Map-assisted target detection for sensor calibration
AssigneeGM Cruise Holdings LLC
InventorSyed Aziz Enam
Life of this patent
Estimating the term…
Abstract
System, methods, and computer-readable media for extrinsically calibrating sensors on an autonomous vehicle (AV) by leveraging map data. By leveraging offline data, such as the map data on the AV, the extrinsic calibration of the sensors may be performed on road to update or check previous calibrations. Static objects are determined in the scene by leveraging a map with a semantic map layer and sensors are extrinsically calibrating using a portion of a set of static points associated with the determined static objects by determining relative positions and angles the sensors.
Claims20
A method for calibrating a sensor using targets for calibration the sensor using data from a map, the method comprising:
- receiving, from one or more sensors of an autonomous vehicle (AV), sensor data made up of points reflected off of objects detected in an environment surrounding the AV while the AV is in motion;
- determining, at the AV, static objects in the environment by correlating the sensor data with objects represented in a map with a semantic map layer stored on the AV, the semantic map layer including labels providing semantic meaning to objects represented in the map;
- selecting a plurality of different relative positions and relative angles between a first sensor and a forward motion vector of the AV, and determining a number of static points for each relative position and relative angle, the number of static points being from a static object determined by the semantic map layer, wherein each static point is determined to be static based on ego-motion compensation;
- selecting a relative position and a relative angle from the plurality of different relative positions and relative angles, the selected relative position and the selected relative angle associated with a highest number of determined static points after the ego-motion compensation;
- and
- setting, at the AV, the first sensor to be at the selected relative position and the selected relative angle from the forward motion vector to extrinsically calibrate the first sensor.
A system for calibrating a sensor using targets for calibration the sensor using data from a map comprising:
- a storage configured to store instructions;
- and
- a processor configured to execute the instructions and cause the processor to:
- receive, from one or more sensors of an autonomous vehicle (AV), sensor data made up of points reflected off of objects detected in an environment surrounding the AV while the AV is in motion;
- determine, at the AV, static objects in the environment by correlating the sensor data with objects represented in a map with a semantic map layer stored on the AV, the semantic map layer including labels providing semantic meaning to objects represented in the map;
- select a plurality of different relative positions and relative angles between a first sensor and a forward motion vector of the AV, and determine a number of static points for each relative position and relative angle, the number of static points being from a static object determined by the semantic map layer, wherein each static point is determined to be static based on ego-motion compensation;
- select a relative position and a relative angle from the plurality of different relative positions and relative angles, the selected relative position and the selected relative angle associated with a highest number of determined static points after the ego-motion compensation;
- and
- set, at the AV, the first sensor to be at the selected relative position and the selected relative angle from the forward motion vector to extrinsically calibrate the first sensor.
A non-transitory computer-readable medium comprising instructions, the instructions, when executed by a computing system, cause the computing system to:
- receive, from one or more sensors of an autonomous vehicle (AV), sensor data made up of points reflected off of objects detected in an environment surrounding the AV while the AV is in motion;
- determine, at the AV, static objects in the environment by correlating the sensor data with objects represented in a map with a semantic map layer stored on the AV, the semantic map layer including labels providing semantic meaning to objects represented in the map;
- select a plurality of different relative positions and relative angles between a first sensor and a forward motion vector of the AV, and determine a number of static points for each relative position and relative angle, the number of static points being from a static object determined by the semantic map layer, wherein each static point is determined to be static based on ego-motion compensation;
- select a relative position and a relative angle from the plurality of different relative positions and relative angles, the selected relative position and the selected relative angle associated with a highest number of determined static points after the ego-motion compensation;
- and
- set, at the AV, the first sensor to be at the selected relative position and the selected relative angle from the forward motion vector to extrinsically calibrate the first sensor.
Dependencies are inferred from the claim text.