Methods, systems, and apparatuses for storage analysis and management
Inventors
Gregory, Whitney Ann • Bandi, Hemanth • Li, Dan • Dhaka, Devang • Kumar, Anshul
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Described herein are methods, systems, and apparatuses for storage analysis and management using computer vision techniques. An imaging device may capture a series of images of a plurality of containers. The series of images may be analyzed by a classification model to determine a current capacity of each container. When the classification model determines that a current capacity of a container(s) satisfies a threshold(s), at least one remedial action may be performed.
Core Innovation
The invention relates to computer vision for storage analysis and management using an imaging device that captures at least one first image including a plurality of containers. A computing device receives the at least one first image and identifies, via a classification model, the plurality of containers present within the at least one first image. For each container, the computing device determines a quantity of pixels within the at least one first image associated with the corresponding container, where the quantity of pixels is indicative of a current capacity of the corresponding container.
The computing device determines, based on a reference quantity of pixels and the quantity of pixels within the at least one first image indicative of the current capacity of a first container of the plurality of containers, that a first threshold quantity is satisfied. The reference quantity of pixels is associated with the first threshold quantity. The computing device then causes, based on the first threshold quantity being satisfied, at least one remedial action to be performed.
In some embodiments, the remedial action includes outputting an alert and/or other remedial output. The system can further associate multiple containers with portions of a first image, determine which image portion corresponds to a given container, and identify identifiers present within the determined portions. The invention also supports training the classification model using training images containing reference pixel quantities associated with known capacities and current-capacity pixel quantities.
Claims Coverage
The independent claims are directed to a method, an apparatus, and a non-transitory computer-readable storage medium. The inventive features center on pixel-quantity-based container capacity estimation, threshold comparison against a reference quantity of pixels associated with a threshold quantity, and remedial action when the threshold is satisfied.
Classification model container identification in images
Identifying, via a classification model, a plurality of containers present within the at least one first image.
Pixel quantity indicative of current capacity
Determining, for each container of the plurality of containers, a quantity of pixels within the at least one first image associated with the corresponding container, wherein the quantity of pixels is indicative of a current capacity of the corresponding container.
Reference pixel quantity and threshold satisfaction determination
Determining, based on a reference quantity of pixels and the quantity of pixels within the at least one first image indicative of the current capacity of a first container of the plurality of containers, that a first threshold quantity is satisfied, wherein the reference quantity of pixels is associated with the first threshold quantity.
Threshold-triggered remedial action
Causing, based on the first threshold quantity being satisfied, at least one remedial action to be performed.
Background subtraction technique for threshold quantity
Determining the first threshold quantity using a background subtraction technique.
Identifier extraction from image portions corresponding to containers
Associating multiple containers with portions of a first image, determining which image portions correspond to each first container, and identifying identifiers present within the determined portions.
Training using reference and current pixel-quantity examples
Receiving a plurality of training images that include reference quantity of pixels associated with known container capacity and quantity of pixels indicative of current capacity, and training the classification model using the training images.
Across the independent claims, the core claim coverage is the same: a classification model identifies containers in received images, pixel quantities associated with each container indicate current capacity, a reference quantity of pixels associated with a threshold quantity is compared to determine satisfaction, and remedial action is caused when satisfied. Dependent aspects further specify background subtraction, container-to-image portion mapping with identifier identification, and training using reference versus current pixel-quantity training images.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?