Object tracking method and apparatus

1. An object tracking method comprising:

generating a feature map of a search image and generating a feature map of a target image;

obtaining an object classification result and a basic bounding box based on the feature map of the search image and the feature map of the target image;

obtaining an auxiliary bounding box based on the feature map of the search image;

obtaining a final bounding box based on the basic bounding box and the auxiliary bounding box; and

tracking an object based on the object classification result and the final bounding box.

2. The object tracking method of claim 1 , wherein the obtaining of the object classification result and the basic bounding box comprises:

transforming the feature map of the target image to a first feature map for object classification and a second feature map for localization, using a depthwise convolution;

transforming the feature map of the search image to a third feature map for object classification and a fourth feature map for localization, using a depthwise convolution;

obtaining the object classification result based on the first feature map and the third feature map; and

obtaining the basic bounding box by performing bounding box regression based on the second feature map and the fourth feature map.

3. The object tracking method of claim 2 , wherein the transforming of the feature map of the target image to the first feature map for object classification and the second feature map for localization comprises:

performing a convolution operation on the feature map of the target image;

performing a first depthwise convolution operation and a second depthwise convolution operation on a result of the convolution operation;

generating the first feature map for object classification by concatenating the result of the convolution operation and a result of the first depthwise convolution operation; and

generating the second feature map for localization by concatenating the result of the convolution operation and a result of the second depthwise convolution operation.

4. The object tracking method of claim 2 , wherein the transforming of the feature map of the search image to the third feature map for object classification and the fourth feature map for localization comprises:

performing a convolution operation on the feature map of the search image;

performing a first depthwise convolution operation and a second depthwise convolution operation on a result of the convolution operation;

generating the third feature map for object classification by concatenating the result of the convolution operation and a result of the first depthwise convolution operation; and

generating the fourth feature map for localization by concatenating the result of the convolution operation and a result of the second depthwise convolution operation.

5. The object tracking method of claim 1 , wherein the obtaining of the auxiliary bounding box comprises:

performing a convolution operation on the feature map of the search image;

performing a first depthwise convolution operation on a result of the convolution operation;

performing a second depthwise convolution operation on a concatenation of the result of the convolution operation and a result of the first depthwise convolution operation; and

obtaining the auxiliary bounding box based on a result of the second depthwise convolution operation.

6. The object tracking method of claim 5 , wherein the obtaining of the auxiliary bounding box comprises:

performing a first convolution operation on the result of the second depthwise convolution operation; and

performing a second convolution operation on a result of the first convolution operation.

7. The object tracking method of claim 1 , wherein the obtaining of the auxiliary bounding box comprises:

obtaining the auxiliary bounding box based on a category of an object detected from a previous frame.

8. The object tracking method of claim 1 , wherein the obtaining of the final bounding box comprises:

obtaining the final bounding box based on a weighted sum of the basic bounding box and the auxiliary bounding box.

9. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the object tracking method of claim 1 .

10. An object tracking apparatus comprising:

a memory comprising instructions; and

a processor configured to execute the instructions,

wherein, when the instructions are executed by the processor, the processor is configured to:

generate a feature map of a search image and generate a feature map of a target image;

obtain an object classification result and a basic bounding box based on the feature map of the search image and the feature map of the target image;

obtain an auxiliary bounding box based on the feature map of the search image;

obtain a final bounding box based on the basic bounding box and the auxiliary bounding box; and

track an object based on the object classification result and the final bounding box.

11. The object tracking apparatus of claim 10 , wherein the processor is configured to:

transform the feature map of the target image to a first feature map for object classification and a second feature map for localization, using a depthwise convolution;

transform the feature map of the search image to a third feature map for object classification and a fourth feature map for localization, using a depthwise convolution;

obtain the object classification result based on the first feature map and the third feature map; and

obtain the basic bounding box by performing the bounding box regression based on the second feature map and the fourth feature map.

12. The object tracking apparatus of claim 11 , wherein the processor is configured to:

perform a convolution operation on the feature map of the target image;

perform a first depthwise convolution operation and a second depthwise convolution operation on a result of the convolution operation;

generate the first feature map for object classification by concatenating the result of the convolution operation and a result of the first depthwise convolution operation; and

generate the second feature map for localization by concatenating the result of the convolution operation and a result of the second depthwise convolution operation.

13. The object tracking apparatus of claim 11 , wherein the processor is configured to:

perform a convolution operation on the feature map of the search image;

perform a first depthwise convolution operation and a second depthwise convolution operation on a result of the convolution operation;

generate the third feature map for object classification by concatenating the result of the convolution operation and a result of the first depthwise convolution operation; and

generate the fourth feature map for localization by concatenating the result of the convolution operation and a result of the second depthwise convolution operation.

14. The object tracking apparatus of claim 10 , wherein the processor is configured to:

perform a convolution operation on the feature map of the search image;

perform a first depthwise convolution operation on a result of the convolution operation;

perform a second depthwise convolution operation on a concatenation of the result of the convolution operation and a result of the first depthwise convolution operation; and

obtain the auxiliary bounding box based on a result of the second depthwise convolution operation.

15. The object tracking apparatus of claim 13 , wherein the processor is configured to:

perform a first convolution operation on the result of the second depthwise convolution operation; and

perform a second convolution operation on a result of the first convolution operation.

16. The object tracking apparatus of claim 10 , wherein the processor is configured to:

obtain the auxiliary bounding box based on a category of an object detected from a previous frame.

17. The object tracking apparatus of claim 10 , wherein the processor is configured to:

obtain the final bounding box based on a weighted sum of the basic bounding box and the auxiliary bounding box.