Annotation of infrared images for machine learning using beamsplitter-based camera system and methods
Systems and methods include an infrared camera configured to capture an infrared image of a scene, a visible light camera configured to capture a visible light image of the scene, and a logic device configured to simultaneously capture a pair of images of the scene comprising the infrared image of the scene and the visible image of the scene, align the pair of images so that a pixel location in one of the pair of images has a corresponding pixel location in the other image, classify the visible image, annotate the infrared image based, at least in part, on the classification of the visible image, and add the annotated infrared image to a neural network training dataset for use in training a neural network for infrared image classification.
1. A method comprising:
simultaneously capturing a pair of images of a scene comprising an infrared image of the scene and a visible image of the scene, wherein the pair of images is captured from substantially a same perspective through separate optical paths of a beamsplitter;
aligning the pair of images so that a pixel location in one of the pair of images has a corresponding pixel location in the other image, and both corresponding pixels represent the same position in the scene;
classifying the visible image, including an identification of one or more objects in the visible image;
classifying the infrared image, including an identification of one or more objects present in the infrared image;
annotating the infrared image based, at least in part, on the classification of the visible image and the classification of the infrared image; and
adding the annotated infrared image to a neural network training dataset for use in training a neural network for infrared image classification.
2. The method of claim 1 , wherein simultaneously capturing the pair of images of the scene further comprises capturing the infrared image using an infrared camera and capturing the visible image using a visible light camera.
3. The method of claim 1 , wherein simultaneously capturing the pair of images of the scene further comprises capturing the pair of images of the scene through the beamsplitter which is arranged to reflect a first image of the scene towards a first camera and pass through a second image of the scene to a second camera.
4. The method of claim 1 , wherein aligning the pair of images so that the pixel location in one of the pair of images has a corresponding pixel location in the other image further comprises determining a transformation between the pair of images so that the pixel location in each image represents a same location in the scene; and applying the transformation to the image pair to align the images.
5. The method of claim 1 , wherein classifying the visible image comprises receiving an image classification from a user and/or automatically generating a visible image classification using a trained neural network.
6. The method of claim 1 , further comprising classifying the infrared image by receiving an infrared image classification from a user and/or automatically generating an infrared image classification using a trained neural network.
7. The method of claim 1 , wherein the infrared image is annotated based, at least in part, on a classification of the infrared image.
8. The method of claim 1 , wherein annotating the infrared image comprises:
presenting a user interface displaying the visible image, the infrared image and corresponding annotations; and
receiving user input modifying one or more annotation.
9. The method of claim 1 , wherein annotating the infrared image comprises applying visible image classification information to the infrared image.
10. The method of claim 1 , further comprising training a neural network to classify an infrared image using the neural network training dataset including the annotated infrared image.
11. A system comprising:
an infrared camera configured to capture an infrared image of a scene;
a visible light camera configured to capture a visible light image of the scene;
a beamsplitter arranged to reflect a first image of the scene towards the infrared camera and pass through a second image of the scene to the visible light camera; and
a logic device configured to:
simultaneously capture a pair of images of the scene comprising the infrared image of the scene and the visible image of the scene, wherein the pair of images is captured from substantially a same perspective through separate optical paths of the beamsplitter;
align the pair of images so that a pixel location in one of the pair of images has a corresponding pixel location in the other image, and both corresponding pixels represent the same position in the scene;
classify the visible image, including an identification of one or more objects in the visible image;
classify the infrared image, including an identification of one or more objects in the infrared image;
annotate the infrared image based, at least in part, on the classification of the visible image and the classification of the infrared image; and
add the annotated infrared image to a neural network training dataset for use in training a neural network for infrared image classification.
12. The system of claim 11 , wherein the logic device is configured to align the pair of images wherein the pixel location in one of the pair of images has a corresponding pixel location in the other image further by determining a transformation between the pair of images so that the pixel location in each image represents a same location in the scene; and applying the transformation to the image pair to align the images.
13. The system of claim 11 , wherein the logic device is configured to classify the visible image by receiving an image classification from a user and/or automatically generating a visible image classification using a trained neural network.
14. The system of claim 11 , wherein the logic device is further configured to classify the infrared image by receiving an infrared image classification from a user and/or automatically generating an infrared image classification using a trained neural network.
15. The system of claim 11 , wherein the logic device is further configured to annotate the infrared image based, at least in part, on a classification of the infrared image.
16. The system of claim 11 , wherein the logic device is further configured to annotate the infrared image by presenting a user interface displaying the visible image, the infrared image and corresponding annotations; and receiving input from a user modifying one or more annotation.
17. The system of claim 11 , wherein the logic device is further configured to annotate the infrared image by applying visible image classification information to the infrared image.
18. The system of claim 11 , wherein the logic device is further configured to train a neural network to classify an infrared image using the neural network training dataset including the annotated infrared image.