Method and system for fill level determination
A method for fill level determination, which can include receiving a set training set, training a neural network, selecting reference images, and/or determining a container fill level. A system for fill level determination, which can include a computing system, one or more containers, and/or one or more content sensors.
1. A method for fullness metric assessment, comprising:
training a fullness metric classifier configured to classify fullness of container interiors by providing the classifier with a training set comprising a plurality of images of different container types in different states of fill;
receiving, from a content sensor, a subject image depicting a subject container interior of a subject container; and
using the fullness metric classifier, determining a fullness metric associated with the subject image, comprising:
providing the subject image and a reference image to the fullness metric classifier as an input, wherein the reference image depicts the subject container interior in the target fullness state of waste; and
in response to providing the subject image and the reference image to the fullness metric classifier, receiving, from the fullness metric classifier, information indicative of the fullness metric, wherein the information indicative of the fullness metric is determined based on the training set.
2. The method of claim 1 , wherein:
an empty container state is defined by a fullness metric value of zero; and
the target fullness state is defined by a fullness metric value less than a threshold value.
3. The method of claim 1 , wherein, in the target fullness state, the subject container interior defines an occupied volume fraction less than 20%, wherein the occupied volume fraction is equal to an occupied volume of the subject container interior divided by a volumetric capacity of the subject container interior.
4. The method of claim 1 , wherein:
the subject container interior defines a container floor;
the fullness metric is defined as an occupied floorspace fraction of the subject container; and
the occupied floorspace fraction is equal to an occupied floor area of the container floor divided by a total floor area of the container floor.
5. The method of claim 4 , wherein, in the target fullness state, the subject container interior defines an occupied floorspace fraction less than 10%.
6. The method of claim 4 , wherein the subject container is an intermodal freight container.
7. The method of claim 1 , wherein the fullness metric is defined as an occupied volume fraction of the subject container, wherein the occupied volume fraction is equal to an occupied volume of the subject container interior divided by a volumetric capacity of the subject container interior.
8. The method of claim 1 , wherein the fullness metric classifier comprises a neural network.
9. The method of claim 8 , wherein, the subject image and the reference image each define a respective set of spatial dimensions, and wherein providing the subject image and a reference image to the fullness metric classifier as the input comprises:
generating a convolutional neural network (CNN) input comprising the subject image and the reference stacked along a channel dimension, wherein each spatial dimension of the respective sets is orthogonal to the channel dimension; and
providing the CNN input to the neural network, wherein the neural network is a CNN.
10. The method of claim 1 , further comprising, before receiving the reference image: selecting the reference image from a set of candidate reference images, wherein each candidate reference image of the set depicts the subject container interior in the target fullness state.
11. The method of claim 1 , further comprising, before providing the subject image to the fullness metric classifier, determining, using a statistical classifier, that the subject image is valid, comprising determining that the subject image depicts a container interior, wherein providing the subject image to the fullness metric classifier is performed in response to determining that the subject image is valid.
12. The method of claim 11 , wherein the fullness metric classifier comprises a first trained neural network, wherein the statistical classifier comprises a second trained neural network distinct from the fullness metric classifier.