Method for improving detection efficiency of detected target
Technical Field
The invention relates to a computer vision technology, in particular to a method for improving detection efficiency of a detected target.
Background
Object detection and recognition is a very important research direction in the field of computer vision, and is to distinguish an object from an uninteresting part in an image, judge whether the object exists, and determine the position of the object if the object exists, and the recognition of the object is a computer vision task. With the rapid development of the internet, artificial intelligence technology and intelligent hardware, a large amount of image data exists in human life, so that the computer vision technology plays an increasingly greater role in human life, and the research on computer vision is more and more intense.
In practical applications, the computational power of the target detection and recognition calculation is constant, that is, the number of input continuous image sequence sets that can be processed in a unit time is constant. The calculation for target detection and identification usually needs the hardware support of a display card, and the price is generally expensive. This limits enterprise-level applications of target detection and identification to some extent. How to improve the processing efficiency of the detection algorithm and reduce the cost of the product becomes a problem to be solved at present.
In view of this, the present disclosure provides a method for improving the efficiency of detecting a target. Global detection statistical learning is carried out on the detected and identified target data to obtain the region range where the target is likely to appear; cutting the picture to be detected and identified according to the area, reducing a plurality of cut pictures, and splicing into a large picture; and finally, inputting the spliced large image into an algorithm model for target detection and identification. The effect of detecting a plurality of pictures at one time is achieved, and therefore the processing efficiency of the detection algorithm is improved. And finally, mapping the detection result back to the corresponding original image, extracting the target detected in the original image and then carrying out target identification. The method aims to adapt to the target detection and recognition algorithm model and improve the single processing efficiency of the algorithm model.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a method for improving the detection efficiency of a detection target, aiming at the defects in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a method for improving detection efficiency of a detected target comprises the following steps:
1) taking a scene as a unit, carrying out statistical learning on the detected and identified target data in a machine learning mode, and training a region range in which a target possibly appears in the scene; the scene is a fixed scene;
2) identifying the scene of the picture to be detected, and cutting the picture to be detected according to the identified scene result and the area range learned in the step 1);
3) reducing and numbering the cut pictures, and then splicing a plurality of reduced pictures; processing the pictures according to the serial number sequence during picture splicing;
4) sending the spliced pictures to a target detection algorithm model for target detection;
5) a plurality of target results obtained by target detection are reversely transformed one by one and mapped to corresponding original images;
6) and cutting the target information corresponding to the cut picture, and performing subsequent target identification.
According to the scheme, the reduction ratio in the step 3) is set as follows: the reduced proportion is less than or equal to the minimum resolution ratio of the detectable target in the target detection and recognition algorithm model/the minimum resolution ratio of the recognizable target. The optimum value is the minimum value of the reduction ratio, i.e., the minimum resolution at which the target can be detected/the minimum resolution at which the target can be recognized.
According to the scheme, the resolution of the picture spliced in the step 3) is not greater than the resolution of the input picture of the target detection algorithm model.
The invention has the following beneficial effects: the method comprises the steps of carrying out global detection statistical learning on target data which are detected and identified to obtain a region range in which a target is likely to appear; cutting the picture to be detected and identified according to the area, reducing a plurality of cut pictures, and splicing into a large picture; and finally, inputting the spliced large image into an algorithm model for target detection and identification. The method and the device achieve the purpose of detecting multiple pictures at one time and improve the processing efficiency of the detection algorithm.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method of an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, a method for improving detection efficiency of a detection target includes the following steps:
step 1: and taking a scene as a unit, and performing statistical learning on the detected and identified target data in a machine learning mode to train out the region range in which the target may appear in the scene.
Step 2: and (4) cutting the picture to be detected according to the area range learned in the step (1), and then reducing. The reduction ratio is as follows: the minimum resolution at which the target can be detected/identified (in the target detection and identification algorithm model). The basis is as follows: the pixel requirement for the target to be identified is higher than the pixel requirement for the target to be detected (the minimum resolution requirement for the target to be identified is 70x70, and the minimum resolution requirement for the target to be detected is 35x 35), usually by a ratio of several times. And the detected target can be identified to be valid target data, and only the target which cannot be identified can be detected to belong to invalid data.
And step 3: and (3) generating a plurality of pictures processed in the step (2), and splicing the pictures. The resolution of the spliced picture is not greater than the resolution of the input picture of the target detection algorithm model (if greater than this, the target detection algorithm will shrink the picture, possibly resulting in that the target data cannot be normally detected). And then the target is sent to a target detection algorithm model for target detection.
And 4, step 4: and (4) performing reverse transformation on the plurality of target results detected in the step (3) one by one, and mapping the target results to corresponding original drawings. Corresponding to the corresponding reduced picture; then, scaling up the picture to correspond to the cut picture; and then cutting the target information corresponding to the cut picture, and performing subsequent target identification.
One example of the inventive method for re-identification of a person:
step S101) determines a detection area by means of global detection of the past mass data.
Step S102) cutting the original picture based on the area determined in the step S101) to obtain a cut picture; and reducing the cut picture by a reduction ratio of 4 to obtain a reduced picture.
The minimum resolution for detecting the target comes from the target detection model;
the minimum resolution capable of identifying the target is the minimum resolution capable of identifying the target and is from the target identification model;
step S103) splicing the reduced pictures generated in the step S102) to obtain spliced pictures. And then the target is sent to a target detection model for target detection.
Step S104) finding the corresponding (generated in step S102) reduced pictures one by one according to the plurality of target results detected in step S103); and then, the image is enlarged in scale and is generated corresponding to the (step S101). So as to carry out face recognition on the detected face area in the original image in the following process.
And (5) comparing the detection result obtained by performing face detection on the spliced picture in the step S104) with the detection result obtained by performing face detection identification on the original picture, and finding that the number and the area position of the faces of the spliced picture and the original picture are consistent. The purpose of returning the detection result of 4 pictures in one detection can be achieved.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.