RGBD-based single-view human body measurement method, device and computer-readable storage medium
Through the RGBD-based single-view anthropometric measurement method, the RGBD data acquisition system and FPGA hardware acceleration are used to solve the problem of high cost and low accuracy of single-view anthropometric human parameters measurement, and fully automatic and low-cost high-precision human parameters are achieved. Measurement.
Patent Information
- Application Number
- CN201911213033.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-12-02
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2039-12-02
AI Technical Summary
The measurement of single-view human parameter has problems such as high equipment cost, low accuracy and complex operation, making it difficult to obtain a high-quality human body 3D point cloud model.
Using a single-view anthropometric method based on RGBD, images of multiple angles are obtained through depth cameras and RGB cameras, the position of the human body is determined, the model is established and the human body parameters are obtained, and the depth data is achieved using Zynq's RGB-D data acquisition system and FPGA hardware acceleration. Compute and JPEG compression to improve accuracy and frame rate.
On the basis of maintaining high accuracy, it realizes automatic acquisition of human body parameters, such as neck circumference, waist circumference, hip circumference and wingspan. It is low cost and simple to operate, and is suitable for single-view anthropometric measurement.
Smart Images

Figure CN111008966B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of image technology, and in particular to a method and device for measuring human body from a single perspective based on RGBD, and a computer-readable storage medium. Background Art
[0002] Human body parameter measurement has always been one of the important fields of computer vision, with a huge number of application scenarios. According to the number of cameras used, human body parameter measurement can be divided into single-view and multi-view. Multi-view human body parameter measurement mainly uses multiple calibrated cameras to obtain depth data from multiple angles of the user at the same time, so that a relatively accurate human body point cloud map can be obtained to calculate various parameters of the human body. The measurement solution based on a single view requires the user or the measuring device to rotate to obtain human body data from various angles to calculate human body parameters.
[0003] Multi-view human body parameter measurement requires multiple devices, expensive systems, complex calibration processes, and large space requirements, making it difficult to apply on a large scale. Measurement solutions based on a single view are often cheaper, with a simple calibration process and more flexible applications. However, single-view solutions have more technical challenges than multi-view solutions. First, the depth map obtained from a single consumer-level depth acquisition device has a lot of noise and low accuracy. Second, single-view solutions cannot obtain data from all angles of the human body at the same time, and require the device to rotate or the user to rotate. During the measurement process, it is difficult for the user to avoid slight shaking. These two difficulties make it difficult for single-view solutions to obtain high-quality 3D point cloud models of the human body, and thus require more complex algorithms to obtain human body parameters. Summary of the invention
[0004] The present application provides a single-view human body measurement method, device and computer-readable storage medium based on RGBD, which can solve the problems that the current human body parameter measurement uses multiple devices, which requires relatively high costs and a complicated process, and the use of a single device for measurement has relatively large noise, low accuracy and requires auxiliary equipment for shooting, which is not simple enough.
[0005] Wherein, the technical solution adopted by the present invention is:
[0006] A single-view human body measurement method based on RGBD, the human body measurement method comprising: acquiring images: respectively acquiring a depth map and a color map of a human body to be measured in a set direction through a depth camera and an RGB camera; locating a measurement position: determining a set part of the human body in the depth map and the color map of the set direction as the measurement position; modeling the measurement position: modeling the measurement position according to a preset model and information of the depth map and the color map corresponding to the measurement position; and acquiring human body parameters: acquiring the human body parameters of the measurement position according to the modeling result.
[0007] A single-view human body measurement device based on RGBD, the device comprising: an image acquisition module, used for respectively acquiring a depth map and a color map of a human body to be measured in a set direction through a depth camera and an RGB camera; a positioning measurement position module, used for determining a set part of the human body in the depth map and the color map of the set direction as a measurement position; measurement position modeling, used for modeling the measurement position according to a preset model and information of the depth map and the color map corresponding to the measurement position; and human body parameter acquisition, used for acquiring the human body parameters of the measurement position according to the modeling result.
[0008] A computer-readable storage medium stores a computer program, which implements the human body measurement method as described above when executed by a processor.
[0009] The beneficial effects of the present application are as follows: the present application determines the positions of various parts of the human body in pictures at various angles by acquiring depth maps and color maps at multiple angles, extracts key information of each measurement position for modeling, obtains the shape information of the cross section of each measurement position, and thus obtains the parameters of each measurement position. While maintaining a high measurement accuracy, the parameters of the human body's neck circumference, waist circumference, hip circumference, wrist circumference and arm span can be automatically obtained based on a single perspective, and the cost is relatively low and the operation is relatively simple. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0011] Figure 1 A hardware structure block diagram of an electronic device is shown according to an exemplary embodiment.
