RU2789286C2 - Portable device for obtainment of anthropometric data and method for collection of anthropometric data - Google Patents

Abstract

FIELD: computer technology.

SUBSTANCE: present invention relates to a portable device for obtainment of anthropometric data and to a method for collection of anthropometric data. The portable device for obtainment of anthropometric data contains an image sensor and a processing unit. The processing unit contains, in turn: a reception module made with the possibility of reception of the first human images from the image sensor; a processor for processing of impersonal images, made with the possibility of creation of the second images from corresponding first images by application of a border detection procedure; a communication module made with the possibility of sending of the second images.

EFFECT: reduction in a computational load, as well as increase in computing power.

15 cl, 12 dwg

Description

The field of technology to which the present invention relates

The present invention relates to a portable anthropometric data acquisition device and to a method for collecting anthropometric data.

Background of the Invention

As you know, the collection of anthropometric indicators can be of interest in many areas, from trade to healthcare. For example, clothing e-commerce systems can assist shoppers by prompting them to select a size based on a set of personal data that includes relevant body measurements. Personal data may be submitted to an automatic categorization system and the size may be selected based on general best fit criteria. This service is not only convenient for customers, but also significantly reduces the number of returned goods, which is also beneficial for suppliers in terms of cost savings. Another example, in dietetics, monitoring of body parameters can help in monitoring the patient and selecting treatment depending on his response.

Clearly, the benefits of using anthropometric data would be greatly enhanced if direct purchasers or patients could collect data on the spot with simple tools that do not require special training or skills to produce reasonably accurate results. Anthropometric data is typically collected through image acquisition and analysis, and data collection can take advantage of the widespread use of highly versatile devices such as smartphones. Image capture tools have become so handy that end users are almost always able to crop and obtain images of sufficient quality to perform the necessary analysis procedures.

However, the delegation of the task of obtaining images to end users entails other problems that must be taken into account. On the one hand, image processing, which is resource intensive in itself, must also take into account the fact that field imaging usually suffers from poor conditions in terms of lighting, contrast with the background, user posture, and the like. Such conditions must be compensated by improving processing methods, which usually leads to an increase in the computational load. To ensure sufficient computing power, only basic capture functions are dedicated to image analysis on end user devices, while the main analysis procedures are performed at remote sites accessible via an Internet connection. On the other hand, sharing images with remote sites raises privacy and personal data protection concerns as sensitive information is involved and legal requirements are becoming more stringent. In essence, information relating to the personal aspect is sent without the control of the end user. Clearly, information processing is critical, especially for medical applications.

Description of the invention

Thus, it is an object of the present invention to provide a portable anthropometric data acquisition device and anthropometric data collection method that overcomes or at least alleviates the described limitations.

The present invention provides a portable anthropometric data acquisition device and an anthropometric data collection method as defined in claims 1 and 19, respectively.

Brief description of the figures

The present invention will be described with reference to the accompanying drawings, illustrating some non-limiting embodiments of the same, in which:

- figure 1 is a simplified schematic diagram of the system for obtaining anthropometric data;

- figure 2 is a more detailed schematic diagram of portable devices for obtaining anthropometric data included in the system shown in figure 2, and made in accordance with one of the embodiments of the present invention;

- figures 3a and 3b show masks for receiving data used in the device shown in figure 2;

- Fig.4 is a simplified block diagram of the steps of the method in accordance with one embodiment of the present invention;

- Figs. 5a and 5b show images obtained in the device shown in Figs. 2;

- Fig.6 is a more detailed block diagram of part of the steps of the method shown in Fig.4;

- Fig.7 is a more detailed schematic diagram of a component of the device shown in Fig.2;

- Fig.8 is a more detailed schematic diagram of a component of the system shown in Fig.1; And

- Figs. 9 and 10 show images obtained in the component shown in Figs. 8.

Preferred embodiment of the invention

In FIG. 1, the anthropometric data acquisition system is indicated generally by the reference number 1 and comprises one or more portable anthropometric data acquisition devices 2 in accordance with one embodiment of the present invention and a server 3. The portable acquisition devices 2 and the server 3 are interconnected via a wide area network. 4, for example, via the Internet. The portable acquisition devices 2 are capable of capturing images of people and performing preliminary image processing steps, which are hereinafter referred to as client-side processing, in which data sent to the server to perform corresponding feature extraction (server-side processing) is cleared of information about personality.

