Method, apparatus, and storage medium for multi-target tracking

Abstract

The present invention provides a method, apparatus, and storage medium for multi-target tracking. [Solution] The method involves performing target whole-body detection to determine multiple target whole-body detection frames and determining multiple target head detection frames in the current input image, performing whole-body trajectory association to determine whole-body identifiers for the multiple target whole-body detection frames, performing head trajectory association to determine head identifiers for the multiple target head detection frames, determining multiple target whole-body prediction frames corresponding to the multiple target head detection frames based on the position and size of the multiple target head detection frames, determining target whole-body related frames among the multiple target whole-body detection frames based on the area of ​​the multiple target whole-body prediction frames corresponding to the multiple target head detection frames in the input image, and updating the target whole-body trajectory set based on the whole-body identifiers of the target whole-body related frames of the multiple target head detection frames.

Claims

1. A method for multi-target tracking, The steps include: performing target whole-body detection to determine multiple target whole-body detection frames in the current input image, and performing target head detection to determine multiple target head detection frames in the current input image; The steps include: performing whole-body trajectory association to determine the whole-body identifiers of the multiple target whole-body detection frames and updating the target whole-body trajectory set; The steps include: performing head trajectory association to determine the head identifiers of the plurality of target head detection frames and updating the target head trajectory set; The steps include determining a plurality of target whole-body prediction frames corresponding to the plurality of target head detection frames based on the positions and sizes of the plurality of target head detection frames, A step of determining a target whole-body related frame among the multiple target whole-body detection frames of the multiple target head detection frames based on the region of the input image of the multiple target whole-body prediction frames corresponding to the multiple target head detection frames, A method comprising the step of updating the target whole-body trajectory set based on the whole-body identifier of the target whole-body related frame of the plurality of target head detection frames.

2. The step of updating the target whole-body trajectory set based on the whole-body identifier of the target whole-body related frame of the plurality of target head detection frames is: The method according to claim 1, further comprising the step of, for each trajectory in the target head trajectory set, replacing the full-body identifier of the target full-body association frame at the current trajectory point of the trajectory with the full-body identifier of the target full-body association frame at the immediately preceding trajectory point of the trajectory.

3. The method according to claim 1, wherein a single target detection model is used to determine the plurality of whole-body target detection frames and the plurality of head target detection frames in the current input image.

4. The method according to claim 1, wherein determining the whole-body identifiers of the plurality of target whole-body detection frames and determining the head identifiers of the plurality of target head detection frames are performed based on a target tracking algorithm.

5. A first Kalman filter is used to determine the whole-body identifiers of the multiple target whole-body detection frames in the current input image. The method according to claim 4, wherein a second Kalman filter, different from the first Kalman filter, is used to determine the head identifiers of the plurality of target head detection frames in the current input image.

6. The step of determining a target whole-body prediction frame corresponding to one of the multiple target head detection frames is: The steps include determining the horizontal coordinate component of the target whole-body prediction frame by linearly combining the horizontal coordinate component and width of the position of the target head detection frame, The steps include determining the vertical coordinate component of the target whole-body prediction frame by linearly combining the vertical coordinate component of the position of the target head detection frame and its height, The steps include: expanding the width of the target head detection frame to determine the width of the target whole body prediction frame; The method according to claim 1, comprising the step of determining the height of the target whole body prediction frame by enlarging the height of the target head detection frame.

7. The step of determining the target whole-body related frame from among the plurality of target whole-body detection frames of the plurality of target head detection frames is: The steps include determining a cross-over union matrix based on the plurality of target whole-body prediction frames and the plurality of target whole-body detection frames, The step includes applying a Hungarian algorithm to the crossed overunion matrix to determine the target whole-body related frame corresponding to each target head detection frame, The method according to claim 1, wherein each element in the cross-overunion matrix is ​​a cross-overunion of a corresponding target whole-body detection frame from the plurality of target whole-body detection frames and a corresponding target whole-body prediction frame from the plurality of target whole-body prediction frames.

8. Performing the aforementioned whole-body trajectory association involves associating the current target whole-body detection frame with one target whole-body trajectory in the generated target whole-body trajectory set, based on the Hungarian algorithm. The method according to claim 1, wherein performing the head trajectory association includes associating the current target head detection frame with one target head trajectory in the generated set of target head trajectories, based on a Hungarian algorithm.

9. A device for multi-target tracking, The memory where the instructions are stored, An apparatus comprising: at least one processor configured to execute the instructions to realize the method according to any one of claims 1 to 8.

10. A computer-readable non-temporary storage medium on which a program is stored, wherein when the program is executed by the computer, the computer... The steps include: performing target whole-body detection to determine multiple target whole-body detection frames in the current input image, and performing target head detection to determine multiple target head detection frames in the current input image; The steps include: performing whole-body trajectory association to determine the whole-body identifiers of the multiple target whole-body detection frames and updating the target whole-body trajectory set; The steps include: performing head trajectory association to determine the head identifiers of the plurality of target head detection frames and updating the target head trajectory set; The steps include determining a plurality of target whole-body prediction frames corresponding to the plurality of target head detection frames based on the positions and sizes of the plurality of target head detection frames, A step of determining a target whole-body related frame among the multiple target whole-body detection frames of the multiple target head detection frames based on the region of the input image of the multiple target whole-body prediction frames corresponding to the multiple target head detection frames, A storage medium that causes the following steps to be performed: updating the target whole body trajectory set based on the whole body identifier of the target whole body related frame of the plurality of target head detection frames.

Images