Clothes sorting system based on computer vision

Abstract

The invention discloses a clothing sorting system based on computer vision, which belongs to the technical field of clothing identification, and specifically comprises the following steps: acquiring clothing specification information of a target position, acquiring a monitoring image of the target position at each acquisition point, and generating a front monitoring image sequence and an upper monitoring image sequence; analyzing the front monitoring image sequence, generating a positioning frame of any piece of clothes from the front monitoring image, acquiring positioning frame track information of the clothes, and selecting a corresponding monitoring image with the minimum interaction ratio of the positioning frame of the clothes from the positioning frame track information as a reference image of the clothes; obtaining an upper monitoring image corresponding to the reference image, carrying out image recognition on the reference image and the upper monitoring image to obtain clothes reference information, carrying out similarity comparison on the clothes reference information and clothes specification information, and if the similarity is lower than a preset threshold value, judging that the clothes are abnormal and triggering an alarm; the overall production efficiency of clothes sorting is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of clothing recognition, and in particular to a clothing sorting system based on computer vision. Background Art

[0002] RFID intelligent hanging sorting system (FMS system) uses the rapid identification capability of RFID and the rapid flow capability of the automatic conveyor line to hang the whole piece of clothing or cut pieces on the hanger with RFID electronic tags, and install a fast reading reader on the sorting line to automatically send it to the next sorting operation exit according to the sorting process input in advance. It greatly reduces the non-production time of handling, picking, searching, etc.

[0003] In modern clothing production, the hanging production system is widely used in the processing and sorting of clothing due to its high efficiency and flexibility. By hanging the clothing on a special carrier and moving it with automated conveying equipment, seamless connection between different processes can be achieved. In order to ensure the smooth operation of the production line and accurately identify various types of clothing, a chip is usually installed on each carrier, which records specific information about the clothing it carries, such as style, size, color, and the processes required to be performed. The entry of this information often relies on manual operation or preliminary automation means. Although it improves efficiency to a certain extent, it also introduces potential error risks.

[0004] Since the judgment of clothing types in the entire production process is completely based on the data information in the carrier chip, once the clothing is hung incorrectly or the data entry is deviated, it will cause chaos in the subsequent production process. For example, clothing that should have been sent to the embroidery station may be mistakenly sent to the cutting area, which not only affects production efficiency, but may also lead to a decline in the quality of the finished product, and even require rework, increasing production costs and time costs. Summary of the invention

[0005] The purpose of the present invention is to provide a clothing sorting system based on computer vision to solve the following technical problems:

[0006] Since the judgment of clothing types in the entire production process is completely based on the data information in the carrier chip, once the clothing is hung incorrectly or the data entry is deviated, it will cause chaos in the subsequent production process. For example, clothing that should have been sent to the embroidery station may be mistakenly sent to the cutting area, which not only affects production efficiency, but may also lead to a decline in the quality of the finished product, and even require rework, increasing production costs and time costs.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A computer vision-based clothing sorting system, a data acquisition module, uses a pre-configured radio frequency identification device to acquire clothing specification information at a target location, and sets acquisition points directly in front of and directly above the target location, respectively, acquires monitoring images of the target location at each acquisition point, and generates a front monitoring image sequence and an upper monitoring image sequence; wherein the clothing specification information includes standard clothing images, standard clothing sizes, standard clothing types, standard clothing colors, and standard clothing materials;

[0009] A data analysis module is used to analyze the front monitoring image sequence, generate a positioning frame of any clothing from the front monitoring image and obtain the positioning frame trajectory information of the clothing, and select the corresponding monitoring image with the smallest positioning frame interaction ratio of the clothing from the positioning frame trajectory information as the reference image of the clothing;

[0010] The result generation module is used to obtain the upper monitoring image corresponding to the reference image, perform image recognition on the reference image and the upper monitoring image, obtain clothing reference information, and perform similarity comparison between the clothing reference information and clothing specification information. If the similarity is lower than a preset threshold, it is determined to be abnormal and an alarm is triggered; if the similarity is greater than or equal to the preset threshold, it is determined to be qualified and no alarm is triggered.