One exemplary portable receiving device 2 is shown in FIG. 2 and will be referred to below, it being understood that other portable receiving devices 2 contain the same components. The portable receiving device 2 includes a processing unit 5, a display 6 and an image sensor 7. According to one embodiment, the portable receiving device 2 can be integrated into a smartphone, tablet computer or laptop, advantageously with an integrated touch screen as display 6 and a camera as image sensor 7. In this case, the display 6 does not have to be touch sensitive.

The processing unit 5 is configured to implement the acquisition module 8, the image processor 10 and the communication module 11.

The acquisition module 8 activates the display 6 in real time during acquisition and captures images from the image sensor 7 in response to capture commands sent by the user through an interface that may include virtual or hard buttons (not shown). The real-time output signals of the image sensor 7 are continuously displayed on the display 6. The acquisition module 8 also sends information to the display 6 for visual presentation to the user to help correct acquisition.

The images captured by the receiving module 8 are supplied to the image processor 10 . The acquisition module 8 may be provided with some processing capabilities to perform basic operations such as white balance, brightness and contrast adjustment, and conversion to a compressed format. Alternatively, the raw format images are sent to the image processor 10 which performs all the necessary processing steps.

The image processor 10 extracts modified client images from the received images. The retrieved client images are sent to the server 3 via the communication module 11 for server-side processing, which may therefore be dependent on higher level resources.

The acquisition of the image is carried out by the acquisition module 8, which provides information to direct the cropped person to take a pose in a range of acceptable poses while the operator takes a picture from the back and a picture from the side. The standard pose significantly reduces the computational load associated with image processing. More specifically (FIG. 3a), the acquisition module 8 starts acquiring the back image by overlaying the back image mask 13 on the images displayed on the display 6 in real time. Thus, the person may be directed to assume a posture corresponding to the back mask 13, which in one embodiment may require a standing position with the legs apart and arms away from the torso. When the posture is correct, the operator activates the shooting commands, and the image from the back of the IMG _B (figure 2) is captured by the acquisition module 8 and sent to the image processor 10. The image from the back of the IMG _B may be temporarily stored in a memory unit built into either the acquisition unit 8 or the image processor 10, or in the general purpose memory unit of the portable acquisition device 2.

After the image from the back of the IMG _B has been captured and stored or sent to the image processor 10, the acquisition module 8 starts acquiring the side image and overlays the side image mask 14 (FIG. 3b) on the images displayed on the display 6 in real time . Again, the person is directed to take the pose corresponding to the side view mask 13, which requires a standing position with arms along the torso. When the posture is correct, the operator activates the shooting command, and the image from the side of the IMG _S is captured and sent to the image processor 10 or stored like the image from the back of the IMG _B .

Each of the IMG _B back image and the IMG _S side image is then processed by the image processor 10 to detect edges and restore contours at essentially the same stage. For simplicity, reference will hereinafter be made to the image from the back of IMG _B and it will be understood that the same applies to the image from the side of IMG _S unless otherwise noted.

As shown in FIG. 4, after obtaining a side IMG _S image in a format suitable for processing, the image processor 10 cleans up portions of the side IMG _S image corresponding to areas outside the back mask 13 (block 100), thereby reducing noise. . Edge detection is then performed (block 110) and contour reconstruction (block 120). Thus, a client image from the back of FIMG _B is obtained and transmitted to the server 3, including only the outline of the body and possibly residual background noise. Examples of a back client image of FIMG _B and a side client image of FIMG _S are shown in FIGS. 5a and 5b, respectively. At the boundary detection stage, features of the body that may allow identification of the depicted person are lost.

Edge detection can be done in several ways, but in general it is preferable that the thickness of the outline be fairly constant, as it can be a critical parameter in server-side post-processing. The image processor 10 may be configured to combine various edge detection routines to improve performance.

According to one embodiment (FIG. 6), the image processor 10 previously applies a noise reduction process to the image from the back of IMG _B to reduce the effect of high frequency noise (block 112). The denoising process may be median filtering. It has been found that a kernel size of 15 pixels in the median filtering operator provides acceptable filtering performance without compromising the accuracy of subsequent edge detection.

