Data augmentation for obstruction learning
Disclosed is data augmentation for obstruction learning. In some embodiments, a method comprises: obtaining images of obstacles that obstruct an object of interest; randomly scaling the images; extracting portions of the randomly scaled images at random positions in the randomly scaled images; and training a machine learning model using the extracted portions. In some embodiments, a method comprises: capturing images of obstacles that are at least partially obstructing an object of interest (e.g., a traffic light); processing the images using a machine learning model that is trained on a dataset that includes selected portions of augmented images of obstacles (e.g., large vehicles) that obstruct the object of interest; predicting a state of the object of interest based on output of the machine learning model; planning a trajectory for a vehicle; and causing the vehicle to travel the trajectory.
1 . A method comprising:
obtaining, with at least one processor, images of obstacles that obscure an object of interest;
randomly scaling, with at least one processor, the images;
extracting, with the at least one processor, portions of the randomly scaled images at random positions in the randomly scaled images; and
training, with the at least one processor, a machine learning model using the portions.
2 . The method of claim 1 , wherein the extracted portions include a margin around the obstacle.
3 . The method of claim 1 , wherein the images include narrow field of view (FOV) images, and the method further comprises:
inputting, with the at least one processor, the narrow FOV images into an image segmentation network; and
filtering manually obstacles detected by the image segmentation network.
4 . The method of claim 1 , wherein the extracted portions are squares of pixels extracted from the randomly scaled images.
5 . The method of claim 1 , wherein the randomly scaled images include medium field of view (FOV) images, and the method further comprises augmenting color or brightness of the randomly scaled images.
6 . The method of claim 1 , wherein the object of interest is a traffic light, and the machine learning model is trained to predict a state of the traffic light.
7 . The method of claim 6 , wherein the predicted state of the traffic light is an unknown state due to obstruction of the traffic light by an obstacle.
8 . A method comprising:
capturing, with at least one sensor, images of obstacles that are at least partially obstructing an object of interest;
processing, with at least one processor, the images using a machine learning model that is trained on a dataset that includes portions of augmented images of obstacles that obstruct the object of interest, wherein the portions of augmented images are extracted from randomly scaled images;
predicting, with the at least one processor, a state of the object of interest based on output of the machine learning model;
planning, with the at least one processor, a trajectory for a vehicle; and
causing, with a control circuit of the vehicle, the vehicle to travel the trajectory.
9 . The method of claim 8 , wherein the portions of augmented images are polygons of pixels extracted from random positions in the augmented images.
10 . The method of claim 9 , wherein the polygons are squares of pixels extracted from the random positions in the augmented images.
11 . The method of claim 8 , wherein the augmented images include medium field of view (FOV) images, and the method further comprises augmenting color or brightness of the medium FOV images.
12 . The method of claim 8 , wherein the augmented images include narrow field of view (FOV) images and medium FOV images that are randomly scaled.
13 . The method of claim 8 , wherein the object of interest is a traffic light, and the machine learning model is trained to predict a state of the traffic light.
14 . The method of claim 13 , wherein the predicted state of the traffic light is an unknown state due to obstruction of the traffic light by an obstacle.
15 . A vehicle comprising:
at least one sensor;
at least one processor;
memory storing instructions that when executed by the at least one processor, causes the at least one processor to perform operations comprising:
capturing, with the at least one sensor, images of obstacles that are at least partially obstructing an object of interest;
processing, with the at least one processor, the images using a machine learning model that is trained on a dataset that includes selected portions of augmented images of obstacles that obstruct the object of interest, wherein the portions of augmented images are extracted from randomly scaled images;
predicting, with the at least one processor, a state of the object of interest based on output of the machine learning model;
planning, with the at least one processor, a trajectory for a vehicle; and
causing, with a control circuit of the vehicle, the vehicle to travel the trajectory.
16 . The vehicle of claim 15 , wherein the selected portions of augmented images are polygons of pixels extracted from randomly selected positions in the augmented images.
17 . The vehicle of claim 16 , wherein the polygons are squares of pixels.
18 . The vehicle of claim 15 , wherein the augmented images include medium field of view (FOV) images, and the operations comprise augmenting color or brightness of the medium FOV images.
19 . The vehicle of claim 15 , wherein the augmented images include narrow field of view (FOV) images and medium FOV images.
20 . The vehicle of claim 15 , wherein the object of interest is a traffic light, and the machine learning model is trained to predict a state of the traffic light.