[0012] Figure 2 A flowchart of a single-view human body measurement method based on RGBD is shown according to an exemplary embodiment.
[0013] Figure 3 is based on Figure 2 The depth map and color map obtained from the capture.
[0014] Figure 4 yes Figure 2 A flowchart of the stereo calibration correction steps in one embodiment.
[0015] Figure 5 yes Figure 2 The corresponding embodiment is a flowchart of step S230 in an embodiment.
[0016] Figure 6 It is a schematic diagram of the effect of human body segmentation in step S230.
[0017] Figure 7 It is a schematic diagram of the effect of measuring the position positioning in step S230.
[0018] Figure 8 yes Figure 5 The corresponding embodiment is a flowchart of step S233 in an embodiment.
[0019] Fig. 9 yes Figure 8 The flowchart of step S2331 in the corresponding embodiment in one embodiment.
[0020] Fig.10 yes Figure 2 The corresponding embodiment is a flowchart of step S230 in an embodiment.
[0021] Fig.11 yes Fig.10 A schematic diagram of the effect of modeling in step S251 in the corresponding embodiment.
[0022] Fig.12 yes Fig.10 Schematic diagram of the effect of performing model correction in step S252 in the corresponding embodiment.
[0023] Fig.13 It is a measurement schematic diagram of the RGBD-based single-view human body measurement method of the present application.
[0024] Fig.14 It is a principle block diagram of single-view human body measurement based on RGBD according to an exemplary embodiment.
[0025] The above-mentioned drawings have shown clear embodiments of the present invention, which will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0026] Here, exemplary embodiments will be described in detail, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.
[0027] As mentioned above, the prior art also uses multi-viewing angles or single-viewing angles to measure human body parameters. However, no matter whether the measurement is performed from a single-viewing angle or from multiple-viewing angles, the following problems exist:
[0028] First, if multi-viewing angles are used to measure human body parameters, multiple devices are required to shoot simultaneously, which is costly, complicated to implement, and difficult to apply on a large scale.
[0029] Second, if a single perspective is used to measure human body parameters, the accuracy is not high and other mechanical equipment or manual assistance of single-perspective equipment is required for shooting.
[0030] In order to overcome the shortcomings of the prior art, the present invention proposes a single-view human body parameter measurement algorithm based on RGB-D (three-dimensional depth data plus two-dimensional texture data) and an RGB-D data acquisition system based on Zynq, realizing a convenient and flexible system that can measure human body parameters in complex environments. This RGBD-based single-view human body measurement method is implemented by a computer program, and correspondingly, the constructed face recognition device can be stored in an electronic device with a von Neumann system architecture to be executed in the electronic device, thereby realizing face recognition. For example, the electronic device can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a server, etc., which is not limited here.
[0031] Figure 1 is a hardware structure framework diagram of an electronic device according to an exemplary embodiment. It should be noted that the electronic device is only an example adapted to the present invention and cannot be considered to provide any limitation on the scope of use of the present invention. The electronic device cannot be interpreted as needing to rely on or must have Figure 1 One or more components of an exemplary electronic device shown in FIG.
[0032] The hardware structure of the electronic device may vary greatly due to different configurations or performances, such as Figure 1 As shown, the electronic device 100 includes: a power supply 110 , an interface 130 , at least one memory 150 , a camera sensor 170 , and at least one central processing unit (CPU) 190 .
[0033] The power supply 110 is used to provide operating voltage for various hardware devices on the electronic device 100 .
[0034] The interface 130 includes at least one wired or wireless network interface 131 , at least one serial-to-parallel conversion interface 133 , at least one input-output interface 135 , and at least one USB interface 137 , etc., for communicating with external devices.
[0035] The memory 150 is a carrier for storing resources, which may be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon include an operating system 151, an application 153 and data 155, etc. The storage method may be temporary storage or permanent storage. Among them, the operating system 151 is used to manage and control the hardware devices and application 153 on the electronic device 100 to enable the central processor 190 to calculate and process the massive data 155. It may be WindowsServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc. The application 153 is a computer program that performs at least one specific task based on the operating system 151, which may include at least one module ( Figure 1 Each module may include a series of computer-readable instructions for the electronic device 100. The data 155 may be photos, pictures, etc. stored in a disk.
[0036] The camera sensor 170 includes: a 3D sensor and a 2D sensor. The 3D sensor is used to capture 3D images, and the 2D sensor is used to capture 2D images.