Next (block 114), the edge of the image from the back of IMG _B is first extracted using the Canny edge detector process, which is described in Canny, J., "A Computational Approach to Edge Detection", IEEE Trans. Pattern Analysis and Machine Intelligence, 8(6):679-698, 1986. The Canny edge detector is very sensitive to areas of low contrast in an image that may actually be present. This aspect can be critical, since it is not always possible to select a background that provides sufficient contrast when obtaining an image in the field. To avoid loss of information, the contour of the image from the back of IMG _B is re-derived by the processor 10 using another edge detector process (block 116), and the result of the re-edge extraction is combined with the result of the first contour extraction by the Canny edge detector in block 120 in FIG. According to one embodiment, the image processor 10 uses a structured random forest for edge detection, which is also referred to as a structured edge detector. The structured edge detector is less sensitive to local areas of low contrast and is also more robust to noise than the Canny edge detector, thus eliminating the disadvantages of the latter.

FIG. 7 shows an exemplary structure of an image detector 10 implementing the process of FIGS. 4 and 6. According to the illustrated embodiment, the image detector 10 comprises a noise reduction filter 15 receiving an image from the back of the IMG _B (and an image from the side of the IMG _S ) and transmitting it in parallel. into a Canny edge detector and a structured edge detector, referred to herein as 16 and 17, respectively. The first output I _{1 of} the Canny edge detector 16 and the second output I ₂ of the structured edge detector 17 are then combined in an edge recovery module 18 which combines the results of the first and second edge extraction and outputs a rear client image FIMG _B and a side client image FIMG _S .

According to one embodiment, contour recovery module 18 combines the first output I _{1 of} the Canny edge detector 16 and the second output I ₂ of the structured edge detector 17 by addition, thresholding, and binarization. More precisely, the brightness values of the first output I ₁ and the second output I ₂ are added and the result is compared with a threshold τ, which may be selected as a fraction of the maximum response of the image sensor 7, for example 10%. The combined output defines a client image from the back of FIMG _B , which is assigned a first logical value L ₁ (eg, high) if the sum of the first output I ₁ and the second output I ₂ exceeds a threshold τ, and a second logical value L ₁ (eg, low) in otherwise:

Similarly, the FIMG _S side client image is defined as follows:

When the side image of the IMG _S is transmitted to the Canny edge detector 16 and the structured edge detector 17 (the output of which is labeled I ₁ ' and I ₂ ' respectively).

According to another embodiment, probabilistic models may be used when sufficient processing power is available in the portable receiving devices 2 to have an acceptable response delay for users.

Each patch in the image IMG _B (or IMG _S ) is assigned a label 1∈[L1, L2] ^IMGB that minimizes the energy function E(l). The energy function contains the first component and the second component and looks like this:

The first component of E _UNARY (l) is a function that seeks to assign to each pixel a value corresponding to the output of the edge detectors with relative probabilities w ₁ and w2 respectively

Where

and w ₁ and w ₂ are weight parameters that determine to what extent the Canny edge detector 16 and structured edge detector 17 will influence the result.

The second component of E _PAIRWISE (l) seeks to have neighboring pixels match each other on the label, adding a constant mismatch penalty, and it is defined as follows:

Where

and PAIRS(I) includes all pairs of adjacent image pixels.

The weight parameter w _P determines how strong the correspondence between neighboring pixels should be. Optimal solution

is approximated by a trust propagation algorithm using tree reweighting.

Thus, from the back image of IMG _B and the side image of IMG _S , the back client image of FIMG _B and the side client image of FIMG _S , respectively, are obtained and sent to the server 3 for server-side processing. The FIMG _B back client image and the FIMG _S side client image include a complete and accurate body contour of the person being cropped, and no personal information about the identity is transmitted without the control of the user. Personal information (such as facial features) is effectively removed during the boundary detection and contouring stages without the need for special processing, and the body contour remains largely anonymous. As already mentioned, exemplary FIMG _B back client image and FIMG _S side client image are shown in FIGS. 5a and 5b, respectively, which also show a back image mask 13 and a side image mask 14 to intuitively illustrate how image acquisition is performed. and processing on the client side. However, the back mask 13 and the side mask 14 need not themselves be part of the back client image FIMG _B and the side client image FIMG _S .

As shown in FIG. 8, the server 3 includes a communication module 20 that receives a back client image FIMG _B and a side client image FIMG _S , and components for server-side processing, including a key point detector 21, a projection measurer 22, and a 23 three-dimensional prediction measurements.

