Granted Patent B2

US 12,432,428 · App. 18/644,707 · Granted Sep 30, 2025

Inventors: Ariel Naim (Modi'in, IL); Igal Dvir (Modi'in, IL); Shmuel Peleg (Modi'in, IL)

Assignee: BRIEFCAM LTD.

H04N21/8549G06T7/194G06T2207/20081

Abstract

Claims (56)

1. A method of generating an interaction-aware video synopsis, the method comprising:

obtaining, using a computer processor, a source video containing a plurality of source objects;

extracting, using the computer processor, the source objects from the source video;

detecting, using the computer processor, at least one object-interaction between at least two source objects, wherein said object interaction comprises an interaction between the at least two objects in the scene which is visible in the source video;

generating, using the computer processor, synopsis objects by sampling respective source objects; and

generating, using the computer processor, a synopsis video having an overall play time shorter than the overall play time of the source video, by determining a play time for each one of the synopsis objects,

wherein at least two synopsis objects which are played at least partially simultaneously in the synopsis video, are generated from source objects that are captured at different times in the source video, and

maintaining a relative display time in the synopsis video of at least two of the synopsis objects created from the at least two source objects, responsive to the detecting of the at least one object-interaction between the at least two source objects, so the relative display time is the same as the relative display time of the at least two source objects in the source video,

wherein the maintaining of the relative display time in the synopsis video of the at least one object-interaction between the at least two source objects is achieved by either:

a) representing multiple interacting object tubes as if they are a single combined tube, and whenever a new appearance time is determined for the single combined tube, all interacting tubes therein maintain an original relative time; or

b) marking interacting objects and constraining the generating of the video synopsis, by keeping the relative time of iterating objects also in the generated video synopsis.

2. The method according to claim 1 , wherein the extracting of the source objects from the source video is carried out by training a machine learning model to distinguish between foreground objects and a background object.

3. The method according to claim 1 , wherein the detection of the at least one object-interaction between at least two source objects is carried out by training a machine learning model to detect object interaction.

4. The method according to claim 1 , wherein each one of the source objects is represented by a respective tube and wherein the method further comprising merging the respective tubes for the at least two source objects, detected as having an object-interaction therebetween into a single tube.

5. The method according to claim 4 , further comprising using the single tube to guarantee that the relative display time in the synopsis video of the at least two synopsis objects created from the at least two source objects, detected as having an object-interaction therebetween is the same as the relative display time of the at least two source objects in the source video.

6. The method according to claim 1 , wherein each one of the source objects is represented by a respective tube and wherein the method further comprising tagging the respective tubes for the at least two source objects, detected as having an object-interaction therebetween, as associated with a same object interaction.

7. The method according to claim 6 , further comprising using the tagging to guarantee that the relative display time in the synopsis video of the at least two synopsis objects created from the at least two source objects, detected as having an object-interaction therebetween is the same as the relative display time of the at least two source objects in the source video.

8. A non-transitory computer readable medium for generating an interaction-aware video synopsis, the computer readable medium comprising a set of instructions that, when executed, cause at least one computer processor to:

obtain a source video containing a plurality of source objects;

extract the source objects from the source video;

detect at least one object-interaction between at least two source objects wherein said object interaction comprises an interaction between the at least two objects in the scene which is visible in the source video;

generate synopsis objects by sampling respective source objects; and

generate a synopsis video having an overall play time shorter than the overall play time of the source video, by determining a play time for each one of the synopsis objects,

wherein at least two synopsis objects which are played at least partially simultaneously in the synopsis video, are generated from source objects that are captured at different times in the source video, and

maintain a relative display time in the synopsis video of at least two of the synopsis objects created from the at least two source objects, responsive to the detecting of the at least one object-interaction between the at least two source objects, so the relative display time is the same as the relative display time of the at least two source objects in the source video,

wherein the maintaining of the relative display time in the synopsis video of the at least one object-interaction between the at least two source objects is achieved by either:

a) representing multiple interacting object tubes as if they are a single combined tube, and whenever a new appearance time is determined for the single combined tube, all interacting tubes therein maintain an original relative time; or

b) marking interacting objects and constraining the generating of the video synopsis, by keeping the relative time of iterating objects also in the generated video synopsis.

9. The non-transitory computer readable medium according to claim 8 , wherein the extracting of the source objects from the source video is carried out by training a machine learning model to distinguish between foreground objects and a background object.

