CN110807494B

CN110807494B - Quick positioning method for repeated textures in industrial vision

Info

Publication number: CN110807494B
Application number: CN201911085199.1A
Authority: CN
Inventors: 许琦; 何志权; 何志海
Original assignee: Shenzhen Deepvision Creative Technology Ltd
Current assignee: Shenzhen Deepvision Creative Technology Ltd
Priority date: 2019-11-08
Filing date: 2019-11-08
Publication date: 2023-05-30
Anticipated expiration: 2039-11-08
Also published as: CN110807494A

Abstract

The invention provides a rapid positioning method of repeated textures in industrial vision, which comprises the steps of arranging a plurality of filter layers in cascade connection, wherein each filter layer filters the filtering result of the last filter layer again; obtaining a plurality of finally left discrete areas through layer-by-layer filtration; sliding a window in each discrete region to obtain the matching degree between the current picture block and the target picture block and reserving the position with the highest matching degree; and outputting the finally obtained multiple discrete positions as positioning results. The operator adopted by the filtering algorithm is self-adaptively selected and has high degree of distinction to the target, so that the system has good operation effect and wide practical application.

Description

Quick positioning method for repeated textures in industrial vision

Technical Field

The invention relates to the field of industrial vision image processing, in particular to a rapid positioning method of repeated textures in industrial vision.

Background

The industrial vision field often needs to locate a repeated texture block product (as shown in fig. 1), and the repeated texture block is mainly characterized by no changes in scaling, rotation, deformation, illumination and the like, and the main appearing fields are as follows: the LED bracket, the FPC circuit board, the cloth and the like contain repeated textures.

There are two conventional approaches: one is feature point matching, which requires computing features on the entire picture and then finding a location similar to the target block; the other is to slide a window on the whole picture to obtain the matching degree between the current picture block and the target picture block.

Both of these methods have the problem of being time consuming: the calculation process of the features is complex, the algorithms of the sift, surf, haaris corner points, the ORB points and other feature points are complex, and the images in the industrial vision field are large in size, so that a large amount of time is consumed for running the algorithm on the whole picture.

In summary, there is currently no efficient and effective method and system available, and the present invention proposes an algorithm and system for fast locating repeated texture blocks.

Disclosure of Invention

The invention provides a rapid positioning method for repeated textures in industrial vision, which aims to solve at least one technical problem.

In order to solve the above problems, as an aspect of the present invention, there is provided a rapid positioning method of repetitive textures in industrial vision, including providing a plurality of filter layers arranged in cascade, each filter layer filtering a filtering result of a previous filter layer again; obtaining a plurality of finally left discrete areas through layer-by-layer filtration; sliding a window in each discrete region to obtain the matching degree between the current picture block and the target picture block and reserving the position with the highest matching degree; and outputting the finally obtained multiple discrete positions as positioning results.

Preferably, the whole system uses a mask graph to represent the filtering result so as to facilitate the transmission of filtering information between upper and lower filter layers, each filter layer traverses the non-zero position in the mask graph, and the position which does not meet the threshold requirement is set to be zero in the mask graph according to the calculation result.

Preferably, before the filtering algorithm runs, a plurality of positions are randomly found in the picture, the similarity between the positions and the target small picture is calculated, the position with particularly high similarity is eliminated, then all candidate operators are calculated for each position in sequence, the discrimination degree of the result is calculated for each operator, all operators are ordered from high to low according to the discrimination degree, and the operators are sequentially arranged in a filter layer of the system.

Preferably, the method comprises the steps of:

step 1: reading in an image, if the image is the first time entering the system, executing the image sequentially, and if the image is not the first time entering the system, jumping to the step 4;

step 2: the user uses a mouse to draw a rectangular frame as a target and sets a filtering proportion parameter P%;

step 3: calculating N operators prepared in advance based on Haar features and contrast features between point pairs to obtain N groups of calculation results; sequencing each group of calculation results according to the principle of approaching a target value, finding out the Nth P% calculation result, and storing the Nth calculation result as a filtering threshold th of the operator; calculating scoring indexes of each group, sorting operators from high to low according to the scoring indexes, selecting M highest operators, and arranging the operators into a filter layer in sequence;

step 4: performing multi-layer filtering on the image, traversing non-zero positions in the mask graph for each filtering, operating an operator on each position, and setting the mask to be zero for positions which do not meet the threshold th requirement of the operator;

step 5: and filtering to obtain a plurality of discrete areas, solving the matching degree between the picture block at the current position and the target picture block in each discrete area by using a sliding window method, and reserving the position with the highest matching degree, so as to finally obtain a plurality of discrete positions as output results.

