Target detection method, apparatus and device for continuous images, and storage medium
A method, an apparatus, and a device for target detection in consecutive images, and a computer-readable storage medium. A second frame is divided into multiple sub-images, before a target in the second frame in a video sequence is detected through a target-detecting network model. A first frame is searched, according to a preset rule for motion estimation, for a corresponding image block matched with each sub-image. Pixels of a sub-image, of which the matched image block is found in the first frame, are replaced with preset background pixels. Hence, a target repeating in both frames is replaced. Finally, the second frame subject to the replacement is inputted in to the target-detecting network model, to obtain a bounding box of a target object of the second frame and a category of such target object. An algorithm for target detection in consecutive images is optimized.
1. A method for target detection in consecutive images, comprising:
inputting a first frame in a video sequence into a target-detecting network model, to obtain a bounding box of a target object of the first frame and a category of the target object of the first frame;
dividing a second frame in the video sequence into a plurality of first sub-images;
searching, according to a preset rule for motion estimation, the first frame for an image block matching with each of the plurality of first sub-images, to determine a position of the target object in the second frame;
replacing pixels of a first sub-image in the plurality of first sub-images with preset background pixels, wherein the image block matched with the first sub-image is found in the first frame; and
inputting the second frame, in which the pixels of the first sub-image are replaced, into the target-detecting network model, to obtain a bounding box of a target object of the second frame and a category of the target object of the second frame;
wherein the second frame is subsequent and adjacent to the first frame.
2. The method according to claim 1 , wherein the target-detecting network model is a YOLO-v3 network model or a SSD (Single Shot Multibox Detector) network model.
3. The method according to claim 2 , wherein after inputting the second frame in which the pixels of the first sub-image are replaced into the target-detecting network model, the method further comprises:
dividing a third frame in the video sequence into a plurality of second sub-images;
searching, according to the preset rule for motion estimation, the first frame and the second frame for image blocks matching with each of the plurality of second sub-images, to determine a position of the target object in the third frame;
replacing pixels of a second sub-image in the plurality of second sub-images with the preset background pixels, wherein the image blocks matched with the second sub-image are found in the first frame and the second frame; and
inputting the third frame, in which the pixels of the second sub-image are replaced, into the target-detecting network model, to obtain a bounding box of a target object of the third frame and a category of the target object of the third frame;
wherein the third frame is subsequent and adjacent to the second frame.
4. The method according to claim 1 , wherein before dividing the second frame in the video sequence into the plurality of first sub-images, the method further comprises:
denoising the second frame in the video sequence that is acquired, to remove noise interference in the second frame.
5. A device for target detection in consecutive images, comprising:
a memory, storing a computer program;
a processor, configured to implement the method according to claim 1 , when executing the computer program.
6. A computer-readable storage medium, storing a program for target detection in consecutive images, wherein:
the program when executed by a processor implements the method according to claim 1 .
7. An apparatus for target detection in consecutive images, comprising:
a first-frame inputting module, configured to input a first frame in a video sequence into a target-detecting network model, to obtain a bounding box of a target object of the first frame and a category of the target object of the first frame;
an image matching module, configured to:
divide a second frame in the video sequence into a plurality of first sub-images, and
search, according to a preset rule for motion estimation, the first frame for an image block matching with each of the plurality of first sub-images, to determine a position of the target object in the second frame,
wherein the second frame is subsequent and adjacent to the first frame;
a background replacing module, configured to replace pixels of a first sub-image in the plurality of first sub-images with preset background pixels, wherein the image block matched with the first sub-image is found in the first frame; and
a second-frame inputting module, configured to input the second frame, in which the pixels of the first sub-image are replaced, into the target-detecting network model, to obtain a bounding box of a target object of the second frame and a category of the target object of the second frame.
8. The apparatus according to claim 7 , wherein the target-detecting network model is a YOLO-v3 network model or a SSD (Single Shot Multibox Detector) network model.
9. The apparatus according to claim 7 , further comprising a third-frame processing module, wherein the third-frame processing module comprises:
a previous-frame matching sub-module, configured to
divide a third frame in the video sequence into a plurality of second sub-images, and
search, according to the preset rule for motion estimation, the first frame and the second frame for image blocks matching with each of the plurality of second sub-images, to determine a position of the target object in the third frame,
wherein the third frame is subsequent and adjacent to the second frame;
a repeated-target replacing sub-module, configured to replace pixels of a second sub-image in the plurality of second sub-images with the preset background pixels, wherein the image blocks matched with the second sub-image are found in the first frame and the second frame; and
a third-frame inputting sub-module, configured to input the third frame, in which the pixels of the second sub-image are replaced, into the target-detecting network model, to obtain a bounding box of a target object of the third frame and a category of the target object of the third frame.
10. The apparatus according to claim 7 , further comprising:
a denoising module, configured to denoise the second frame and the third frame in the video sequence, to remove noise interference in the second frame and the third frame.