Garment size measuring system based on machine vision

Abstract

The invention discloses a garment size measurement system based on machine vision, and the system comprises the following steps: A, image collection: flatly laying a garment on an operation table, enabling an image collection module to horizontally move along the operation table at a constant speed, collecting a garment image, and generating a measurement image; step B, image preprocessing: preprocessing the measurement image, and selecting a region of interest through ROI (Region of Interest); step C, edge detection and deformation analysis: using a Canny edge detection algorithm for the measurement image, fitting an edge curve through a polynomial, analyzing curvature change to identify a deformation area, if the deformation area is identified, executing step D, and if the deformation area is not identified, executing step E; through the machine vision and image processing technology, detection and calculation of the garment size are automatically completed, manual measurement is not needed, the measurement efficiency is greatly improved, and the requirement for large-scale automatic measurement is met.

Description

Technical Field

[0001] The invention relates to the technical field of clothing size measurement, and in particular to a clothing size measurement system based on machine vision. Background Art

[0002] With the booming development of e-commerce and online shopping, consumers are increasingly inclined to choose clothes online. However, online clothing cannot be tried on or measured manually, which makes consumers' online shopping choices somewhat blind and unable to accurately choose clothes that fit them.

[0003] Traditional clothing size measurement relies entirely on manual work, which is inefficient and cannot meet the needs of modern efficient automation. There are also some automatic clothing measurement systems based on computer vision on the market, but they are mainly used for clothing with simpler structures. For clothing images with complex structures or deformations, the measurement accuracy cannot be guaranteed.

[0004] Therefore, it is of great practical significance to develop a new automatic clothing measurement system based on machine vision, which can perform efficient and accurate intelligent analysis, processing and measurement of clothing images with complex structures, ensure measurement accuracy, realize intelligent and automated clothing manufacturing, and meet the development needs of the modern clothing industry. This is the current trend and urgent need in the development of automatic clothing measurement technology. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a clothing size measurement system based on machine vision in order to better and effectively solve the above problems.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A clothing size measurement system based on machine vision comprises the following steps:

[0008] Step A, image acquisition: the garment is laid flat on the operating table, and the image acquisition module moves horizontally along the operating table at a uniform speed to acquire garment images and generate a measurement diagram;

[0009] Step B, image preprocessing: preprocess the measurement image and select the region of interest through ROI;

[0010] Step C, edge detection and deformation analysis: Use the Canny edge detection algorithm on the measurement image, fit the edge curve through a polynomial, analyze the curvature change to identify the deformation area, if the deformation area is identified, execute step D, if not, execute step E;

[0011] Step D, deformation compensation: polynomial fitting is used again to describe the deformation in the deformation area, the edge is fitted using the least squares method, and the edge curve is generated. Assuming that the edge is an ideal straight line before deformation, the correction coordinates are calculated by reverse transformation to generate the real contour;

[0012] Step E, template image matching: establish a clothing template library, store the standard template image, mark the key measurement points on the template image, and match the measurement image with the standard template image to complete the key measurement point marking and generate the marked image;

[0013] Step F, size measurement: calculate the distance relationship between key measurement points on the annotation diagram to obtain the clothing size.

[0014] Preferably, the Canny edge detection algorithm in step C comprises the following specific steps:

[0015] Step C1, dividing the continuous edge into multiple local parts, collecting a series of discrete edge points in each local part, performing local curve fitting on these discrete points using a polynomial, fitting them into a quadratic or cubic low-order polynomial, and generating a curve according to the low-order polynomial;

[0016] Step C2, analyzing the curvature change of the curve by the second-order derivative, and determining whether the second-order derivative value of the curve changes. If so, it is considered that there is a deformation region, otherwise, there is no deformation region.