[0011] As a further solution of the present invention: in the data analysis module, the specific process of generating the positioning frame is:

[0012] S1, grayscale processing is performed on the front monitoring image to obtain a monitoring grayscale image, a rectangular coordinate system is established with the central pixel of the monitoring grayscale image as the origin, the coordinates (x, y) of all pixels in the monitoring grayscale image are generated, and the grayscale values ​​Gi(x, y) of all pixels in the monitoring grayscale image are identified, i=1,...,n, i represents the order of the monitoring grayscale image where the pixel is located in the sequence, and the grayscale difference H of the pixel at the same position between adjacent monitoring grayscale images is calculated by the grayscale difference formula H=|Gi+1(x, y)-Gi(x, y)|. If H is greater than a preset threshold, it indicates that the corresponding pixel is a clothing pixel. If H is less than or equal to the preset threshold, it indicates that the corresponding pixel is a background pixel.

[0013] S2, merging adjacent clothing pixels, and circumscribing the merged clothing pixel area with a minimum circumscribed rectangle to obtain a pixel rectangle of any clothing in the monitored grayscale image, and marking the pixel rectangle as a positioning frame of the clothing.

[0014] As a further solution of the present invention: in the data analysis module, the specific process of obtaining the reference image is as follows:

[0015] Select any surveillance image, calculate the total overlapping area S of the clothing positioning frame and any other clothing positioning frame, calculate the area X of the clothing positioning frame, and calculate the positioning frame interaction ratio of the clothing in the current surveillance image according to the calculation formula C=S / X. Repeat the above process to obtain the positioning frame interaction ratio of the clothing in all surveillance images, and select the surveillance image corresponding to the minimum positioning frame interaction ratio MINC as the reference image of the clothing.

[0016] As a further solution of the present invention: if there are two or more monitoring images with the same positioning frame interaction ratio, then the monitoring image with the least number of positioning frames is selected as the reference image of the clothing.

[0017] As a further solution of the present invention: the data analysis module also includes calculating the positioning frame area S of the clothing in each front monitoring image, and calibrating the maximum positioning frame area MAXS as the first reference area S'; calculating the positioning frame area S of the clothing in each upper monitoring image, and obtaining the positioning frame length C and width V corresponding to the minimum positioning frame area MINS.

[0018] As a further solution of the present invention: in the result generation module, the specific process of obtaining the clothing reference information is as follows:

[0019] S11, converting the reference image into an HSV color space to obtain an HSV color matrix and performing a Haar transformation, normalizing the matrix after the Haar transformation to obtain a 256-dimensional color feature vector A;

[0020] S12, grayscale processing the reference image to obtain a grayscale image, extracting texture features from the grayscale image to obtain roughness, contrast and orientation values ​​of the grayscale image and normalizing the image to obtain a texture feature vector B;

[0021] S13, according to the calculation formula The actual positioning frame area Q of the clothing is calculated, where S' is the first reference area, f is the focal length of the camera at the collection point directly in front of the target position, and D is the horizontal distance between the collection point directly in front and the target position;

[0022] S14, according to the calculation formula The actual positioning frame width V' of the clothing is calculated, where V is the positioning frame width corresponding to the minimum positioning frame area MINS, F is the focal length of the camera at the collection point directly above the target position, d is the vertical distance between the collection point directly above and the target position, and P is the pixel density.

