Panoramic camera systems
A camera system captures images from a set of cameras to generate binocular panoramic views of an environment. The cameras are oriented in the camera system to maximize the minimum number of cameras viewing a set of randomized test points. To calibrate the system, matching features between images are identified and used to estimate three-dimensional points external to the camera system. Calibration parameters are modified to improve the three-dimensional point estimates. When images are captured, a pipeline generates a depth map for each camera using reprojected views from adjacent cameras and an image pyramid that includes individual pixel depth refinement and filtering between levels of the pyramid. The images may be used generate views of the environment from different perspectives (relative to the image capture location) by generating depth surfaces corresponding to the depth maps and blending the depth surfaces.
1. A method comprising:
receiving a training data set comprising a plurality of training images, each training image associated with a predetermined depth map describing a depth of objects at locations corresponding to the training image;
generating a noisy depth map for each training image of the set of training images by applying random noise to the predetermined depth map associated with that training image;
generating, by applying a set of transformations to each training image of the set of training images, a feature map for each training image of the set of training images;
determining an improved depth map for each noisy depth map by filtering the noisy depth map using a joint bilateral filter guided by the feature map associated with the noisy depth map;
comparing, for each training image of the set of training images, the improved depth map to the predetermined depth map; and
training the set of transformations based on the comparison between the predetermined depth maps and the improved depth maps.
2. The method of claim 1 , wherein the set of transformations comprises a multilevel deep learning model.
3. The method of claim 2 , wherein training the set of transformations based on the comparison between the predetermined depth maps and the improved depth maps comprises training the multilevel deep learning model using the predetermined depth maps as ground truth.
4. The method of claim 1 , wherein the training images and the predetermined depth maps are generated based on a prerendered scene.
5. A non-transitory computer readable storage medium comprising instructions which, when executed by a processor, cause the processor to perform steps of:
receiving a training data set comprising a plurality of training images, each training image associated with a predetermined depth map describing a depth of objects at locations corresponding to the training image;
generating a noisy depth map for each training image of the set of training images by applying random noise to the predetermined depth map associated with that training image;
generating, by applying a set of transformations to each training image of the set of training images, a feature map for each training image of the set of training images;
determining an improved depth map for each noisy depth map by filtering the noisy depth map using a joint bilateral filter guided by the feature map associated with the noisy depth map;
comparing, for each training image of the set of training images, the improved depth map to the predetermined depth map; and
training the set of transformations based on the comparison between the predetermined depth maps and the improved depth maps.
6. The non-transitory computer readable storage medium of claim 5 , wherein the set of transformations comprises a multilevel deep learning model.
7. The non-transitory computer readable storage medium of claim 6 , wherein training the set of transformations based on the comparison between the predetermined depth maps and the improved depth maps comprises training the multilevel deep learning model using the predetermined depth maps as ground truth.
8. The non-transitory computer readable storage medium of claim 5 , wherein the training images and the predetermined depth maps are generated based on a prerendered scene.
9. A system comprising:
a processor; and
a non-transitory computer readable storage medium comprising instructions which, when executed by the processor, cause the processor to:
receive a training data set comprising a plurality of training images, each training image associated with a predetermined depth map describing a depth of objects at locations corresponding to the training image; and
generate a noisy depth map for each training image of the set of training images by applying random noise to the predetermined depth map associated with that training image;
generate, by applying a set of transformations to each training image of the set of training images, a feature map for each training image of the set of training images;
determine an improved depth map for each noisy depth map by filtering the noisy depth map using a joint bilateral filter guided by the feature map associated with the noisy depth map;
compare, for each training image of the set of training images, the improved depth map to the predetermined depth map; and
train the set of transformations based on the comparison between the predetermined depth maps and the improved depth maps.
10. The system of claim 9 , wherein the set of transformations comprises a multilevel deep learning model.
11. The system of claim 10 , wherein training the set of transformations based on the comparison between the predetermined depth maps and the improved depth maps comprises training the deep multilevel learning model using the predetermined depth maps as ground truth.
12. The system of claim 9 , wherein the training images and the predetermined depth maps are generated based on a prerendered scene.