[0017] Preferably, the steps of generating the true contour in step D are as follows:

[0018] Step D1, obtaining the slope of the edge curve by polynomial derivation, and obtaining the normal slope by slope calculation;

[0019] Step D2, projecting the points on the edge curve onto the ideal straight line in the direction of the normal slope, calculating and obtaining the corresponding points on the ideal straight line for each point on the edge curve, and then generating a true contour.

[0020] Preferably, the establishment of the clothing template library in step E comprises the following specific steps:

[0021] Step E1, collecting different clothing images and classifying the clothing images according to clothing types;

[0022] Step E2: training the clothing classification model through a convolutional neural network and embedding it into the clothing template library.

[0023] Preferably, step F obtains the clothing size, specifically by multiplying the distance between adjacent key measurement points by a proportional factor to generate the clothing size.

[0024] Preferably, step B pre-processes the measurement image, specifically, using Gaussian filtering to perform denoising and gray-scale conversion.

[0025] The beneficial effects of the present invention are: 1. Automatic and efficient measurement of clothing size is achieved

[0026] Through machine vision and image processing technology, clothing size detection and calculation are automatically completed without manual measurement, which greatly improves measurement efficiency and meets the needs of large-scale automated measurement.

[0027] 2. Ensure measurement accuracy

[0028] For clothing areas that may be deformed, a deformation compensation algorithm is used to ensure measurement accuracy, meet the measurement requirements of clothing manufacturing and sales, and is superior to the measurement accuracy of similar equipment currently available.

[0029] 3. Wide application, supports measurement of all kinds of clothing

[0030] By establishing a clothing template library and intelligent matching algorithm, it can adapt to the measurement of different styles of clothing, broaden the application scope of the system, and meet the needs of measuring the size of various types of clothing.

[0031] 4. Simplify equipment implementation, reduce costs, and facilitate promotion

[0032] The system can be implemented by simply setting up an operating table, image acquisition equipment and a computer to execute the algorithm program. The system has low cost and is easy to promote and apply. It can provide technical support for clothing production and e-commerce companies and improve the level of automation and precision. DETAILED DESCRIPTION

[0033] The present invention will be further described below.

[0034] A clothing size measurement system based on machine vision of the present invention comprises the following steps:

[0035] Step A, image acquisition: the garment is laid flat on the operating table, and the image acquisition module moves horizontally along the operating table at a uniform speed to acquire garment images and generate a measurement diagram;

[0036] Among them, the calibrated camera is used to shoot the clothing to be measured to obtain clothing pictures. The photographed clothing should be laid flat on the operating table, the camera is located on the side of the table, the lens is located on the central axis parallel to the long side of the table, and is horizontal to the table to eliminate errors in subsequent recognition caused by the tilt of the lens during shooting. Considering that some clothes may be long, if the camera position is relatively low, it is difficult to shoot the whole clothing. If the position is higher, the position of the device is more demanding, and the clothing image will be smaller when shooting short clothes. For this reason, the lens is located on a sliding track at a certain distance above the table. When shooting is required, the camera slides at a uniform speed on the track to capture a long picture for the next analysis step.

[0037] Step B, image preprocessing: preprocess the measurement image and select the region of interest through ROI;

[0038] Step C, edge detection and deformation analysis: Use the Canny edge detection algorithm on the measurement image, fit the edge curve through a polynomial, analyze the curvature change to identify the deformation area, if the deformation area is identified, execute step D, if not, execute step E;

[0039] Step D, deformation compensation: polynomial fitting is used again to describe the deformation in the deformation area, the edge is fitted using the least squares method, and the edge curve is generated. Assuming that the edge is an ideal straight line before deformation, the correction coordinates are calculated by reverse transformation to generate the real contour;

[0040] Step E, template image matching: establish a clothing template library, store the standard template image, mark the key measurement points on the template image, and match the measurement image with the standard template image to complete the key measurement point marking and generate the marked image;

[0041] Step F, size measurement: calculate the distance relationship between key measurement points on the annotation diagram to obtain the clothing size.