[0023] As a further solution of the present invention: in the result generation module, the specific process of similarity comparison is:

[0024] S111, convert the standard clothing image into HSV color space to obtain the corresponding HSV color matrix, apply Haar transformation to the HSV color matrix, normalize the matrix after Haar transformation to obtain the color feature vector a of the standard clothing image; grayscale process the standard clothing image to obtain a grayscale image, extract texture features from the grayscale image to obtain the roughness, contrast and orientation values ​​of the grayscale image and normalize them to obtain the texture feature vector b of the standard clothing image; calculate the vector cosine value N of the color feature vector A and the color feature vector a; calculate the vector cosine value M of the texture feature vector B and the texture feature vector b; calculate the first similarity W1 of the clothing according to the calculation formula X1=b*M+d*N; wherein b and d are preset coefficients;

[0025] S112, obtaining standard clothing sizes, constructing a reference rectangle based on the standard clothing sizes and obtaining the area Q of the reference rectangle 参 According to the calculation formula X2 = Q / Q 参 Calculate and obtain the second similarity W2 of the clothing;

[0026] S113, according to the calculation formula X3 = V' / V 标 The third similarity of the clothing is calculated, where V 标 It is the reference standard value corresponding to the standard clothing type;

[0027] S114, calculating the similarity X according to the calculation formula X=αX1+βX2+γX3, wherein α, β and γ are preset coefficients.

[0028] As a further solution of the present invention: if the first similarity is greater than or equal to a preset threshold, directly determining that it is abnormal and triggering an alarm, without repeating S112, S113 and S114.

[0029] Beneficial effects of the present invention:

[0030] The clothing specification information is automatically obtained through the radio frequency identification device, and the front and upper monitoring image sequences are automatically generated using the image acquisition points. The front monitoring image is used to locate and track the clothing, and the upper monitoring image is used for further image recognition and comparison. The positioning frame of the clothing is generated by the front monitoring image and the trajectory information is obtained to accurately track the position changes of the clothing, providing an accurate positioning basis for subsequent image recognition. By analyzing the front monitoring image, the monitoring image with the smallest interaction ratio is selected as the reference image, and the reference image is compared with the clothing specification information to achieve intelligent anomaly detection. When the similarity is lower than the preset threshold, it can automatically determine the anomaly and trigger an alarm, thereby improving the response speed and processing efficiency of abnormal situations, thereby verifying whether the clothing matches the carrier, and being able to promptly discover mismatch errors to avoid them being transported to the wrong process, thereby ensuring the overall production efficiency of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below in conjunction with the accompanying drawings.

[0032] Figure 1 It is a flow chart of a clothing sorting system based on computer vision of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] See also Figure 1 As shown, the present invention is a clothing sorting system based on computer vision, comprising:

[0035] The data acquisition module uses a pre-configured radio frequency identification device to acquire clothing specification information at the target location, and sets collection points in front of and above the target location, respectively, collects monitoring images of the target location at each collection point, and generates a front monitoring image sequence and an upper monitoring image sequence; wherein the clothing specification information includes standard clothing images, standard clothing sizes, standard clothing types, standard clothing colors, and standard clothing materials;

[0036] A data analysis module is used to analyze the front monitoring image sequence, generate a positioning frame of any clothing from the front monitoring image and obtain the positioning frame trajectory information of the clothing, and select the corresponding monitoring image with the smallest positioning frame interaction ratio of the clothing from the positioning frame trajectory information as the reference image of the clothing;

[0037] The result generation module is used to obtain the upper monitoring image corresponding to the reference image, perform image recognition on the reference image and the upper monitoring image, obtain clothing reference information, and perform similarity comparison between the clothing reference information and clothing specification information. If the similarity is lower than a preset threshold, it is determined to be abnormal and an alarm is triggered; if the similarity is greater than or equal to the preset threshold, it is determined to be qualified and no alarm is triggered.

