| CPC G06V 20/05 (2022.01) [G01N 33/18 (2013.01); G06T 7/50 (2017.01); G06V 10/7747 (2022.01); G06V 20/60 (2022.01); H04N 23/56 (2023.01); H04N 23/695 (2023.01); G06T 2207/20081 (2013.01)] | 23 Claims |
|
1. A computer-implemented method, comprising:
obtaining, by a camera, an image of water in a fish enclosure;
detecting, using a blob detector, a plurality of blobs in the image that represent particles suspended in the water in the fish enclosure;
determining a respective size associated with each of the plurality of blobs that represent the particles suspended in the water in the fish enclosure;
generating a histogram that, for each of a plurality of size intervals, includes a count of blobs in the fish enclosure whose size is in the size interval;
providing the histogram of the counts of blobs in the fish enclosure for each of the plurality of size intervals as an input to a machine learning model, the machine learning model having been trained to output values representing estimate amounts of turbidity in fish enclosures using training data that includes (i) histograms generated from images that were previously captured by cameras in the fish enclosures, and (ii) ground truth values that were measured by turbidity sensors in the fish enclosures are not cameras;
in response to providing the histogram as the input to the machine learning model, receiving, from the machine learning model, a value representing an estimated amount of turbidity of the water in the fish enclosure; and
when the value does not satisfy water clarity criteria, providing a signal to an image processor to discard or halt processing of images.
|