[0037] The central processing unit 190 may include one or more processors and is configured to communicate with the memory 150 via a bus for computing and processing the mass data 155 in the memory 150 .
[0038] The central processor 190 of this application selects Xilinx's Zynq7Z020 as the data acquisition platform, and implements the configuration of 3D sensors and 2D sensors on the FPGA, as well as data acquisition cache. For 3D sensors, depth calculation, nonlinear correction, temperature compensation and grayscale compensation are implemented on the FPGA to improve the accuracy of depth data; for 2D sensors, JPEG (Joint Photographic Experts Group) compression is implemented on the FPGA to improve the acquisition and transmission of high-resolution RGB data. This article uses the characteristics of FPGA hardware acceleration to realize the calculation of depth data and the compression and optimization functions of JPEG. Compared with other embedded devices, it is extremely fast and has a high frame rate. At the same time, on the PS (Processing System) end of Zynq7z020, we implement the control of each IP (Intellectual Property) on the PL (Progarmmable Logic) end, acquire RGB-D data, build a TCP (Transmission Control Protocol) server, and realize communication with the computer host and transmission of RGB-D data.
[0039] The Zynq-based RGB-D data acquisition system can quickly acquire image data with a frame rate of 25fps for depth data and color image data.
[0040] As described in detail above, the electronic device 100 applicable to the present invention will complete the human body parameter measurement method by having the central processor 190 read a series of computer-readable instructions stored in the memory 150 .
[0041] In addition, the present invention can also be implemented by hardware circuits or hardware circuits combined with software. Therefore, the implementation of the present invention is not limited to any specific hardware circuits, software, or a combination of the two.
[0042] See also Figure 2 In an exemplary embodiment, a single-view human body measurement method based on RGBD is applicable to an electronic device, and the structure of the electronic device can be as follows Figure 1 shown.
[0043] The human body measurement method can be performed by an electronic device and may include the following steps:
[0044] Step S210: Acquire an image: acquire a depth image and a color image of a set direction of a human body to be measured through a depth camera and an RGB camera respectively.
[0045] In this embodiment, the depth camera and the RGB camera respectively obtain the depth map and color map of the front, back, and left side of the human body to be measured. The depth map includes a depth point cloud map and a depth pseudo-color map. Figure 3 , Figure 3 From left to right are depth pseudo-color image, 2D color image, and depth point cloud image.
[0046] Step S230: Locate the measurement position: determine a set part of the human body in the depth image and the color image in a set direction as the measurement position.
[0047] In this embodiment, the neck, waist, hips, arm span and wrist circumference of the human body can be used as measurement positions.
[0048] Step S250: Modeling the measurement position: Modeling the measurement position according to the preset model and the information of the depth map and color map corresponding to the measurement position.
[0049] Step S270: Obtaining human body parameters: obtaining human body parameters of the measurement position according to the modeling results.
[0050] Specifically, Figure 4 As shown, in a specific implementation of an embodiment, the following steps may be included before step S210: the step of performing stereo calibration correction on the image may also be included, and the step of performing stereo calibration correction on the image may include:
[0051] Step S211: Correct the image distortion caused by the lenses of the depth camera and the RGB camera. Correcting the image distortion can ensure that subsequent measurements are more accurate.
[0052] Step S212: establishing a depth map and a color map having a mutual mapping relationship.
[0053] In this embodiment, the depth map and the color map are mapped to each other, which means finding the RGB value on the color map corresponding to each pixel on the depth map.
[0054] In a specific implementation of an embodiment, step S211 may include the following steps:
[0055] Step S2111: Obtain pictures of the chessboard from various angles of the depth camera and the RGB camera.
[0056] Step S2112: Perform chessboard corner point monitoring on the acquired image to obtain the intrinsic parameter matrix and extrinsic parameter matrix of the depth camera and the RGB camera.
[0057] Step S2113: Read the internal parameter matrix and the external parameter matrix. In this step, the internal parameter matrix and the external parameter matrix are read through the opencv function.
[0058] Step S2114: The images acquired by the depth camera and the RGB camera are subjected to coordinate transformation by using the parameters of the read intrinsic parameter matrix and extrinsic parameter matrix to perform lens distortion correction.
[0059] In a specific implementation of an embodiment, the specific mapping process of step S212 is as follows: Step S2121: First, since the viewing angles of the two cameras are different, the relationship between the coverage ranges of the two cameras and the depth can be found through calibration.