[0038] The clothing specification information is automatically obtained through the radio frequency identification device, and the front and upper monitoring image sequences are automatically generated using the image acquisition points. The front monitoring image is used to locate and track the clothing, and the upper monitoring image is used for further image recognition and comparison. The positioning frame of the clothing is generated by the front monitoring image and the trajectory information is obtained to accurately track the position changes of the clothing, providing an accurate positioning basis for subsequent image recognition. By analyzing the front monitoring image, the monitoring image with the smallest interaction ratio is selected as the reference image, and the reference image is compared with the clothing specification information to achieve intelligent anomaly detection. When the similarity is lower than the preset threshold, it can automatically determine the anomaly and trigger an alarm, thereby improving the response speed and processing efficiency of abnormal situations, thereby verifying whether the clothing matches the carrier, and being able to promptly discover mismatch errors to avoid them being transported to the wrong process, thereby ensuring the overall production efficiency of the production line.

[0039] In a preferred case of this embodiment, in the data analysis module, the specific process of generating the positioning frame is:

[0040] S1, grayscale processing is performed on the front monitoring image to obtain a monitoring grayscale image, a rectangular coordinate system is established with the central pixel of the monitoring grayscale image as the origin, the coordinates (x, y) of all pixels in the monitoring grayscale image are generated, and the grayscale values ​​Gi(x, y) of all pixels in the monitoring grayscale image are identified, i=1,...,n, i represents the order of the monitoring grayscale image where the pixel is located in the sequence, and the grayscale difference H of the pixel at the same position between adjacent monitoring grayscale images is calculated by the grayscale difference formula H=|Gi+1(x, y)-Gi(x, y)|. If H is greater than a preset threshold, it indicates that the corresponding pixel is a clothing pixel. If H is less than or equal to the preset threshold, it indicates that the corresponding pixel is a background pixel.

[0041] S2, merging adjacent clothing pixels, and circumscribing the merged clothing pixel area with a minimum circumscribed rectangle to obtain a pixel rectangle of any clothing in the monitored grayscale image, and marking the pixel rectangle as a positioning frame of the clothing.

[0042] In another preferred situation of this embodiment, in the data analysis module, the specific process of obtaining the reference image is:

[0043] Select any surveillance image, calculate the total overlapping area S of the clothing positioning frame and any other clothing positioning frame, calculate the area X of the clothing positioning frame, and calculate the positioning frame interaction ratio of the clothing in the current surveillance image according to the calculation formula C=S / X. Repeat the above process to obtain the positioning frame interaction ratio of the clothing in all surveillance images, and select the surveillance image corresponding to the minimum positioning frame interaction ratio MINC as the reference image of the clothing.

[0044] In another preferred situation of the present embodiment, if there are two or more monitoring images with the same positioning frame interaction ratio, the monitoring image with the least number of positioning frames is selected as the reference image of the clothing.

[0045] In another preferred situation of the present embodiment, the data analysis module further includes calculating the positioning frame area S of the clothing in each front monitoring image, and calibrating the maximum positioning frame area MAXS as the first reference area S'; calculating the positioning frame area S of the clothing in each upper monitoring image, and obtaining the positioning frame length C and width V corresponding to the minimum positioning frame area MINS.

[0046] In another preferred situation of this embodiment, in the result generating module, the specific process of obtaining the clothing reference information is as follows:

[0047] S11, converting the reference image into an HSV color space to obtain an HSV color matrix and performing a Haar transformation, normalizing the matrix after the Haar transformation to obtain a 256-dimensional color feature vector A;

[0048] S12, grayscale processing the reference image to obtain a grayscale image, extracting texture features from the grayscale image to obtain roughness, contrast and orientation values ​​of the grayscale image and normalizing the image to obtain a texture feature vector B;

[0049] S13, according to the calculation formula The actual positioning frame area Q of the clothing is calculated, where S' is the first reference area, f is the focal length of the camera at the collection point directly in front of the target position, and D is the horizontal distance between the collection point directly in front and the target position;

[0050] S14, according to the calculation formula The actual positioning frame width V' of the clothing is calculated, where V is the positioning frame width corresponding to the minimum positioning frame area MINS, F is the focal length of the camera at the collection point directly above the target position, d is the vertical distance between the collection point directly above and the target position, and P is the pixel density.