[0042] Specifically, step C, Canny edge detection algorithm, the specific steps are:

[0043] Step C1, dividing the continuous edge into multiple local parts, collecting a series of discrete edge points in each local part, performing local curve fitting on these discrete points using a polynomial, fitting them into a quadratic or cubic low-order polynomial, and generating a curve according to the low-order polynomial;

[0044] Step C2, analyzing the curvature change of the curve by the second-order derivative, and determining whether the second-order derivative value of the curve changes. If so, it is considered that there is a deformation region, otherwise, there is no deformation region.

[0045] Specifically, step D generates the true contour, and the specific steps are as follows:

[0046] Step D1, obtaining the slope of the edge curve by polynomial derivation, and obtaining the normal slope by slope calculation;

[0047] Step D2, projecting the points on the edge curve onto the ideal straight line in the direction of the normal slope, calculating and obtaining the corresponding points between each point on the edge curve and the ideal straight line, and then generating a true contour;

[0048] Here, the polynomial is represented as f(x), the derivative is f′(x), and the equation of the normal line can be represented as yf(x i )=m i (xx i ),in, represents the slope of the normal line. Assume that the contour before deformation can be represented by a straight line y=bx+c, and the point on the straight line is represented by (x′ i , y′ j ), project the points on the curve onto the straight line through the normal direction, and calculate each point on the curve (x i ,f(x i )) corresponds to (x′) on the straight line i , y′ i ), thus obtaining the straight line model before edge deformation.

[0049] To calculate the straight line parameters b and c, the first and last coordinate values ​​of the deformation area can be directly selected and substituted into the solution.

[0050] Specifically, step E, establishing a clothing template library, the specific steps are:

[0051] Step E1, collecting different clothing images and classifying the clothing images according to clothing types;

[0052] Step E2: training the clothing classification model through a convolutional neural network and embedding it into the clothing template library.

[0053] Specifically, step F obtains the clothing size by multiplying the distance between adjacent key measurement points by a scale factor to generate the clothing size.

[0054] Specifically, step B pre-processes the measurement image, and the specific steps are to use Gaussian filtering to perform denoising and grayscale conversion.

[0055] Now further analyze the system in combination with specific embodiments and comparative examples

[0056] Example 1

[0057] A clothing size measurement system based on machine vision, measuring clothing measurements without deformation:

[0058] A shirt with a simple structure, rigid material, and not prone to deformation was selected for measurement. After image processing and edge detection, no deformation was detected. The collar, sleeve, and length of the shirt were generated by directly matching with the standard template, calibrating the measurement points, and calculating the distance between key points.

[0059] Example 2

[0060] A clothing size measurement system based on machine vision to measure deformed clothing:

[0061] A soft and easily deformed skirt was selected for measurement. After processing, it was found that there was configuration deformation. Deformation compensation was performed on the deformed area to generate the true contour before deformation. Then, the measurement point calibration was completed through template matching to generate the size data of the skirt such as waist circumference and skirt length.

[0062] Comparative Example 1

[0063] Manually measure the two garments mentioned above 10 times each, and record the average measurement time and error range.

[0064] Comparative Example 2

[0065] Using a vision-based clothing measurement device, the two pieces of clothing were measured 10 times, and the average time and error were recorded.

[0066] The specific measurement results are shown in Table 2

[0067] Average time of 10 measurements 10 times measurement error Example 1 8s ±0.5cm Example 2 15s ±1cm Comparative Example 1 Shirt 60s ±1cm Comparative Example 1 Skirt 90s ±2cm Comparative Example 1 Shirt 12s ±1cm Comparative Example 1 Skirt 30s ±3cm

[0068] Table 2

[0069] In summary, the clothing size measurement system based on machine vision of the present invention has the following features:

[0070] Beneficial effects:

[0071] 1. Realizes automatic and efficient measurement of clothing size

[0072] Through machine vision and image processing technology, clothing size detection and calculation are automatically completed without manual measurement, which greatly improves measurement efficiency and meets the needs of large-scale automated measurement.

