System and method for three-dimensional scanning and for capturing a bidirectional reflectance distribution function

1. A method for generating a three-dimensional (3D) model of an object, comprising:

capturing a plurality of images of the object from a plurality of viewpoints, the images comprising a plurality of color images;

generating a 3D model of the object from the images;

for each patch of a plurality of planar patches corresponding to a surface of the 3D model:

mapping a plurality of image regions of the plurality of images to the patch, each image region comprising at least one color vector, and

computing, for each patch, at least one minimal color vector among the color vectors of the image regions mapped to the patch;

generating a diffuse component of a bidirectional reflectance distribution function (BRDF) for each patch of planar patches of the 3D model in accordance with the at least one minimal color vector computed for each patch;

subtracting the at least one minimal color vector from the color vector of each of the image regions mapped to the patch to compute a plurality of specular images of the patch;

outputting the 3D model with the BRDF for each patch, the BRDF further comprising a specular component separate from the diffuse component;

rendering a plurality of diffuse views based on the 3D model of the object, each view being rendered from a different pose with respect to the 3D model of the object;

generating a feature vector by extracting a plurality of features from the plurality of diffuse views;

supplying the feature vector to a trained convolutional neural network to classify the patches of the object as clean or defective, wherein assigning a classification to the patches of the object is performed by the convolutional neural network, and

wherein the convolutional neural network is trained by:

receiving a plurality of training 3D models of objects and corresponding training classifications,

rendering a plurality of views of the 3D models with controlled lighting to generate training data,

computing a plurality of feature vectors from the views by the convolutional neural network,

computing parameters of the convolutional neural network,

computing a training error metric between the training classifications of the training 3D models with outputs of the convolutional neural network configured based on the parameters,

computing a validation error metric in accordance with a plurality of validation 3D models separate from the training 3D models, and

in response to determining that the training error metric and the validation error metric satisfy a threshold, configuring the neural network in accordance with the parameters.

2. The method of claim 1 , wherein at least one of the plurality of features is computed based on color information of the object in the plurality of diffuse views of the object.

3. The method of claim 1 , wherein at least one of the plurality of features is computed based on shape information of the object in the plurality of diffuse views of the object.

4. The method of claim 1 , wherein at least one of the plurality of features is computed based on color information and shape information of the object in the plurality of diffuse views of the object.

5. The method of claim 1 , wherein the 3D model comprises a point cloud.

6. The method of claim 5 , wherein each patch of the plurality of planar patches corresponds to one or more points of the point cloud.

7. The method of claim 1 , further comprising:

initializing one or more parameters;

rendering the 3D model in accordance with the BRDF set in accordance with the one or more parameters to render a plurality of rendered views of the patch;

computing an error function in accordance with a difference between the rendered views of the patch with the image regions mapped to the patch; and

computing the one or more parameters by iteratively updating the one or more parameters to minimize the error function.

8. The method of claim 1 , wherein the plurality of images are captured by a plurality of different cameras.

9. The method of claim 8 , wherein the plurality of images of the object are captured while the object is on a conveyor belt.

10. The method of claim 1 , wherein the plurality of images of the object from the plurality of viewpoints are captured by a single camera.

11. The method of claim 10 , wherein the camera is a stereoscopic depth camera comprising a first infrared camera, a second infrared camera, and a color camera.

12. The method of claim 1 , wherein the 3D model is a model of less than the entire exterior surface of the object.

13. A system for generating a three-dimensional (3D) model of an object, the system comprising:

a depth camera system;

a processor coupled to the depth camera system; and

memory having instructions stored thereon that, when executed by the processor, cause the processor to:

capture a plurality of images of the object from a plurality of viewpoints, the images comprising a plurality of color images;

generate a 3D model of the object from the images;

for each patch of a plurality of planar patches corresponding to a surface of the 3D model:

map a plurality of image regions of the plurality of images to the patch, each image region comprising at least one color vector, and

compute, for each patch, at least one minimal color vector among the color vectors of the image regions mapped to the patch;

generate a diffuse component of a bidirectional reflectance distribution function (BRDF) for each patch of planar patches of the 3D model in accordance with the at least one minimal color vector computed for each patch;

output the 3D model with the BRDF for each patch, the BRDF further comprising a specular component separate from the diffuse component;

render a plurality of diffuse views based on the 3D model of the object, each view being rendered from a different pose with respect to the 3D model of the object;

generate a feature vector by extracting a plurality of features from the plurality of diffuse views;

supply the feature vector to a trained convolutional neural network to classify the patches of the object as clean or defective, wherein assigning a classification to the patches of the object is performed by the convolutional neural network, and

wherein the convolutional neural network is trained by:

receiving a plurality of training 3D models of objects and corresponding training classifications,

rendering a plurality of views of the 3D models with controlled lighting to generate training data,

computing a plurality of feature vectors from the views by the convolutional neural network,

computing parameters of the convolutional neural network,

computing a training error metric between the training classifications of the training 3D models with outputs of the convolutional neural network configured based on the parameters,

computing a validation error metric in accordance with a plurality of validation 3D models separate from the training 3D models, and

in response to determining that the training error metric and the validation error metric satisfy a threshold, configuring the neural network in accordance with the parameters.

14. The system of claim 13 , wherein at least one of the plurality of features is computed based on color information of the object in the plurality of diffuse views of the object.

15. The system of claim 13 , wherein at least one of the plurality of features is computed based on shape information of the object in the plurality of diffuse views of the object.

16. The system of claim 13 , wherein at least one of the plurality of features is computed based on color information and shape information of the object in the plurality of diffuse views of the object.

17. The system of claim 13 , wherein the 3D model comprises a point cloud.

18. The system of claim 17 , wherein each patch of the plurality of planar patches corresponds to one or more points of the point cloud.

19. The system of claim 13 , wherein the memory further stores instructions that, when executed by the processor, cause the processor to:

compute the specular component of the BRDF for each of the planar patches by:

initializing one or more parameters;

rendering the 3D model in accordance with the BRDF set in accordance with the one or more parameters to render a plurality of rendered views of the patch;

computing an error function in accordance with a difference between the rendered views of the patch with the image regions mapped to the patch; and

computing the one or more parameters by iteratively updating the one or more parameters to minimize the error function.

20. The system of claim 13 , further comprising a plurality of different cameras, wherein the plurality of images are captured by the plurality of different cameras.

21. The system of claim 20 , wherein the plurality of different cameras is configured to image the object while the object is on a conveyor belt.

22. The system of claim 13 , further comprising a single camera, wherein the plurality of images of the object from the plurality of viewpoints are captured by the single camera.

23. The system of claim 22 , wherein the camera is a stereoscopic depth camera comprising a first infrared camera, a second infrared camera, and a color camera.

24. The system of claim 13 , wherein the 3D model is a model of less than the entire exterior surface of the object.