Evaluation of models generated from objects in video
View Patent ↗Models are generated from objects identified in video. Each model is evaluated based on knowledge of the objects determined from video analysis, and preferred models are identified based on the evaluations. In some examples, each model could be evaluated by tracking a movement of each object in the video by using each model to track the object from which it was generated, evaluating an ability of each model to identify the objects in the video that are similar to the object from which it was generated, and determining an amount of false identifications made by each model of different objects in different video that does not include the object from which it was generated.
1. A method of operating an image processing system, the method comprising:
generating a plurality of respective models from identified objects in a first video, wherein said plurality of respective models include descriptors of parts of said identified objects;
evaluating an ability of each of the plurality of respective models to identify other objects in a second video when said other objects are similar to a respective identified object from which said plurality of respective models was generated; and
identifying at least one preferred model from the plurality of respective models based on the evaluating step.
2. The method of claim 1 wherein evaluating an ability of each respective model comprises evaluating knowledge of the objects determined from video analysis of tracked movement of said respective identified object in the second video.
3. The method of claim 1 wherein evaluating an ability of each respective model comprises evaluating knowledge of the objects comprising a number of false identifications made by each model of different objects in different video that does not include the respective identified object from which the respective model was generated.
4. The method of claim 3 wherein the respective models that have a high rate of false identifications are removed from being identified as the at least one preferred model.
5. The method of claim 4 wherein the models determined to have a high rate of false identifications are those ranking in the top 5% of false identifications from the plurality of respective models.
6. The method of claim 1 wherein evaluating each model comprises evaluating knowledge of the identified objects determined from a video analysis that comprises:
tracking a movement in the video by using each respective model to track the respective identified object from which it was generated; and
determining an amount of false identifications made by each respective model of different objects in the first video that may include the object from which it was generated.
7. The method of claim 1 wherein identifying at least one preferred model based on the evaluating comprises identifying one of the respective models having a greatest ability to identify the other objects in the second video that are similar to the respective identified object from which the one of the respective models was generated and having a least amount of false identifications of different objects in the first video.
8. The method of claim 1 wherein the identified objects are identified in the first video by manual identification.
9. The method of claim 1 wherein the identified objects are identified in the first video based on movement detection.
10. The method of claim 1 wherein the identified objects are identified in the first video by human head detection.
11. The method of claim 10 wherein the identified objects in the first video comprise human heads.
12. The method of claim 11 wherein a different model is generated for each different identified human head.
13. The method of claim 1 wherein the identified objects identified in the first video comprise human body parts.
14. A method of image processing, comprising:
generating a plurality of respective models from identified objects in a first video;
evaluating an ability of each of the plurality of respective models to identify other objects from a second video when said other objects are similar to a respective identified object from which said respective model was generated and distinguish different objects from the first video when said first video includes said other objects that are similar; and
identifying at least one preferred model from the plurality of respective models based on the evaluating step.
15. The method of claim 14 further comprising performing the evaluations on each respective model based on knowledge of the respective identified objects determined from video analysis and further tracking a movement of each respective identified object in the first video.
16. The method of claim 15 further comprising performing the evaluations on each respective model based on knowledge of the identified objects determined from video analysis and evaluating an ability of each respective model to identify, from the portions of the second video, other objects in the second video that are similar to the identified object from which the respective model was generated.
17. The method of claim 15 further comprising performing the evaluations on each respective model based on knowledge of the objects determined from video analysis and determining an amount of false identifications made by each respective model of said different objects in the first video that includes the identified object from which the respective model was generated.
18. The method of claim 17 wherein the respective models with a high rate of false identifications are removed from being identified as the at least one preferred model.
19. The method of claim 18 wherein the models determined to have a high rate of false identifications are those ranking in the top 5% of false identifications from the plurality of respective models.
20. The method of claim 14 wherein the identified objects are identified in the first video by manual identification.
21. The method of claim 14 wherein the identified objects are identified in the first video based on movement detection, and wherein the identified objects in the first video comprise objects associated with the identified movement.
22. The method of claim 14 wherein the identified objects are identified in the first video by human head detection, and wherein the identified objects in the first video comprise human heads, further wherein a different model is generated for each different identified human head.