[0051] In another preferred embodiment of the present invention, in the result generation module, the specific process of similarity comparison is as follows:

[0052] S111, convert the standard clothing image into HSV color space to obtain the corresponding HSV color matrix, apply Haar transformation to the HSV color matrix, normalize the matrix after Haar transformation to obtain the color feature vector a of the standard clothing image; grayscale process the standard clothing image to obtain a grayscale image, extract texture features from the grayscale image to obtain the roughness, contrast and orientation values ​​of the grayscale image and normalize them to obtain the texture feature vector b of the standard clothing image; calculate the vector cosine value N of the color feature vector A and the color feature vector a; calculate the vector cosine value M of the texture feature vector B and the texture feature vector b; calculate the first similarity W1 of the clothing according to the calculation formula X1=b*M+d*N; wherein b and d are preset coefficients;

[0053] S112, obtaining standard clothing sizes, constructing a reference rectangle based on the standard clothing sizes and obtaining the area Q of the reference rectangle 参 According to the calculation formula X2 = Q / Q 参 Calculate and obtain the second similarity W2 of the clothing;

[0054] S113, according to the calculation formula X3 = V' / V 标 The third similarity of the clothing is calculated, where V 标 It is the reference standard value corresponding to the standard clothing type;

[0055] S114, calculating the similarity X according to the calculation formula X=αX1+βX2+γX3, wherein α, β and γ are preset coefficients.

[0056] In another preferred situation of this embodiment, if the first similarity is greater than or equal to a preset threshold, it is directly determined to be abnormal and an alarm is triggered without repeating S112, S113 and S114.

[0057] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A clothing sorting system based on computer vision, characterized in that: include: The data acquisition module uses a pre-configured radio frequency identification device to acquire clothing specification information at the target location, and sets collection points in front of and above the target location, respectively, collects monitoring images of the target location at each collection point, and generates a front monitoring image sequence and an upper monitoring image sequence; wherein the clothing specification information includes standard clothing images, standard clothing sizes, standard clothing types, standard clothing colors, and standard clothing materials; A data analysis module is used to analyze the front monitoring image sequence, generate a positioning frame of any clothing from the front monitoring image and obtain the positioning frame trajectory information of the clothing, and select the corresponding monitoring image with the smallest positioning frame interaction ratio of the clothing from the positioning frame trajectory information as the reference image of the clothing; The result generation module is used to obtain the upper monitoring image corresponding to the reference image, perform image recognition on the reference image and the upper monitoring image, obtain clothing reference information, and perform similarity comparison between the clothing reference information and clothing specification information. If the similarity is lower than a preset threshold, it is determined to be abnormal and an alarm is triggered; if the similarity is greater than or equal to the preset threshold, it is determined to be qualified and no alarm is triggered.

2. A computer vision-based clothing sorting system according to claim 1, characterized in that: In the data analysis module, the specific process of generating the positioning frame is as follows: S1, grayscale processing is performed on the front monitoring image to obtain a monitoring grayscale image, a rectangular coordinate system is established with the central pixel of the monitoring grayscale image as the origin, the coordinates (x, y) of all pixels in the monitoring grayscale image are generated, and the grayscale values ​​Gi(x, y) of all pixels in the monitoring grayscale image are identified, i=1,...,n, i represents the order of the monitoring grayscale image where the pixel is located in the sequence, and the grayscale difference H of the pixel at the same position between adjacent monitoring grayscale images is calculated by the grayscale difference formula H=|Gi+1(x, y)-Gi(x, y)|. If H is greater than a preset threshold, it indicates that the corresponding pixel is a clothing pixel. If H is less than or equal to the preset threshold, it indicates that the corresponding pixel is a background pixel. S2, merging adjacent clothing pixels, and circumscribing the merged clothing pixel area with a minimum circumscribed rectangle to obtain a pixel rectangle of any clothing in the monitored grayscale image, and marking the pixel rectangle as a positioning frame of the clothing.

