objects in the lane closure zone, including the objects indicating the lane closure, and understanding related surrounding non-player characters (e.g., other vehicles).
1 . A method for evaluating lane closure zone detection, comprising:
obtaining map data for an area around a lane closure zone, wherein the map data includes travel lane information;
receiving a labeled polygon indicating the lane closure zone;
generating a ground truth lane state for the lane closure zone based on the labeled polygon;
receiving a set of captured data frames of the lane closure zone from a sensor suite comprising a computer vision system, localization sensors, and driving sensors;
identifying at least one lane closure object in each of the set of captured data frames using a perception module of an onboard computer, and wherein the perception module fuses data from one or more interior sensors with data from exterior sensors;
generating a predicted lane closure polygon for the lane closure zone based on the at least one lane closure object and the travel lane information using a zone modeling module of the onboard computer;
generating a predicted lane state for the lane closure zone based on the predicted lane closure polygon using the zone modeling module of the onboard computer;
determining a difference between the ground truth lane state and the predicted lane state; and
generating a performance metric based on the difference.
2 . The method of claim 1 , further comprising classifying the at least one lane closure object for temporary traffic restriction attributes using the perception module of the onboard computer.
3 . The method of claim 1 , wherein generating the predicted lane state includes identifying map elements and associating the predicted lane closure polygon and the at least one lane closure object with the map elements.
4 . The method of claim 1 , wherein the at least one lane closure object includes a plurality of lane closure objects, and wherein generating a predicted lane closure polygon further comprises generating predicted edges of the predicted lane closure polygon based on the plurality of lane closure objects.
5 . The method of claim 1 , wherein receiving a set of captured data frames includes receiving, for each frame in the set of captured data frames, image data, LIDAR data, and RADAR data.
6 . The method of claim 1 , wherein generating the performance metric includes:
determining a first intersection over union value for the labeled polygon and the lane closure zone,
matching the labeled polygon with a first lane in the lane closure zone based on the first intersection over union value,
determining a second intersection over union value for the predicted lane closure polygon, and
matching the predicted lane closure polygon with the first lane in the lane closure zone based on the second intersection over union value.
7 . The method of claim 1 , wherein the ground truth lane state is a first ground truth lane state for a closed lane, wherein the predicted lane state is a first predicted lane state for the closed lane, and further comprising generating a second ground truth lane state for an open lane and generating a second predicted lane state for the open lane.
8 . The method of claim 1 , wherein the lane closure object is an emergency vehicle, and wherein generating the predicted lane state includes determining that the emergency vehicle is stopped and blocking a traffic lane, identifying map elements, and associating the predicted lane closure polygon and the at least one lane closure object with the map elements.
9 . The method of claim 1 , wherein the at least one lane closure object is one of a construction object, a temporary sign, caution tape, and a flare, and wherein generating the predicted lane state includes determining that the at least one lane closure object is blocking a traffic lane, identifying map elements, and associating the predicted lane closure polygon and the at least one lane closure object with the map elements.
10 . A system for evaluating lane closure zone detection, comprising:
a vehicle including:
an onboard computer configured to obtain map data for an area around a lane closure zone, wherein the map data includes travel lane information,
a sensor suite configured to capture sensor data including a set of captured data frames of a lane closure zone, and
a perception module of the onboard computer configured to identify at least one lane closure object in each of the set of captured data frames, and configured to fuse data from one or more interior sensors with data from exterior sensors,
a zone modeling module of the onboard computer configured to generate a predicted lane closure polygon for the lane closure zone based on the at least one lane closure object and travel lane information, and generate a predicted lane state for the lane closure zone based on the predicted lane closure polygon, and
wherein the onboard computer is further configured to transmit the set of captured data frames and the generated predicted lane states to a central computer; and
the central computer configured to:
obtain map data for an area around a lane closure zone, wherein the map data includes travel lane information;
receive a labeled polygon indicating the lane closure zone;
generate a ground truth lane state for the lane closure zone based on the labeled polygon;
receive the set of captured data frames of the lane closure zone;
determine a difference between the ground truth lane state and the predicted lane state; and
generate a performance metric based on the difference.
11 . The system of claim 10 , wherein the perception module is further configured to classify the at least one lane closure object for temporary traffic restriction attributes.
12 . The system of claim 10 , wherein the zone modeling module is configured to generate the predicted lane state by identifying map elements and associating the predicted lane closure polygon and the at least one lane closure object with the map elements.
13 . The system of claim 10 , wherein the at least one lane closure object includes a plurality of lane closure objects, and wherein the zone modeling module is configured to generate the predicted lane closure polygon by generating predicted edges of the predicted lane closure polygon based on the plurality of lane closure objects.
14 . The system of claim 10 , wherein the set of captured data frames includes, for each frame in the set of captured data frames, image data, LIDAR data, and RADAR data.
15 . The system of claim 10 , wherein the ground truth lane state is a first ground truth lane state for a closed lane, wherein the predicted lane state is a first predicted lane state for the closed lane, and
wherein zone modeling module is further configured to generate a second predicted lane state for an open lane, and
wherein the zone modeling module is further configured to generate a second ground truth lane state for the open lane.
16 . The system of claim 10 , wherein the lane closure object is an emergency vehicle, and wherein the perception module is configured to determine that the emergency vehicle is stopped and blocking a traffic lane, identify map elements, associate the predicted lane closure polygon and the at least one lane closure object with the map elements to generate the predicted lane state.
17 . A computer system, comprising:
a computer processor for executing computer program instructions; and
one or more non-transitory computer-readable media storing computer program instructions executable by the computer processor to perform operations comprising:
obtaining map data for an area around a lane closure zone, wherein the map data includes travel lane information;
receiving a labeled polygon indicating the lane closure zone;
generating a ground truth lane state for the lane closure zone based on the labeled polygon;
receiving a set of captured data frames of the lane closure zone from a sensor suite comprising a computer vision system, localization sensors, and driving sensors;
identifying at least one lane closure object in each of the set of captured data frames using a perception module of an onboard computer, and wherein the perception module fuses data from one or more interior sensors with data from exterior sensors;
generating a predicted lane closure polygon for the lane closure zone based on the at least one lane closure object and the travel lane information using a zone modeling module of the onboard computer;
generating a predicted lane state for the lane closure zone based on the predicted lane closure polygon using a zone modeling module of the onboard computer;
determining a difference between the ground truth lane state and the predicted lane state; and
generating a performance metric based on the difference.
18 . The computer system of claim 17 , wherein the instructions further comprise classifying the at least one lane closure object for temporary traffic restriction attributes using the perception module of the onboard computer.
19 . The computer system of claim 17 , wherein generating the predicted lane state includes identifying map elements and associating the predicted lane closure polygon and the at least one lane closure object with the map elements.
20 . The computer system of claim 17 , wherein the ground truth lane state is a first ground truth lane state for a closed lane, wherein the predicted lane state is a first predicted lane state for the closed lane, and wherein the instructions further comprise generating a second ground truth lane state for an open lane and generating a second predicted lane state for the open lane.