The operator adopted by the filtering algorithm is self-adaptively selected and has high degree of distinction to the target, so that the system has good operation effect and wide practical application.

Drawings

Fig. 1 schematically shows a texture map of an LED as is common in the prior art;

FIG. 2 schematically shows a schematic of the primary filtration results;

FIG. 3 schematically shows a schematic of the results of a medium-stage filtration;

FIG. 4 schematically shows a schematic of the final filtration result;

fig. 5 schematically shows an effect diagram of the final positioning.

Detailed Description

Embodiments of the invention are described in detail below with reference to the attached drawings, but the invention can be implemented in a number of different ways, which are defined and covered by the claims.

The invention provides a rapid positioning method of repeated textures in industrial vision, which is characterized in that a basic idea is cascade filtration, a system consists of a plurality of filter layers, each layer filters the filtering result of the previous layer again, only a plurality of discrete areas are remained finally after the filtering is carried out layer by layer, a sliding window in each discrete area is used for solving the matching degree between a current picture block and a target picture block and keeping the position with the highest matching degree, and a plurality of discrete positions are finally obtained as output results.

In order to facilitate the transmission of filtering information between upper and lower layers, the whole system uses a mask graph to represent the filtering result, each layer traverses the non-zero position in the mask graph, and simultaneously positions which do not meet the threshold value requirements are set to be zero according to the calculation result.

In order to improve the speed, instead of the complex feature operators, the Haar features and the contrast features between the point pairs are adopted, and the two features have a plurality of specific operators, for example, the Haar features have top and bottom, left and right, top and middle and bottom, left and middle and right and the like, and the filtering effects of the operators are different, so that before the filtering algorithm operates, a plurality of positions are randomly found in a picture, the similarity between the positions and a target small picture is calculated, the positions with particularly high similarity are eliminated, then all candidate operators are calculated for each position in sequence, the discrimination of the result is calculated for each operator, all operators are ordered from high to low according to the discrimination degree, and the operators are sequentially arranged in a filtering layer of the system.

Because the operators are adaptively selected and have high degree of distinction to the targets, the system has good operation effect and wide practical application.

The invention mainly comprises the following four steps:

step 3: n prepared operators (based on Haar characteristics and contrast characteristics between point pairs) are calculated and realized, N groups of calculation results are obtained, each group of calculation results are sequenced according to a principle of approaching a target value, the N & ltth & gt P & lt/th & gt calculation result is found, and the N & ltth & gt calculation result is stored as a filtering threshold th of the operator; calculating scoring indexes of each group, sorting operators from high to low according to the scoring indexes, selecting M highest operators, and arranging the operators into a filter layer in sequence;

The working process of the cascade filter in step 3 is shown in fig. 2, 3 and 4, and the final positioning effect in step 5 is shown in fig. 5.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A rapid positioning method for repeated textures in industrial vision is characterized by comprising the following steps of

Setting a plurality of filter layers in cascade connection, and filtering the filtering result of the last filter layer again by each filter layer;

obtaining a plurality of finally left discrete areas through layer-by-layer filtration;

sliding a window in each discrete region to obtain the matching degree between the current picture block and the target picture block and reserving the position with the highest matching degree;

outputting the finally obtained multiple discrete positions as positioning results;

the method comprises the following steps:

2. The method of claim 1, wherein the overall system uses a mask map to represent the filtering results to facilitate the transfer of filtering information between upper and lower filter layers, each filter layer traversing non-zero positions in the mask map, and setting non-threshold positions in the mask map to zero according to the calculation results.

3. A method according to claim 1, characterized in that before the filtering algorithm is run, a number of positions are first found randomly in the picture and the similarity with the target small picture is calculated, positions with a particularly high similarity are excluded, then for each position all candidate operators are calculated in turn, for each operator the resulting degree of discrimination is calculated, all operators are ordered from high to low in the degree of discrimination, and these operators are arranged in turn into the filter layer of the system.