[0060] Specifically, two cameras simultaneously shoot white walls from 1m to 2.6m, and save amplitude maps, color maps, and depth data every 20cm. Record the top horizontal coordinate t, bottom horizontal coordinate b, left vertical coordinate l, right vertical coordinate r, and depth data d of the same field of view in the amplitude map and color map obtained each time. The amplitude map data is recorded as (t0, b0, l0, r0, d), and the color map data is recorded as (t1, b1, l1, r1, d), where t, b, l, r, d are all vectors, and the length of the vector is the number of saved data, that is, the length is 9. By fitting the data of each vector with a linear fit. The relationship between the coverage of the two cameras and the depth. For example, the horizontal coordinate T0 of the top within the coverage of the depth camera is a*d+b; where a and b are the parameters obtained by fitting the previous t0 vector, and d is the distance measured in practice, and the other parameters are the same. Get the top horizontal coordinate, bottom horizontal coordinate, left vertical coordinate, and right vertical coordinate of the coverage of the two cameras, and then you can cut according to the above.
[0061] Step S2122: Then cut out the same coverage areas in the two cameras.
[0062] Step S2123: Finally, both images are scaled to a size of 1280*720 through bilinear interpolation.
[0063] Specifically, Figures 5 to 7 As shown, in a specific implementation of an embodiment, step S230 may include the following steps:
[0064] Step S231: Roughly segment the human body region in the depth map. In this embodiment, the depth map is roughly segmented by region growing and threshold segmentation algorithms to obtain the human body region of the approximate depth map. Among them, the region growing and threshold segmentation algorithms are roughly to place a seed in the middle of the depth map for region growing, and set the region growing threshold to 20, which means that the pixels within 20 cm from the center pixel and with continuous positions are segmented out by the threshold segmentation algorithm.
[0065] Step S232: Mapping the roughly segmented depth map to the color map. In this embodiment, the depth map roughly segmented from the depth map is mapped to the color map according to the mapping relationship between the depth map and the color map established in step S212 to obtain a rough human body region in the color map.
[0066] Step S233: finely segment the mapped color image. Figure 6 , Figure 6The upper left picture in the figure is the depth map captured by the 3D sensor, the upper right picture is the color map captured by the RGB sensor, the lower left picture is the situation of roughly segmenting the human body area in the depth map and mapping it to the color map, and the lower right picture is the situation after fine segmentation of the lower left picture. It can be seen that after the above steps, the human body area can be completely and quickly separated from the image.
[0067] Step S234: Determine the measurement position of the human body area according to the finely segmented color image.
[0068] In this embodiment, the measurement position of the human body area can be determined based on human body proportions, wherein the measurement positions include: neck, waist and hips. According to scientific statistics, the ratio of the position of the neck to the human body height is 0.14, the ratio of the position of the waist to the human body height is 0.39, and the ratio of the position of the hips to the human body height is 0.47. Therefore, if the measurement position is the neck, the proportion of the relative human body height in the cut human body area is found to be at the position of 0.14. For example, when the human body height is 2 meters, the position of 0.28 meters starting from the top of the human body area is the position of the neck. Similarly, the positions of the waist and hips can also be located in the human body area in this way. The position of the wrist circumference is determined specifically by scanning the arm in the human body area from left to right, and it will be found that among all the cross-sectional diameters, the cross-sectional diameter at the wrist is at the minimum value of all the cross-sectional diameters. The position of the arm span is determined specifically by scanning the segmented human body contour from left to right, and the maximum value of all the cross-sectional diameters is the arm span. Please refer to Figure 7 , Figure 7 It is the result of measuring position and finding the positions of the neck, waist, hips, wrist circumference and arm span in the human body area.
[0069] Specifically, Figure 8 As shown, in a specific implementation of an embodiment, in step S233, the mapped color image can be finely segmented based on the optimized Grabcut algorithm. Step S233 can specifically include the following steps:
[0070] Step S2331: Mark the foreground and background in the color image obtained after mapping to obtain a mixed Gaussian model of the foreground and background.
[0071] Step S2332: Perform fine segmentation of the foreground and background using the maximum flow algorithm to obtain a finely segmented human body region.
[0072] Specifically, Fig. 9 As shown, in a specific implementation of an embodiment, step S2331 may include the following steps:
[0073] Step S2331a: Map the result of the rough segmentation onto the color map, mark the area that can be mapped onto the color map as the foreground, and then divide the pixels of the foreground into a set number of groups by the kmeans (k-means clustering algorithm), and obtain a mixed Gaussian model for each group. In this way, the mixed Gaussian model of the foreground can be obtained. In this embodiment, the set number of groups is 5, so 5 mixed Gaussian models can be obtained.
