Material handling method, apparatus, and system for identification of a region-of-interest

1. A material handling method comprising:

generating, by an image processing unit, three-dimensional (3-D) point cloud data based on a field-of-view (FOV) of an image capturing unit, wherein the FOV comprises an interior of a container;

identifying, by the image processing unit, a plurality of cluster areas of the container interior;

extracting, by the image processing unit, a first set of cluster areas from the plurality of cluster areas based on orientation data associated with the 3-D point cloud data;

determining, by the image processing unit, a candidate region of the first set of cluster areas, wherein the candidate region comprises a cluster area;

determining a ratio of a cross-sectional area of the candidate region and a cross-sectional area of the container;

determining, by a machine learning unit, a classification score of the candidate region when the determined ratio exceeds a first cross-sectional threshold;

classifying, by the machine learning unit, the candidate region as a region-of-interest in an instance in which the classification score of the candidate region exceeds a first classification threshold; and

defining, by a processor, a navigation path in the container based on a data set associated with the region-of-interest, wherein the navigation path is configured to prevent collision with the region-of-interest.

2. The method of claim 1 , wherein generating the 3-D point cloud data comprises accumulating:

3-D point cloud data generated based on point cloud data of the FOV captured by an image capturing unit positioned on a material handling apparatuses, and

kinematic data collected based on a movement of an article manipulator of the material handling apparatus during a scanning operation of the FOV.

3. The method of claim 2 , further comprising transforming, by the image processing unit, a first format of the 3-D point cloud data to a second format, wherein the first format corresponds to a light detection and ranging frame and the second format corresponds to a transformed frame compatible with the image processing unit.

4. The method of claim 3 , further comprising pre-processing, by the image processing unit, the 3-D point cloud data, wherein the pre-processing comprises:

removing, by the image processing unit, noise data points from the 3-D point cloud data in the second format;

segmenting, by the image processing unit, the 3-D point cloud data from which the noise data points are removed; and

determining, by the image processing unit, the plurality of cluster areas based on the segmentation of the 3-D point cloud data.

5. The method of claim 1 , further comprising:

extracting, by the image processing unit, a second set of cluster areas from the plurality of cluster areas based on at least an orientation of the second set of cluster areas with respect to a material handling apparatus; and

determining, by the image processing unit, a type of each of the second set of cluster areas of the plurality of cluster areas of the container,

wherein the type of each of the second set of cluster areas corresponds to one of a side wall, a floor area, or a ceiling area of the container, wherein the first set of cluster areas of the plurality of cluster areas correspond to a back wall and a product wall of the docked container.

6. The method of claim 5 , further comprising: obtaining, by the image processing unit, updated 3-D point cloud data, wherein the updated 3-D point cloud data comprises 3-D point cloud data from which the second set of cluster areas are clipped.

7. The method of claim 6 , further comprising:

segmenting, by the image processing unit, the updated 3-D point cloud data;

determining, by the image processing unit, the first set of cluster areas based on the segmentation of the updated 3-D point cloud data; and

identifying, by the image processing unit, the candidate region as the cluster area of the first set of cluster areas that is disposed at intersection areas between the first set of cluster areas and the second set of cluster areas, wherein the candidate region defines a 2-D depth map.

8. The method of claim 7 , further comprising:

converting, by the image processing unit, the updated 3-D point cloud data into a range image, wherein the generation of the 2-D depth map is based on the range image and corresponds to depth information associated with a point in the 3-D point cloud data.

9. The method of claim 7 , further comprising:

generating, by the processor, the data set associated with the region-of-interest based on localization of the candidate region; and

transmitting, by the processor, the data set associated with the region-of-interest to a programmable logic controller associated with the material handling apparatus.

10. The method of claim 7 , further comprising:

providing, by the processor, the identified candidate region to a trained single input convolution neural network module;

determining, by the convolution neural network module, a score associated with the identified candidate region;

classifying, by the convolution neural network module, the identified candidate region as a region-of-interest in an instance in which the classification score exceeds a second classification threshold, wherein the region-of-interest is a back wall of the container; and

classifying, by the convolution neural network module, the identified candidate region as a product wall in an instance in which the classification score fails to exceed the second classification threshold.

11. The method of claim 10 , further comprising:

defining, by the processor, a navigation path in the container based on a data set associated with the classified region-of-interest in an instance in which the classification score fails to exceed the second classification threshold; and

controlling, by an article manipulation unit, an article manipulator to follow the navigation path to manipulate a plurality of articles placed in the container in an area corresponding to the product wall.

