Systems and methods for automatic labeling of images for supervised machine learning
A method of automatic labeling of images for supervised machine learning includes obtaining images of roadside objects with a camera mounted to a vehicle, recording a position and orientation of the vehicle within a defined coordinate system while obtaining the images recording position information for each roadside object with the same defined coordinates system as used while recording the position and orientation of the vehicle, and correlating a position of each of the obtained images of the roadside objects with the position information of each roadside object in view of the recorded position and orientation of the vehicle. The images are labeled to identify the roadside objects in view of the correlated position of each of the obtained images of the roadside objects.
1. A method of generating input-output training or validating pairs for supervised machine learning comprising:
obtaining images of roadside objects with a camera mounted to a vehicle;
recording a position and orientation of the vehicle within a defined coordinate system while obtaining the images;
obtaining a position for each roadside object from a high definition map;
converting at least one of the position for each roadside object or the recorded position of the vehicle so as to have the same defined coordinates system;
following the converting, determining a position in each of the obtained images of the roadside objects with the position of each roadside object in view of the recorded position and orientation of the vehicle; and
labeling the images to identify the roadside objects in each of the obtained images in view of the corresponding determined position of the roadside objects to create labeled images, each obtained image and corresponding labeled image forming an input-output pair to train or validate a machine learning algorithm to detect roadside objects in images.
2. The method as recited in claim 1 , wherein the roadside objects comprises signs.
3. The method as recited in claim 2 , further comprising determining a relative position of the vehicle to a position of the sign.
4. The method as recited in claim 3 , the method comprising correcting a recorded orientation between the sign and the vehicle for optical characteristics of the camera.
5. The method as recited in claim 1 , wherein the labeling comprises providing, in each labeled image, a marking or outline on each of the roadside objects.
6. The method as recited in claim 1 , wherein the defined coordinate system is a Cartesian coordinate system.
7. The method as recited in claim 1 , wherein the recording a position and orientation of the vehicle is performed using a DGPS/IMU.
8. The method as recited in claim 1 , wherein the recording a position and orientation of a vehicle and recording position information for each roadside object are performed using a memory device of the vehicle.
9. The method as recited in claim 1 , further comprising
obtaining a first timestamp data set associated with the images;
obtaining a second timestamp data set associated with the position and orientation information; and
wherein the determining a position of the roadside objects in each of the obtained images includes synchronizing the first timestamp data set and the second timestamp data set.
10. The method as recited in claim 9 , wherein the roadside objects comprises signs, and the high definition map includes positional information for each sign.
11. The method as recited in claim 1 , the labeling comprises providing a marking or outline on each sign.
12. The method as recited in claim 11 , the method comprising identifying at least one corner of each sign with a Cartesian coordinate.
13. The method as recited in claim 1 , wherein determining the position of the roadside objects in each of the obtained images uses a perspective transform based on parameters of the camera to determine a position of one or more corners of each roadside object.
14. The method as recited in claim 1 , further comprising correcting a misalignment between roadside objects appearing in at least one labeled image and the corresponding label therefor, including, for each roadside object in the at least one labeled image, searching for at least one roadside object edge corresponding to the object in the at least one labeled image and correcting alignment of the corresponding label relative to the at least one roadside object in the at least one labeled image based upon the at least one roadside object edge found.
15. A system for generating input-output training or validating pairs for supervised machine learning comprising:
a camera mounted to a vehicle and configured to capture images of objects external to the vehicle;
a DGPS/IMU which determines position and orientation information of the vehicle during the capturing of the images of the objects;
a memory device for recording the images and the associated position and orientation information;
a high definition (HD) map including HD map positional information for the objects; and
at least one computing device programmed to synchronize the capture of the images; with the positional and orientation information of the vehicle, and to label the objects in the images based upon the HD map positional information of the objects and the position and orientation information of the vehicle to provide a set of labeled images in which the objects are labeled, each captured image and the corresponding labeled image forming an input-output training pair for a training a machine learning algorithm.
16. The system as recited in claim 15 , comprising
at least one clock for providing a first timestamp data set associated with the captured images and a second timestamp dataset associated with the position and orientation information, wherein the at least one computing device synchronizes the captured images based on the first timestamp data set and the positional and orientation information of the vehicle based on the second timestamp data set.
17. The system as recited in claim 15 , wherein the at least one computing device includes a vehicle controller in communication with the camera, the DGPS/IMU and the HD map.
18. The system as recited in claim 17 , wherein the vehicle controller includes the memory device.
19. The system as recited in claim 15 , wherein the DGPS/IMU determines a position of one or more corners of the objects in each of the obtained images using a perspective transform based on parameters of the camera.
20. The system as recited in claim 15 , wherein the roadside objects comprise signs.
21. The system as recited in claim 15 , wherein the at least one computing device is further configured to
obtain a first timestamp data set associated with the images;
obtain a second timestamp data set associated with the position and orientation information; and
wherein determining a position of the objects in each of the obtained images includes synchronizing the first timestamp data set and the second timestamp data set.
22. The system as recited in claim 15 , wherein the at least one computing device is further programmed to correct a misalignment between one or more objects appearing in at least one labeled image and the corresponding label therefor, including, for each object in the at least one labeled image, searching for at least one object edge corresponding to the object in the at least one labeled image and correcting alignment of the corresponding label relative to the at least one object in the at least one labeled image based upon the at least one object edge found.