Method and system for classification of an object in a point cloud data set
A method for classifying an object in a point cloud includes computing first and second classification statistics for one or more points in the point cloud. Closest matches are determined between the first and second classification statistics and a respective one of a set of first and second classification statistics corresponding to a set of N classes of a respective first and second classifier, to estimate the object is in a respective first and second class. If the first class does not correspond to the second class, a closest fit is performed between the point cloud and model point clouds for only the first and second classes of a third classifier. The object is assigned to the first or second class, based on the closest fit within near real time of receiving the 3D point cloud. A device is operated based on the assigned object class.
1. A method, comprising:
retrieving, by one or more processors, a three dimensional (3D) point cloud representing an object, the 3D point cloud comprising a plurality of point cloud data points;
determining, by the one or more processors, at least one feature variable based on at least one point cloud data point of the 3D point cloud;
determining, by the one or more processors, a first classification statistic based on the at least one feature variable;
determining, by the one or more processors, a second classification statistic based on the at least one feature variable;
assigning, by the one or more processors to the object, a selected object class from a plurality of object classes by determining a closest match between the object and the selected object class using the first classification statistic and the second classification statistic; and
generating, by the one or more processors, a control signal to control operation of an autonomous vehicle based on the selected object class.
2. The method of claim 1 , wherein assigning the selected object class comprises:
determining, by the one or more processors, a first candidate closest match between a first object class of the plurality of object classes and the object using the first classification statistic;
determining, by the one or more processors, a second candidate closest match between a second object class of the plurality of object classes and the object using the second classification statistic; and
assigning, by the one or more processors to the object, the first object class as the selected object class responsive to the first object class being the same as the second object class.
3. The method of claim 2 , wherein assigning the selected object class to the object comprises:
determining, by the one or more processors responsive to the first object class not being the same as the second object class, a closest fit between the 3D point cloud and one of a first model point cloud of the first object class or a second model point cloud of the second object class; and
assigning, by the one or more processors to the object, one of the first object class or the second object class as the selected object class based on the one of the first object class or the second object class corresponding to the closest fit.
4. The method of claim 3 , wherein at least one of the first model point cloud or the second model point cloud is occluded such that the at least one of the first model point cloud or the second model point cloud represents a first portion of a model object corresponding to the at least one of the first model point cloud or the second model point cloud and does not represent a second portion of the model object different than the first portion.
5. The method of claim 1 , wherein the object is unclassified prior to assigning the selected object class to the object.
6. The method of claim 1 , wherein determining the first classification statistic comprises determining, by the one or more processors, the first classification statistic based on a histogram of at least a subset of the at least one point cloud data point in each of a plurality of bins corresponding to a range of values of the at least one feature variable.
7. The method of claim 1 , wherein determining the second classification statistic comprises determining, by the one or more processors, the second classification statistic based on a covariance matrix based on the at least one feature variable.
8. The method of claim 1 , wherein a number N of the plurality of object classes is less than one hundred.
9. The method of claim 1 , further comprising updating, by the one or more processors, a training database corresponding to the plurality of object classes using the 3D point cloud responsive to the 3D point cloud not matching any of the plurality of object classes.
10. The method of claim 1 , further comprising presenting, by a display device, information based on the selected object class.
11. A light detection and ranging (LIDAR) system, comprising:
a sensor configured to:
generate a transmitted signal using a laser source;
output the transmitted signal;
receive a return signal responsive to the transmitted signal; and
output a data signal representing at least one point cloud data point representing an object corresponding to the return signal; and
a processing circuit configured to:
determine at least one feature variable based on the at least one point cloud data point;
determine a first classification statistic based on the at least one feature variable;
determine a second classification statistic based on the at least one feature variable;
assign a selected object class from a plurality of object classes to the object by determining a closest match between the object and the selected object class using the first classification statistic and the second classification statistic; and
generate a control signal to control operation of an autonomous vehicle based on the selected object class.
12. The LIDAR system of claim 11 , wherein the sensor is configured to provide the transmitted signal as an optical signal that comprises an optical pulse in an optical frequency band.
13. The LIDAR system of claim 12 , wherein the processing circuit is configured to assign the first object class as the selected object class by:
determining a first candidate closest match between a first object class of the plurality of object classes and the object using the first classification statistic;
determining a second candidate closest match between a second object class of the plurality of object classes and the object using the second classification statistic; and
assigning the first object class as the selected object class to the object responsive to the first object class being the same as the second object class.
14. The LIDAR system of claim 13 , wherein the processing circuit is configured to assign the first object class as the selected object class by:
determining, responsive to the first object class not being the same as the second object class, a closest fit between the 3D point cloud and one of a first model point cloud of the first object class or a second model point cloud of the second object class; and
assigning one of the first object class or the second object class as the selected object class to the object based on the one of the first object class or the second object class corresponding to the closest fit.
15. The LIDAR system of claim 11 , wherein the processing circuit is configured to:
determine the first classification statistic based on a histogram of at least a subset of the at least one point cloud data point in each of a plurality of bins corresponding to a range of values of the at least one feature variable; and
determine the second classification statistic based on a covariance matrix based on the at least one feature variable.
16. The LIDAR system of claim 11 , wherein a number N of the plurality of object classes is less than one hundred.
17. The LIDAR system of claim 11 , further comprising a display device configured to present information based on the selected object class.
18. The LIDAR system of claim 11 , wherein the sensor is configured to perform at least one of frequency modulation or phase modulation to generate the transmitted signal.
19. An autonomous vehicle control system, comprising:
a processing circuit configured to:
determine at least one feature variable based on the at least one point cloud data point;
determine a first classification statistic based on the at least one feature variable;
determine a second classification statistic based on the at least one feature variable;
assign a selected object class from a plurality of object classes to the object by determining a closest match between the object and the selected object class using the first classification statistic and the second classification statistic; and
generate a control signal to control operation of an autonomous vehicle based on the selected object class.
20. The autonomous vehicle control system of claim 19 , wherein the processing circuit is configured to receive the at least one point cloud data point from a light detecting and ranging (LIDAR) sensor.