Object corner detection
Techniques for determining detection boxes representing objects in an environment using improved corner detections are disclosed herein. Autonomous vehicle sensors can capture data in an environment that may include separate objects, such as large and/or articulated vehicles. In an example, the data can include a plurality of points that may be processed to identify object corners. Locations of points in clusters of points associated with particular object corners may be aggregated to determine a predicted corner point that may be used to adjust the detection box for that object, thereby generating a more accurate detection box for the object. A vehicle computing system can control the vehicle using the adjusted detection box.
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 sensor data associated with a vehicle traversing an environment;
determining a predicted corner point location by:
determining based at least in part on the sensor data, a plurality of corner points associated with a corner in the environment and associated with a corner point classification; and
determining, based at least in part on locations of the plurality of corner points, the predicted corner point location;
determining, based at least in part on the sensor data and independently from determining the predicted corner point location, a detection box representing an object associated with an object classification distinct from the corner point classification;
modifying, based at least in part on the predicted corner point location, at least one extent parameter of the detection box by associating a corner point of the detection box with the predicted corner point location to determine an adjusted detection box representing the object; and
controlling the vehicle based at least in part on the adjusted detection box.
2 . The system of claim 1 , wherein determining the predicted corner point location comprises averaging location parameters associated with the plurality of corner points to determine a location parameter of the predicted corner point location.
3 . The system of claim 1 , wherein determining the plurality of corner points comprises:
determining, based at least in part on the sensor data, a second plurality of corner points;
determining corner location parameters for individual corner points of the second plurality of corner points; and
determining, as the plurality of corner points, a subset of the second plurality of corner points having a same corner location parameter.
4 . The system of claim 3 , wherein the same corner location parameter comprises one of:
a vertical corner location parameter,
a horizontal corner location parameter, or
a height corner location parameter.
5 . The system of claim 1 , wherein modifying the at least one extent parameter of the detection box to determine the adjusted detection box comprises modifying a first location parameter representing a location a corner of the detection box to represent a second location parameter of the predicted corner point location.
6 . The system of claim 1 , wherein determining the plurality of corner points comprises:
determining, based at least in part on the sensor data, a second plurality of corner points;
determining that the second plurality of corner points are interior corner points associated with an articulated vehicle object type; and
excluding the second plurality of corner points.
7 . A method comprising:
receiving sensor data associated with a vehicle traversing an environment;
determining, based at least in part on the sensor data and a plurality of corner points determined in the environment, a predicted corner point;
determining, based at least in part on the sensor data and independently from determining the predicted corner point, a detection box representing an object in the environment;
modifying, based at least in part on the predicted corner point, the detection box; and
controlling the vehicle based at least in part on the detection box.
8 . The method of claim 7 , further comprising:
determining, based at least in part on the sensor data, a second predicted corner point;
modifying, based at least in part on the second predicted corner point, a second detection box representing the object; and
controlling the vehicle further based at least in part on the second detection box.
9 . The method of claim 7 , wherein determining the predicted corner point is further based at least in part on determining that a corner point classification confidence value for the predicted corner point meets or exceeds a threshold.
10 . The method of claim 7 , wherein determining the predicted corner point comprises determining that the predicted corner point is associated with one or more exterior corner points associated with an articulated vehicle object type.
11 . The method of claim 7 , wherein determining the predicted corner point comprises:
determining the plurality of corner points are associated with a corner point classification;
determining, based at least in part on the sensor data, a classification of an object associated with the predicted corner point, wherein the classification of the object is distinct from the corner point classification; and
modifying, further based at least in part on the classification of the object, the detection box.
12 . The method of claim 7 , wherein controlling the vehicle comprises:
determining a predicted trajectory for the object based at least in part on the detection box; and
controlling the vehicle based at least in part on the predicted trajectory.
13 . The method of claim 7 , wherein controlling the vehicle comprises:
determining a trajectory for the vehicle based at least in part on the detection box; and
controlling the vehicle using the trajectory.
14 . The method of claim 7 , wherein:
determining the predicted corner point based at least in part on the sensor data comprises determining the predicted corner point based at least in part on a first subset of the sensor data associated with a first sensor type; and
determining the detection box based at least in part on the sensor data comprises determining the detection box based at least in part on a second subset of the sensor data associated with a second sensor type that is distinct from the first sensor type.
15 . One or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, perform operations comprising:
receiving environmental data associated with a vehicle traversing an environment;
determining, based at least in part on the environmental data and a plurality of corner points determined in the environment, a predicted corner point;
determining, based at least in part on the environmental data and independently from determining the predicted corner point, a detection box representing an object in the environment;
refining, based at least in part on the predicted corner point, the detection box; and
determining a vehicle control based at least in part on the detection box.
16 . The one or more non-transitory computer-readable media of claim 15 , wherein determining the predicted corner point comprises:
determining that the plurality of corner points are associated with a same relative corner location parameter; and
determining the predicted corner point from among the plurality of corner points.
17 . The one or more non-transitory computer-readable media of claim 16 , wherein the same relative corner location parameter comprises one of:
a vertical corner location parameter,
a horizontal corner location parameter, or
a height corner location parameter.
18 . The one or more non-transitory computer-readable media of claim 15 , wherein refining the detection box comprises modifying, based at least in part on the predicted corner point, a first extent of the detection box.
19 . The one or more non-transitory computer-readable media of claim 18 , wherein determining the detection box comprises maintaining a second extent of the detection box.
20 . The one or more non-transitory computer-readable media of claim 15 , wherein determining the vehicle control comprises:
determining a top-down image of the environment comprising the detection box; and
determining the vehicle control based at least in part on the top-down image.