Object detection dataset construction method using image entropy and data processing device performing the same

1. A data processing device, comprising:

a processor, configured to:

receive multiple images;

choose processing priorities of the received multiple images using image entropy and construct an object detection dataset from a corresponding image according to the chosen processing priorities as a constructed object detection dataset;

store the constructed object detection dataset;

perform an inspection to determine whether to re-process the constructed object detection dataset using image entropy;

determine that the constructed object detection dataset is a final object detection dataset responsive to the constructed object detection dataset passing the inspection;

erase at least one object from an image of the constructed object detection dataset using object information;

compute image entropy of the image from which at least one object is erased; and

determine an image to be re-processed using the computed image entropy.

2. The data processing device of claim 1 , wherein the processor is further configured to:

compute image entropy of each image using an image segmentation algorithm or an object detection model; and

choose a predetermined number of images to be preferentially processed in descending order of the computed image entropy or when the computed image entropy is greater than or equal to a preset reference value.

3. The data processing device of claim 2 , wherein the image entropy computation using the image segmentation algorithm comprises:

segmenting each image into at least one color appearing in the corresponding image; and

computing image entropy by dividing the number of segmentation color types by preset color sensitivity.

4. The data processing device of claim 2 , wherein the image entropy computation using the object detection model comprises:

inputting a predetermined image to the object detection model;

detecting objects through the object detection model; and

computing image entropy by adding scores of the detected objects.

5. The data processing device of claim 1 , wherein the processor is further configured to:

extract multiple frame images from the received multiple images;

classify the extracted multiple frame images into section images of preset sections;

compute average image entropy of each section from the section images of a corresponding section; and

choose a predetermined number of section images to be preferentially processed in descending order of the computed average image entropy or when the computed average image entropy is greater than or equal to a preset reference value.

6. The data processing device of claim 1 , wherein, when image entropy is computed using an image segmentation algorithm, the processor is further configured to erase at least one object through filling with ambient color or filling with a surrounding image, and

wherein when image entropy is computed using an object detection model, the processor is further configured to erase at least one object through one of a black-out, a white-out, or a filling with an average color of a corresponding area.

7. The data processing device of claim 1 , wherein, when the image entropy of the image from which at least one object is erased is greater than or equal to a preset reference value, the processor is further configured to determine whether the corresponding image is to be re-processed, and

wherein the preset reference value varies depending on a situation.

8. A method of constructing an object detection dataset using a data processing device, the method comprising:

receiving multiple images;

choosing processing priorities of the multiple images using image entropy;

constructing the object detection dataset from a corresponding image according to the chosen processing priorities as a constructed object detection dataset;

performing an inspection to determine whether to re-process the constructed object dataset using image entropy, wherein the performing of the inspection comprises:

erasing at least one object from an image of the object detection dataset using object information;

computing image entropy of the image from which at least one object is erased; and

determining an image to be re-processed using the computed image entropy;

determining that the constructed object detection dataset is a final object detection dataset responsive to a passing of the inspection;

re-processing the constructed object detection dataset responsive to a failing of the inspection to generate a reprocessed object detection dataset;

performing the inspection on the reprocessed object detection dataset; and

determining that the reprocessed object detection dataset is the final object detection dataset responsive to the reprocessed object detection dataset passing the inspection.

9. The method of claim 8 , wherein the choosing of the processing priority comprises:

computing image entropy of each image using an image segmentation algorithm or an object detection model; and

choosing a predetermined number of images to be preferentially processed in descending order of the computed image entropy or when the computed image entropy is greater than or equal to a preset reference value.

10. The method of claim 8 , wherein the choosing of the processing priority comprises:

extracting multiple frame images from the multiple images;

classifying the extracted multiple frame images into section images of preset sections;

computing average image entropy of each section from the section images of a corresponding section; and

choosing a predetermined number of section images to be preferentially processed in descending order of the computed average image entropy or when the computed average image entropy is greater than or equal to a preset reference value.

11. The method of claim 8 , wherein the determining of an image to be re-processed comprises determining that a corresponding image is to be re-processed when the image entropy of the image from which at least one object is erased is greater than or equal to a preset reference value, and

wherein the preset reference value varies depending on a situation.