3. The computer vision-based clothing sorting system according to claim 1, characterized in that: In the data analysis module, the specific process of obtaining the reference image is as follows: Select any surveillance image, calculate the total overlapping area S of the clothing positioning frame and any other clothing positioning frame, calculate the area X of the clothing positioning frame, and calculate the positioning frame interaction ratio of the clothing in the current surveillance image according to the calculation formula C=S / X. Repeat the above process to obtain the positioning frame interaction ratio of the clothing in all surveillance images, and select the surveillance image corresponding to the minimum positioning frame interaction ratio MINC as the reference image of the clothing.

4. The computer vision-based clothing sorting system according to claim 3, characterized in that: It also includes selecting the monitoring image with the least number of positioning frames as the reference image of the clothing if there are two or more monitoring images with the same positioning frame interaction ratio.

5. The computer vision-based clothing sorting system according to claim 1, characterized in that: The data analysis module also includes calculating the positioning frame area S of the clothing in each front monitoring image, and marking the maximum positioning frame area MAXS as the first reference area S'; The positioning frame area S of the clothing in each upper monitoring image is calculated, and the positioning frame length C and width V corresponding to the minimum positioning frame area MINS are obtained.

6. The computer vision-based clothing sorting system according to claim 5, characterized in that: In the result generation module, the specific process of obtaining clothing reference information is as follows: S11, converting the reference image into an HSV color space to obtain an HSV color matrix and performing a Haar transformation, normalizing the matrix after the Haar transformation to obtain a 256-dimensional color feature vector A; S12, grayscale processing the reference image to obtain a grayscale image, extracting texture features from the grayscale image to obtain roughness, contrast and orientation values ​​of the grayscale image and normalizing the image to obtain a texture feature vector B; S13, according to the calculation formula The actual positioning frame area Q of the clothing is calculated, where S' is the first reference area, f is the focal length of the camera at the collection point directly in front of the target position, and D is the horizontal distance between the collection point directly in front and the target position; S14, according to the calculation formula The actual positioning frame width V' of the clothing is calculated, where V is the positioning frame width corresponding to the minimum positioning frame area MINS, F is the focal length of the camera at the collection point directly above the target position, d is the vertical distance between the collection point directly above and the target position, and P is the pixel density.

7. The computer vision-based clothing sorting system according to claim 1, characterized in that: In the result generation module, the specific process of similarity comparison is as follows: S111, converting the standard clothing image into an HSV color space to obtain a corresponding HSV color matrix, applying Haar transformation to the HSV color matrix, normalizing the matrix after Haar transformation, and obtaining a color feature vector a of the standard clothing image; Perform grayscale processing on the standard clothing image to obtain a grayscale image, perform texture feature extraction on the grayscale image to obtain the roughness, contrast and orientation values ​​of the grayscale image and perform normalization processing to obtain the texture feature vector b of the standard clothing image; calculate the vector cosine value N of the color feature vector A and the color feature vector a; calculate the vector cosine value M of the texture feature vector B and the texture feature vector b; calculate the first similarity W1 of the clothing according to the calculation formula X1=b*M+d*N; wherein b and d are preset coefficients; S112, obtaining standard clothing sizes, constructing a reference rectangle based on the standard clothing sizes and obtaining the area Q of the reference rectangle 参 According to the calculation formula X2 = Q / Q 参 Calculate and obtain the second similarity W2 of the clothing; S113, according to the calculation formula X3 = V' / V 标 The third similarity of the clothing is calculated, where V 标 It is the reference standard value corresponding to the standard clothing type; S114, calculating the similarity X according to the calculation formula X=αX1+βX2+γX3, wherein α, β and γ are preset coefficients.

8. The computer vision-based clothing sorting system according to claim 7, characterized in that: It also includes directly determining that an abnormality is present and triggering an alarm if the first similarity is greater than or equal to a preset threshold, without repeating S112, S113 and S114.

Images