[0073] 2. Ensure measurement accuracy

[0074] For clothing areas that may be deformed, a deformation compensation algorithm is used to ensure measurement accuracy, meet the measurement requirements of clothing manufacturing and sales, and is superior to the measurement accuracy of similar equipment currently available.

[0075] 3. Wide application, supports measurement of all kinds of clothing

[0076] By establishing a clothing template library and intelligent matching algorithm, it can adapt to the measurement of different styles of clothing, broaden the application scope of the system, and meet the needs of measuring the size of various types of clothing.

[0077] 4. Simplify equipment implementation, reduce costs, and facilitate promotion

[0078] The system can be implemented by simply setting up an operating table, image acquisition equipment and a computer to execute the algorithm program. The system has low cost and is easy to promote and apply. It can provide technical support for clothing production and e-commerce companies and improve the level of automation and precision.

[0079] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A clothing size measurement system based on machine vision, characterized in that: The following steps are included: Step (A), image acquisition: the garment is laid flat on the operating table, and the image acquisition module moves horizontally along the operating table at a uniform speed to acquire the garment image and generate a measurement diagram; Step (B), image preprocessing: preprocess the measurement image and select the region of interest through ROI; Step (C), edge detection and deformation analysis: using the Canny edge detection algorithm on the measurement image, fitting the edge curve through a polynomial, analyzing the curvature change to identify the deformation area, if the deformation area is identified, executing step (D), if not, executing step (E); Step (D), deformation compensation: polynomial fitting is used again to describe the deformation in the deformation area, the edge is fitted using the least squares method, and the edge curve is generated. Assuming that the edge is an ideal straight line before deformation, the correction coordinates are calculated by reverse transformation to generate the true contour; Step (E), template image matching: establish a clothing template library, store a standard template image, mark key measurement points on the template image, and match the measurement image with the standard template image to complete the key measurement point marking and generate a marked image; Step (F), size measurement: calculate the distance relationship between key measurement points on the annotation diagram to obtain the clothing size.

2. The clothing size measurement system based on machine vision according to claim 1, characterized in that: Step (C), Canny edge detection algorithm, the specific steps are: Step (C1), dividing the continuous edge into a plurality of local parts, collecting a series of discrete edge points in each local part, performing local curve fitting on these discrete points using a polynomial, fitting them into a quadratic or cubic low-order polynomial, and generating a curve according to the low-order polynomial; Step (C2), analyzing the curvature change of the curve by the second-order derivative, and determining whether the second-order derivative value of the curve changes. If a change occurs, it is considered that a deformation region exists, otherwise, no deformation region exists.

3. The clothing size measurement system based on machine vision according to claim 1, characterized in that: Step (D) generates the true contour. The specific steps are as follows: Step (D1), obtaining the slope of the edge curve by polynomial derivation, and obtaining the normal slope by slope calculation; Step (D2), projecting the points on the edge curve onto the ideal straight line in the direction of the normal slope, calculating the corresponding points on each point on the edge curve and the ideal straight line, and then generating a true contour.

4. The clothing size measurement system based on machine vision according to claim 1, characterized in that: The specific steps of establishing a clothing template library in step (E) are as follows: Step (E1), collecting different clothing images and classifying the clothing images according to clothing types; Step (E2), training the clothing classification model through a convolutional neural network and embedding it into the clothing template library.

5. The clothing size measurement system based on machine vision according to claim 1, characterized in that: The step (F) of obtaining the clothing size comprises the following steps: multiplying the distance between adjacent key measurement points by a proportional factor to generate the clothing size.

6. The clothing size measurement system based on machine vision according to claim 1, characterized in that: The measurement image is preprocessed in step (B), specifically, the steps of using Gaussian filtering to remove noise and grayscale.