Systems and methods for identifying and labeling objects in images using an index
A method for automatically identifying and labeling objects in images using an index includes accessing first video frames captured at a first physical location. The method further includes identifying a plurality of first objects from the first video frames and generating first composite vectors for the plurality of first objects. The method further includes storing the first composite vectors in an index. The method further includes accessing second video frames captured at a second physical location. The method further includes identifying a plurality of second objects from the second video frames and generating second composite vectors for the plurality of second objects. The method further includes determining, using the index and the second composite vectors for the plurality of second objects, a plurality of similar objects. The method further includes displaying images of one or more of the plurality of similar objects in a graphical user interface.
1. A system comprising:
one or more memory units; and
one or more computer processors communicatively coupled to the one or more memory units and configured to perform operations comprising:
access a plurality of first video frames of a first video captured at a first physical location;
identify a plurality of first objects from the plurality of first video frames;
generate a plurality of first composite vectors for the plurality of first objects;
store the plurality of first composite vectors in an index;
access a plurality of second video frames of a second video captured at a second physical location;
identify a plurality of second objects from the plurality of second video frames;
generate a plurality of second composite vectors for the plurality of second objects;
determine, using the index and the plurality of second composite vectors for the plurality of second objects, a plurality of similar objects; and
display images of one or more of the plurality of similar objects in a graphical user interface;
wherein the plurality of first composite vectors and the plurality of second composite vectors are linear combinations of a plurality of vectors generated for each of the plurality of first and second objects.
2. The system of claim 1 , wherein:
identifying the plurality of first objects from the plurality of first video frames comprises:
generating a plurality of first masks, each first mask comprising neighboring pixels that are determined to be related; and
extracting the plurality of first objects from the plurality of first video frames based on the generated plurality of first masks; and
identifying the plurality of second objects from the plurality of second video frames comprises:
generating a plurality of second masks, each second mask comprising neighboring pixels that are determined to be related; and
extracting the plurality of second objects from the plurality of second video frames based on the generated plurality of second masks.
3. The system of claim 1 , wherein the plurality of vectors generated for each of the plurality of first and second objects comprises:
an appearance vector that is a mathematical representation of an appearance of a particular first or second object;
a behavior vector that is a mathematical representation of a behavior or movement of the particular first or second object; and
a shape vector that is a mathematical representation of a size or shape of the particular first or second object.
4. The system of claim 1 , the operations further comprising displaying a user-editable label for the plurality of similar objects in the graphical user interface.
5. The system of claim 1 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises calculating a plurality of similarity scores for the plurality of second objects using the index.
6. The system of claim 5 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises:
accessing a user-selection of a selected second object of the plurality of second objects;
calculating similarity scores between the selected second object and other objects within the index; and
selecting the other objects within the index whose calculated similarity scores with the selected second object meets a predetermined similarity value.
7. The system of claim 5 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises automatically calculating similarity scores between each of the plurality of second objects and every other object within the index.
8. A method by a computing system for automatically identifying and labeling objects in images using an index, the method comprising:
accessing a plurality of first video frames of a first video captured at a first physical location;
identifying a plurality of first objects from the plurality of first video frames;
generating a plurality of first composite vectors for the plurality of first objects;
storing the plurality of first composite vectors in an index;
accessing a plurality of second video frames of a second video captured at a second physical location;
identifying a plurality of second objects from the plurality of second video frames;
generating a plurality of second composite vectors for the plurality of second objects;
determining, using the index and the plurality of second composite vectors for the plurality of second objects, a plurality of similar objects; and
displaying images of one or more of the plurality of similar objects in a graphical user interface;
wherein the plurality of first composite vectors and the plurality of second composite vectors are linear combinations of a plurality of vectors generated for each of the plurality of first and second objects.
9. The method of claim 8 , wherein:
identifying the plurality of first objects from the plurality of first video frames comprises:
generating a plurality of first masks, each first mask comprising neighboring pixels that are determined to be related; and
extracting the plurality of first objects from the plurality of first video frames based on the generated plurality of first masks; and
identifying the plurality of second objects from the plurality of second video frames comprises:
generating a plurality of second masks, each second mask comprising neighboring pixels that are determined to be related; and
extracting the plurality of second objects from the plurality of second video frames based on the generated plurality of second masks.
10. The method of claim 8 , wherein the plurality of vectors generated for each of the plurality of first and second objects comprises:
an appearance vector that is a mathematical representation of an appearance of a particular first or second object;
a behavior vector that is a mathematical representation of a behavior or movement of the particular first or second object; and
a shape vector that is a mathematical representation of a size or shape of the particular first or second object.
11. The method of claim 8 , further comprising displaying a user-editable label for the plurality of similar objects in the graphical user interface.
12. The method of claim 8 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises calculating a plurality of similarity scores for the plurality of second objects using the index.
13. The method of claim 12 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises:
accessing a user-selection of a selected second object of the plurality of second objects;
calculating similarity scores between the selected second object and other objects within the index; and
selecting the other objects within the index whose calculated similarity scores with the selected second object meets a predetermined similarity value.
14. The method of claim 12 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises automatically calculating similarity scores between each of the plurality of second objects and every other object within the index.
15. One or more computer-readable non-transitory storage media embodying instructions that, when executed by a processor, cause the processor to perform operations comprising:
access a plurality of first video frames of a first video captured at a first physical location;
identify a plurality of first objects from the plurality of first video frames;
generate a plurality of first composite vectors for the plurality of first objects;
store the plurality of first composite vectors in an index;
access a plurality of second video frames of a second video captured at a second physical location;
identify a plurality of second objects from the plurality of second video frames;
generate a plurality of second composite vectors for the plurality of second objects;
determine, using the index and the plurality of second composite vectors for the plurality of second objects, a plurality of similar objects; and
display images of one or more of the plurality of similar objects in a graphical user interface;
wherein the plurality of first composite vectors and the plurality of second composite vectors are linear combinations of a plurality of vectors generated for each of the plurality of first and second objects.
16. The one or more computer-readable non-transitory storage media of claim 15 , wherein:
identifying the plurality of first objects from the plurality of first video frames comprises:
generating a plurality of first masks, each first mask comprising neighboring pixels that are determined to be related; and
extracting the plurality of first objects from the plurality of first video frames based on the generated plurality of first masks; and
identifying the plurality of second objects from the plurality of second video frames comprises:
generating a plurality of second masks, each second mask comprising neighboring pixels that are determined to be related; and
extracting the plurality of second objects from the plurality of second video frames based on the generated plurality of second masks.
17. The one or more computer-readable non-transitory storage media of claim 15 , wherein the plurality of vectors generated for each of the plurality of first and second objects comprises:
an appearance vector that is a mathematical representation of an appearance of a particular first or second object;
a behavior vector that is a mathematical representation of a behavior or movement of the particular first or second object; and
a shape vector that is a mathematical representation of a size or shape of the particular first or second object.
18. The one or more computer-readable non-transitory storage media of claim 15 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises calculating a plurality of similarity scores for the plurality of second objects using the index.
19. The one or more computer-readable non-transitory storage media of claim 18 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises:
accessing a user-selection of a selected second object of the plurality of second objects;
calculating similarity scores between the selected second object and other objects within the index; and
selecting the other objects within the index whose calculated similarity scores with the selected second object meets a predetermined similarity value.
20. The one or more computer-readable non-transitory storage media of claim 18 , wherein determining, using the index and the plurality of second composite vectors for the plurality of second objects, the plurality of similar objects comprises automatically calculating similarity scores between each of the plurality of second objects and every other object within the index.