Systems and methods for joint learning of complex visual inspection tasks using computer vision

1. A method for performing automatic visual inspection, comprising:

capturing visual information of an object using a scanning system comprising a plurality of cameras;

extracting, by a computing system comprising a processor and memory, one or more feature maps from the visual information using one or more feature extractors;

classifying, by the computing system, the object by supplying the one or more feature maps to a complex classifier to compute a classification of the object, the complex classifier comprising:

a plurality of simple classifiers, each simple classifier of the plurality of simple classifiers being configured to compute outputs representing a characteristic of the object, and

one or more logical operators configured to combine the outputs of the plurality of simple classifiers to compute the classification of the object;

outputting, by the computing system, the classification of the object as a result of the automatic visual inspection;

wherein each simple classifier of the plurality of simple classifiers is configured by a corresponding threshold parameter of a plurality of threshold parameters, wherein the plurality of threshold parameters are jointly trained; and

wherein the plurality of threshold parameters are jointly trained by:

sampling a parameter space to select a plurality of sets of threshold parameters to configure the plurality of simple classifiers,

computing a True Positive rate (TPr) and a False Positive rate (FPr) for each set of threshold parameters of the plurality of sets of threshold parameters by:

configuring the complex classifier by configuring the plurality of simple classifiers based on the set of threshold parameters; and

computing the TPr and the FPr for the configuration by supplying the configured complex classifier with a validation set of data,

identifying a Pareto front comprising best performing sets of configuration parameters in accordance with the TPr and FPr for each set of the sets of configuration parameters, and

selecting a set of configuration parameters from the Pareto front in accordance with a rule set in accordance with a domain.

2. The method of claim 1 , wherein the one or more feature extractors comprise one or more convolutional neural networks.

3. The method of claim 1 , wherein the plurality of simple classifiers comprises one or more neural networks.

4. The method of claim 3 , wherein the plurality of simple classifiers comprises one or more support vector machines, and

wherein at least one logical operation is configured to combine an output of the one or more neural networks and an output of the one or more support vector machines.

5. The method of claim 1 , wherein the plurality of simple classifiers comprises one or more regression model.

6. The method of claim 1 , wherein the plurality of simple classifiers comprises one or more label-based classifier configured to perform on text detection.

7. The method of claim 1 wherein the visual information comprises color images, grayscale images, or depth maps.

8. The method of claim 7 , wherein the visual information comprises at least one depth map,

wherein the at least one depth map is captured by a depth camera system of the plurality of cameras.

9. The method of claim 8 , wherein the depth camera system comprises:

a time-of-flight depth camera;

a structured light depth camera;

a stereo depth camera comprising:

at least two color cameras;

a stereo depth camera comprising:

at least two color cameras; and

a color projector;

a stereo depth camera comprising:

at least two infrared cameras; or

a stereo depth camera comprising:

at least two infrared cameras;

an infrared projector; and

a color camera.

10. The method of claim 8 , wherein the plurality of simple classifiers comprises a classifier based on mathematical modeling of the at least one depth map.

11. The method of claim 1 , wherein a feature map of the one or more feature maps is provided as input to at least two of the plurality of simple classifiers.

12. The method of claim 1 , wherein the classification of the object comprises an identification of a category of a plurality of categories of objects.

13. The method of claim 1 , wherein the classification of the object comprises an identification of one or more properties of the object based on the visual information.

14. A visual inspection system comprising:

a scanner system comprising a plurality of cameras; and

a computing system connected to the scanner system over a computer network, the computing system comprising a processor and memory storing instructions that, when executed by the processor, cause the processor to:

control the scanner system to capture visual information of an object;

extract one or more feature maps from the visual information using one or more feature extractors;

classify, by the computing system, the object by supplying the one or more feature maps to a complex classifier to compute a classification of the object, the complex classifier comprising:

a plurality of simple classifiers, each simple classifier of the plurality of simple classifiers being configured to compute outputs representing a characteristic of the object, and

one or more logical operators configured to combine the outputs of the plurality of simple classifiers to compute the classification of the object;

output, by the computing system, the classification of the object as a result of an automatic visual inspection of the object;

wherein each simple classifier of the plurality of simple classifiers is configured by a corresponding threshold parameter of a plurality of threshold parameters, wherein the plurality of threshold parameters are jointly trained; and

wherein the plurality of threshold parameters are jointly trained by:

sampling a parameter space to select a plurality of sets of threshold parameters to configure the plurality of simple classifiers,

computing a True Positive rate (TPr) and a False Positive rate (FPr) for each set of threshold parameters of the plurality of sets of threshold parameters by:

configuring the complex classifier by configuring the plurality of simple classifiers based on the set of threshold parameters; and

computing the TPr and the FPr for the configuration by supplying the configured complex classifier with a validation set of data,

identifying a Pareto front comprising best performing sets of configuration parameters in accordance with the TPr and FPr for each set of the sets of configuration parameters, and

selecting a set of configuration parameters from the Pareto front in accordance with a rule set in accordance with a domain.

15. The visual inspection system of claim 14 , wherein the scanner system comprises at least one color camera.

16. The visual inspection system of claim 14 , wherein the scanner system comprises at least one depth camera.

17. The visual inspection system of claim 14 , further comprising a user device comprising a display device, the user device being configured to display:

the classification of the object; and

at least one characteristic of the object computed by at least one simple classifier of the plurality of simple classifiers.

18. The visual inspection system of claim 14 , wherein the computing system is configured to control a conveyor system to redirect movement of the object in accordance with the classification.