10. The non-transitory computer readable medium according to claim 8 , wherein the detection of the at least one object-interaction between at least two source objects is carried out by training a machine learning model to detect object interaction.

11. The non-transitory computer readable medium according to claim 8 , wherein each one of the source objects is represented by a respective tube and wherein the computer readable medium further comprising a set of instructions that, when executed, cause the at least one computer processor to merge the respective tubes for the at least two source objects, detected as having an object-interaction therebetween into a single tube.

12. The non-transitory computer readable medium according to claim 11 , and wherein the computer readable medium further comprising a set of instructions that, when executed, cause the at least one computer processor to use the single tube to guarantee that the relative display time in the synopsis video of the at least two synopsis objects created from the at least two source objects, detected as having an object-interaction therebetween is the same as the relative display time of the at least two source objects in the source video.

13. The non-transitory computer readable medium according to claim 8 , wherein each one of the source objects is represented by a respective tube and wherein the computer readable medium further comprising a set of instructions that, when executed, cause the at least one computer processor to tag the respective tubes with a tag for the at least two source objects, detected as having an object-interaction therebetween, as associated with a same object interaction.

14. The non-transitory computer readable medium according to claim 13 , and wherein the computer readable medium further comprising a set of instructions that, when executed, cause the at least one computer processor to use the tag to guarantee that the relative display time in the synopsis video of the at least two synopsis objects created from the at least two source objects, detected as having an object-interaction therebetween is the same as the relative display time of the at least two source objects in the source video.

15. A system for generating an interaction-aware video synopsis, the system comprising:

a server comprising:

a processing device;

a memory device; and

an interface for communicating with a video camera,

wherein the memory device comprising a set of instructions that, when executed, cause at the processing device to:

obtain from the video camera a source video containing a plurality of source objects;

extract the source objects from the source video;

detect at least one object-interaction between at least two source objects, wherein said object interaction comprises an interaction between the at least two objects in the scene which is visible in the source video;

generate synopsis objects by sampling respective source objects; and

generate a synopsis video having an overall play time shorter than the overall play time of the source video, by determining a play time for each one of the synopsis objects,

wherein at least two synopsis objects which are played at least partially simultaneously in the synopsis video, are generated from source objects that are captured at different times in the source video, and

maintain a relative display time in the synopsis video of at least two of the synopsis objects created from the at least two source objects, responsive to the detecting of the at least one object-interaction between the at least two source objects, so the relative display time is the same as the relative display time of the at least two source objects in the source video,

wherein the maintaining of the relative display time in the synopsis video of the at least one object-interaction between the at least two source objects is achieved by either:

a) representing multiple interacting object tubes as if they are a single combined tube, and whenever a new appearance time is determined for the single combined tube, all interacting tubes therein maintain an original relative time; or

b) marking interacting objects and constraining the generating of the video synopsis, by keeping the relative time of iterating objects also in the generated video synopsis.

16. The system according to claim 15 , wherein the extracting of the source objects from the source video is carried out by training a machine learning model to distinguish between foreground objects and a background object.

17. The system according to claim 15 , wherein the detection of the at least one object-interaction between at least two source objects is carried out by training a machine learning model to detect object interaction.

18. The system according to claim 15 , wherein each one of the source objects is represented by a respective tube and wherein the memory device further comprising a set of instructions that, when executed, cause the processing device to merge the respective tubes for the at least two source objects, detected as having an object-interaction therebetween into a single tube.

19. The system according to claim 18 , and wherein the memory device further comprising a set of instructions that, when executed, cause the processing device to use the single tube to guarantee that the relative display time in the synopsis video of the at least two synopsis objects created from the at least two source objects, detected as having an object-interaction therebetween is the same as the relative display time of the at least two source objects in the source video.

20. The system according to claim 15 , wherein each one of the source objects is represented by a respective tube and wherein the memory device further comprising a set of instructions that, when executed, cause the processing device to tag the respective tubes with a tag for the at least two source objects, detected as having an object-interaction therebetween, as associated with a same object interaction.

21. The system according to claim 20 , and wherein the memory device further comprising a set of instructions that, when executed, cause the processing device to use the tag to guarantee that the relative display time in the synopsis video of the at least two synopsis objects created from the at least two source objects, detected as having an object-interaction therebetween is the same as the relative display time of the at least two source objects in the source video.

Assignments (1)

Continuity (2)

References Cited (27)