12. The method of claim 7 , wherein the classification of the candidate region further comprises:

obtaining, by the machine learning unit, a plurality of key interest points in the candidate region,

wherein in an instance in which a count of the plurality of key interest points exceeds a defined value, the candidate region corresponds to the region-of-interest,

wherein in an instance in which the count of the plurality of key interest points fails to exceed the defined value, the candidate region does not correspond to the region-of-interest.

13. The method of claim 12 , wherein the classification of the candidate region further comprises:

determining, by the machine learning unit, a plurality of key descriptors based on the plurality of key interest points;

determining, by the machine learning unit, a distribution of the plurality of key descriptors; and

generating, by the machine learning unit, a feature vector based on the distribution of the plurality of key descriptors.

14. The method of claim 13 , wherein the classification of the candidate region further comprises:

training, by the machine learning unit, a machine learning model to provide the classification score; and

classifying, by the machine learning unit, the candidate region as a product wall in an instance in which the classification score of the feature vector fails to exceed the first classification threshold.

15. The method of claim 1 , further comprising:

controlling, by an article manipulation unit, an article manipulator to follow the navigation path and manipulate a plurality of articles placed in the container.

16. The method of claim 1 , further comprising:

localizing, by the image processing unit, the candidate region as the region-of-interest in an instance in which the difference between the cross-sectional area of the container and the cross-sectional area of the cluster area fails to exceed the first cross-sectional threshold.

17. A material handling apparatus comprising:

an article manipulator;

an image-capturing device positioned on the material handling apparatus; and

a processor communicatively coupled with the article manipulator, an image processing unit, and the image-capturing device, wherein the processor is adapted to:

cause the image processing unit to generate three-dimensional (3-D) point cloud data based on a field-of-view (FOV) of the image capturing unit, wherein the FOV comprises an interior of a container;

cause the image processing unit to identify a plurality of cluster areas of the container interior;

cause the image processing unit to extract a first set of cluster areas from a plurality of cluster areas based on orientation data associated with the 3-D point cloud data;

cause the image processing unit to:

determine a candidate region of the first set of cluster areas, wherein the candidate region comprises a cluster area; and

determine a ratio of a cross-sectional area of the candidate region and a cross-sectional area of the container;

cause a machine learning unit to determine a classification score of the candidate region in an instance in which the determined ratio exceeds a first cross-sectional threshold;

cause the machine learning unit to classify the candidate region as a region-of-interest in an instance in which the classification score of the candidate region exceeds a first classification threshold; and

define a navigation path in the container based on a data set associated with the classified region-of-interest, wherein the navigation path is configured to prevent collision with the region-of-interest.

18. A control system for a material handling apparatus, the control system comprising:

an image-capturing device; and

a processor communicatively coupled to the image-capturing device and an image processing unit, wherein the processor is adapted to:

cause the image processing unit to generate three-dimensional (3-D) point cloud data based on a field-of-view (FOV) of the image capturing unit, wherein the FOV comprises an interior of a container;

cause the image processing unit to identify a plurality of cluster areas of the container interior;

cause the image processing unit to extract a first set of cluster areas from a plurality of cluster areas based on orientation data associated with the 3-D point cloud data;

cause the image processing unit to:

determine a candidate region of the first set of cluster areas, wherein the candidate region comprises a cluster area; and

determine a ratio of a cross-sectional area of the candidate region and a cross-sectional area of the container;

cause a machine learning unit to determine a classification score of the candidate region in an instance in which the determined ratio exceeds a first cross-sectional threshold;

cause the machine learning unit to classify the candidate region as a region-of-interest in an instance in which the classification score of the candidate region exceeds a first classification threshold; and

define a navigation path in the container based on a data set associated with the classified region-of-interest, wherein the navigation path is configured to prevent collision with the region-of-interest.

19. The control system of claim 18 , wherein the processor is further adapted to cause an article manipulation unit to control an article manipulator to follow the navigation path and manipulate a plurality of articles placed in the container.

20. The control system of claim 18 , wherein the processor is further adapted to cause the image processing unit to localize the candidate region as the region-of-interest in an instance in which the difference between the cross-sectional area of the container and the cross-sectional area of the cluster area fails to exceed the first cross-sectional threshold.