Multiple target tracking method and apparatus, calculating device and storage medium
The present disclosure provides a multiple target tracking method and apparatus, a calculating device and a storage medium, so as to solve the problem of inaccurate multiple target tracking in the prior art. The target tracking method comprises: obtaining a prediction box of a target in a current frame according to tracklets of one or more targets in historical frames; performing target detection on the current frame to obtain one or more detection boxes, wherein the detection boxes comprise a high-quality box and a medium-quality box; matching each prediction box with the detection box according to the similarity of the prediction box and the detection box; and in response to a prediction box being unmatched with the high-quality box but matched with the medium-quality box, determining that the target is in a tracking state in the current frame.
1. A multiple target tracking method, comprising:
obtaining a prediction box of a target in a current frame according to a tracklet of the target in historical frames;
performing target detection on the current frame to obtain one or more detection boxes, wherein the detection boxes comprise a high-score box and a medium-score box;
in response to the prediction box being unmatched with the high-score box but matched with the medium-score box, determining that the target is in a tracking state in the current frame;
in response to the prediction box being matched with the medium-score box, maintaining state parameters of the tracklet unchanged; and
in response to the prediction box being matched with the high-score box, determining that the target is in the tracking state in the current frame, and updating the state parameters of the tracklet according to the high-score box.
2. The method according to claim 1 , wherein the tracklet has a target identifier, and the method further comprises:
for a matched high-score box and medium-score box which are matched by the prediction box in the current frame:
establishing an association relationship between the detection box and tracklet; and
outputting the target identifier of the tracklet for the detection box.
3. The method according to claim 2 , further comprising:
for a matched detection box which is matched by the prediction box, inputting information of the matched detection box, and information of a detection box in the tracklet associated with the matched detection box into a preset learning model to obtain an output probability of whether the matched detection box needs to be output, wherein the information comprises at least one of: state parameters of the matched detection box and a score of the matched detection box.
4. The method according to claim 3 , further comprising:
in response to the output probability of the matched detection box being greater than or equal to a preset threshold, outputting the state parameters of the matched detection box.
5. The method according to claim 1 , further comprising at least one of:
in response to the prediction box being unmatched with the high-score box and the prediction box being unmatched with the medium-score box, determining that the target is in a lost state in the current frame; or
in response to the target being in the lost state in continuous multiple frames, determining that the target is in a disappearing state and deleting the tracklet from a tracklet pool.
6. The method according to claim 1 , further comprises:
matching the prediction box with the detection boxes according to similarity of the prediction box and the detection boxes, which further comprises:
extracting the high-score box in the current frame to match with the tracklet for a first time; and
in response to the tracklet being unmatched with the high-score box, extracting the medium-score box from the current frame to match with the tracklet again.
7. The method according to claim 1 , wherein the method further comprises:
calculating an intersection volume according to an intersection area and an intersection height of a first projection box and a second projection box;
calculating the volume of an external convex polygon according to the area of the external convex polygon and a union height; and
calculating the intersection volume according to the volume of the prediction box, the volume of the detection box and the intersection volume.
8. The method according to claim 1 , further comprising:
calculating the state parameters of the target in a next frame according to the state parameters of the target in the current frame and a preset state transfer conversion relationship; and
calculating prediction box information of the target in the next frame according to the state parameters of the target in the next frame and a preset observation conversion relationship.
9. The method according to claim 8 , wherein the state parameters comprise a position and a speed, and calculating the state parameters of the target in the next frame according to the state parameters of the target in the current frame and the preset state transfer conversion relationship comprises:
calculating the state parameters of the target in the next frame according to the state transfer conversion relationship, the position and speed of the target in the current frame, and a time difference of two adjacent frames.
10. The method according to claim 1 , wherein the detection boxes further comprise a low-score box,
the high-score box is a detection box with a score value greater than or equal to a first threshold;
the medium-score box is a detection box with a score value between the first threshold and a second threshold;
the low-score box is a detection box with a score value less than the second threshold, and the second threshold is greater than 0; and
wherein the score is associated with at least one of confidence or accuracy.
11. A calculating device, comprising:
a processor, a memory, and a computer program stored in the memory and capable of running on the processor,
wherein when executing the computer program, the processor performs a multiple target tracking method comprising:
obtaining a prediction box of a target in a current frame according to a tracklet of the target in historical frames;
performing target detection on the current frame to obtain one or more detection boxes, wherein the detection boxes comprise a high-score box and a medium-score box;
in response to the prediction box being unmatched with the high-score box but matched with the medium-score box, determining that the target is in a tracking state in the current frame;
in response to the prediction box being matched with the medium-score box, maintaining state parameters of the tracklet unchanged; and
in response to the prediction box being matched with the high-score box, determining that the target is in the tracking state in the current frame, and updating the state parameters of the tracklet according to the high-score box.
12. The calculating device according to claim 11 , wherein the tracklet has a target identifier, and the method further comprises:
for a matched high-score box and medium-score box which are matched by the prediction box in the current frame:
establishing an association relationship between the detection box and the tracklet; and
outputting the target identifier of the tracklet for the detection box.
13. The calculating device according to claim 12 , wherein the method further comprises:
for a matched detection box which is matched by the prediction box, inputting information of the matched detection box, and information of a detection box in the tracklet into a preset learning model to obtain an output probability of whether the matched detection box needs to be output, wherein the information comprises the at least one of: state parameters of the detection box and a score of the detection box.
14. A non-transitory computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, a multiple target tracking method comprising:
obtaining a prediction box of a target in a current frame according to a tracklet of the target in historical frames;
performing target detection on the current frame to obtain one or more detection boxes, wherein the detection boxes comprise a high-score box and a medium-score box;
in response to the prediction box being unmatched with the high-score box but matched with the medium-score box, determining that the target is in a tracking state in the current frame;
in response to the prediction box being matched with the medium-score box, maintaining state parameters of the tracklet unchanged; and
in response to the prediction box being matched with the high-score box, determining that the target is in the tracking state in the current frame, and updating the state parameters of the tracklet according to the high-score box.
15. The non-transitory computer readable storage medium according to claim 14 , wherein the method further comprises:
in response to the prediction box being unmatched with the high-score box and the prediction box being unmatched with the medium-score box, determining that the target is in a lost state in the current frame; or
in response to the target being in the lost state in continuous multiple frames, determining that the target is in a disappearing state and deleting the tracklet from a tracklet pool.
16. The non-transitory computer readable storage medium according to claim 14 , wherein the method further comprises:
matching the prediction box with the detection boxes according to similarity of the prediction box and the detection boxes, which further comprises:
extracting the high-score box in the current frame to match with the tracklet for an first time; and
in response to the tracklet being unmatched with the high-score box, extracting the medium-score box from the current frame to match with the tracklet again.