The key point detector 21 is configured to identify key points based on the thickness of the contour line, since client images may contain noise, i.e. points that do not correspond to the contour of the body. More specifically, scanning along horizontal lines (eg, from left to right) is first performed on the client image from the back of FIMG _B , and pairs of subsequent lines of equal thickness are considered as the left and right borders of the corresponding body part. For example, if three pairs of lines of the same thickness are detected during horizontal scanning, then three body parts are identified, which can be the left and right borders of the left hand for the first pair, the left and right borders of the torso for the second pair, and the left and right borders of the right hand for the third couples.

When scanning a client image from the back, the FIMG _B 21 keypoint detector identifies the crown, heels, underarms, and perineum. The crown and heels are then identified with a second FIMG _S lateral scan of the client image.

The projection measuring means 22 is configured to identify contour points belonging to the torso based on the position of the armpits and perineum determined by the key point detector 21 (windows 30, 31 in FIGS. 5a and 5b). In order to reduce irregularities that may be introduced by previous processing steps, the projection measuring means 22 approximates the torso lines (32, FIG. 9) with polynomial functions (33, FIG. 10). Given the nature of the torso and, in particular, the expected number of highs and lows, a 4th degree polynomial is sufficient and at the same time not too difficult to fit. Also, possible missing contour parts can be estimated by the projection measurer 22 using polynomial functions such as Bezier polynomials.

Then the projection measuring means 22 measures the projections of the main dimensions of the body on the image plane or on a plane perpendicular to it. Basic body measurements may include, for example, neck, chest, waist, hips, crotch. Bust, waist and hip measurements can be taken in fixed and predefined vertical coordinates (Y-coordinate), for example, as follows:

where CA is the distance between the perineum and armpits.

According to another embodiment, Ychest is taken at the widest part of the upper third of the torso, obtained from the FIMG _S lateral client image; Ywaist is taken at the smallest part of the central third of the torso, obtained from the client image from the back of FIMG _B ; and Yhips is taken at the widest part of the lower third of the torso, obtained from the FIMG _S lateral client image. Other measurements may be added according to design preferences.

The 3D measurement predictor 23 is configured to provide estimates or predictions of actual body measurements based on measurements or projection measurements provided by the projection measurement engine 22 . To this end, the 3D measurement predictor 23 may use either a simplified model, such as a linear model, or a probabilistic model, such as a Gaussian process model. While linear models are simpler and give generally satisfactory results in accordance with industry standards in terms of errors, probabilistic models allow you to supplement the prediction for each measurement with a measure of its uncertainty. Uncertainty can be used to predict estimation errors and trigger appropriate corrective actions to avoid giving users incorrect results and forecasts subject to very high variance.

Gaussian processes can effectively deal with noise, which is important in order to allow the collection of data on images taken in the field by users without special training.

As a rule, Gaussian processes can have the following form:

where m(x)=E[f(x)] is the expected value and k(x,x')=E[(f(x)-m(x))(f(x')-m(x') )] - covariance function.

The function modeled by the Gaussian process is a function that predicts three-dimensional measurements of the body, starting from the length of their projections on the plane, as provided by the means of taking projection measurements 22 . The covariance function allows for a non-linear mapping and gives flexibility in the specific form of non-linearity that is allowed. Due to their mathematical properties, covariance functions can be combined together to produce new ones that exhibit the desired characteristics. In one embodiment, the sum of 3 different covariance functions is:

is a constant covariance function used to model non-zero means for various variables

,

,

- linear covariance function

,

,

where x and x' represent two common points, and the index d indicates the D dimensions of the input space,

- quadratic exponential covariance function, which allows obtaining non-linear smooth mappings

,

,

where x and x' represent two common points, and h is a characteristic length scale, acting as a scaling factor.

As an output, the 3D sizing predictor 23 provides a data set DS that contains estimates of the actual (3D) size of the respective body parts of the person being cropped.

Finally, it is obvious that changes and variations can be made to the described and illustrated device and method without going beyond the scope of the legal protection of the appended claims.

Claims (32)

1. A portable device for obtaining anthropometric data, containing an image sensor (7) and a processing unit (5), and the processing unit (5) contains: an acquisition module (8) configured to receive first images (IMG _B , IMG _S ) of a person from the image sensor (7); an image processor (10) configured to create second images (FIMG _B , FIMG _S ) from the respective first images (IMG _B , IMG _S ) by applying an edge detection procedure; communication module (11) configured to send second images (FIMG _B , FIMG _S ), different in that the image processing processor (10) comprises a first edge detector (16) configured to perform a first edge extraction, a second edge detector (17) configured to perform a second edge extraction, and the first images (IMG _B , IMG _S ) are transmitted in parallel to the first border detector (16) and to the second border detector (17); and the image processor (10) comprises a contour recovery module (18) configured to combine the results of the first contour extraction performed by the first edge detector (16) and the second contour extraction performed by the second edge detector (17); wherein the contour recovery module (18) is configured to assign a label (1) that minimizes the energy function E(l) for each area in the first images (IMG _B , IMG _S ), and the energy function is characterized by the presence of the first component E _UNARY (l) and the second component E _PAIRWISE (l) and is defined as follows: E(l)=E _UNARY (l)+E _PAIRWISE (l)

Where

w ₁ and w ₂ - weight parameters

Where

aw _P - weight parameter. 2. The device according to claim. 1, containing a display (6) operating in real time, in which the output signals of the image sensor (7) are continuously displayed on the display (6), and the acquisition module (8) is additionally configured to overlay visual information on the real-time images on the display (7) to guide the person into a pose within a range of acceptable poses. 3. The apparatus of claim 2, wherein the visual information provides for a first mask (13) and a second mask (14), and the acceptable pose range provides for first human poses that fit the first mask (13) and second human poses that fit to the second mask (14), wherein the first mask (13) requires the person to take a standing position with legs apart and hands away from the body, and the second mask (14) requires the person to take a standing position with arms along the torso. 4. The device according to claim 3, in which the image processor (10) is configured to clean parts of the first images (IMG _B , IMG _S ) corresponding to areas outside one of the first mask (13) and the second mask (14). 5. The device according to any one of paragraphs. 1-4, wherein the first edge detector (16) is a Canny edge detector. 6. The device according to any one of paragraphs. 1-5, wherein the second edge detector (17) is a structured edge detector. 7. The device according to any one of paragraphs. 1-6, in which the contour recovery module (18) is configured to combine the first output (I ₁ , I ₂ ') of the first edge detector (16) and the second output (I ₂ , I ₂ ') of the second edge detector (17) by addition, threshold selection and binarization. 8. The device according to claim 7, in which the contour recovery module (18) is configured to compare the sum of the brightness values of the first output (I ₁ , I ₁ ') and the second output (I ₂ , I ₂ ') with a threshold (τ) and determining the second images (FIMG _B , FIMG _S ) by assigning a first boolean value (L ₁ ) if the sum of the first output (I ₁ , I ₁ ') and the second output (I ₂ , I ₂ ') exceeds a threshold (τ), and second logical value (L ₂ ) otherwise. 9. System for obtaining anthropometric data, containing a server (3) and at least one portable device (2) for obtaining anthropometric data according to any of the previous paragraphs, connected with the possibility of communication with the server (3). 10. The system according to claim 9, in which the server (3) contains a communication module (20) configured to receive second images (FIMG _B , FIMG _S ) from a portable device (2) for obtaining anthropometric data, a detector (21) of key points , configured to identify key points in the second images (FIMG _B , FIMG _S ) and their respective positions based on the contour line thickness, the key points including crown, heels, armpits, perineum. 11. The system according to claim 10, in which the server (3) contains a means (22) for taking projection measurements, configured to identify points of the body contour belonging to the torso, based on the positions of the armpits and perineum, determined by the detector (21) key points. 12. The system according to claim 11, in which the means (22) for taking projection measurements is made with the possibility of approximating the lines of the torso with polynomial functions. 13. The system according to claim 11 or 12, in which the means (22) for taking projection measurements is additionally configured to measure the projections of the selected main body dimensions on the image plane or on a plane perpendicular to it. 14. The system according to claim 11 or 12, in which the means (22) for taking projection measurements is additionally configured to take measurements of the chest, waist and hips in fixed and predetermined vertical coordinates or take measurements of the chest in the widest part of the upper third of the body, measurements waist in the smallest part of the central third of the body and measurements of the hips in the widest part of the lower third of the body. 15. The system according to any one of paragraphs. 10-14, wherein the server (3) comprises a 3D measurement predictor (23) configured to provide estimates or predictions of actual body measurements based on measurements provided by the projection measurement engine (22) using a linear or probabilistic model.

Images