[0074] Step S2331b: The result of the rough segmentation is expanded to a set size and mapped to the color map, and the unmapped area can be marked as the background. Similarly, the background pixels are divided into a set number of groups through the kmeans algorithm, and each group obtains a mixed Gaussian model, so that the mixed Gaussian model of the background can be obtained. In this embodiment, the kernel for the expansion operation is set to a size of 3*3, and the set number of groups is 5, so that 5 mixed Gaussian models can be obtained.
[0075] Step S2331c: Finally, the marked edges are left that have not been classified as foreground or background. The probability of each pixel belonging to the Gaussian model of the foreground and the probability of each pixel belonging to the Gaussian model of the background are calculated, and the pixel is classified as foreground or background according to the probability. In other words, the probability of each pixel belonging to the 5 Gaussian models of the foreground and the probability of each pixel belonging to the 5 Gaussian models of the background are calculated, and the pixel is classified as foreground or background according to the probability.
[0076] This step is optimized compared to the traditional Grabcut algorithm, achieving fully automatic and accurate extraction of human body contour information in general scenes, making the segmentation of human body areas more accurate.
[0077] Specifically, Figures 10 to 12 As shown, in a specific implementation of an embodiment, step S250 may include the following steps:
[0078] Step S251: Acquire and establish a preset model based on the depth information of the set position of the measurement position. In this embodiment, a rhombus is established by acquiring the boundary information of the front and side surfaces of each measurement position, that is, according to the depth information and pixel information of each measurement position, the midpoint of each measurement position on each surface is selected as the four corner points of the rhombus. For example, refer to Fig.11 , a coordinate system is established for the measurement position hip. Two coordinate points of the rhombus (0, -50) and (0, 50) can be determined based on the pixel width of the hip position in the color image obtained from the side and the measured depth information. Then, the other two coordinate points of the rhombus (100, 0) and (-100, 0) can be determined based on the pixel width of the hip position in the color image obtained from the front and the measured depth.
[0079] Step S252: extract and modify the preset model according to the key sparse points of the measurement position to obtain the human body model of the measurement position. Specifically, the key sparse points of each measurement position are selected, added to the diamond model, and then binomial interpolation is performed, so that an accurate human body part model from a top-down angle can be formed. The expression of the model is:
[0080] Among them, x is the point on the horizontal coordinate of the diamond model, and y is the point on the vertical coordinate of the diamond model.
[0081] See also Fig.12 In this embodiment, the method for obtaining key sparse points is to uniformly take 7 depth values of the measurement position as the key sparse points of the measurement position.
[0082] This step can be used to analyze and model each part of the RGBD data at three angles, avoid point cloud alignment operations, and solve the problem of non-rigid changes in the human body.
[0083] Therefore, step S270 includes: integrating the curve of the human body model to obtain the perimeter L of the human body model. The curve integral expression is:
[0084]
[0085] See also Fig.13 , Fig.13 This is a schematic diagram of the results of measuring a human body using the human body measurement method of the present application. It can be seen that the measurement results are 174.14cm in height, 74.64cm in waist circumference, 42.73cm in neck circumference, 85.83cm in hip circumference, and 173.34cm in arm span. The measurement results are consistent with the actual situation, and after verification, the final measurement accuracy is 2.08cm, which shows that the measurement results are relatively accurate.
[0086] See also Fig.14 In an exemplary embodiment, a single-view human body measurement device 500 based on RGBD includes but is not limited to: an image acquisition module 510, a measurement position positioning module 530, a measurement position modeling module 550 and a human body parameter acquisition module 570.
[0087] An image acquisition module 510 is used to acquire a depth image and a color image of a set direction of a human body to be measured through a depth camera and an RGB camera respectively;
[0088] A positioning measurement position module 530 is used to determine a set part of the human body in the depth image and the color image of the set direction as the measurement position;
[0089] Measurement position modeling 550, for modeling the measurement position according to a preset model and information of a corresponding depth map and color map of the measurement position;
[0090] Obtaining human body parameters 570 is used to obtain the human body parameters of the measurement position according to the modeling result.
[0091] It should be noted that, when the single-view anthropometry device provided in the above embodiment performs face recognition processing, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the single-view anthropometry device will be divided into different functional modules to complete all or part of the functions described above.
[0092] In addition, the single-view anthropometry device and the single-view anthropometry method provided in the above embodiments belong to the same concept, and the specific manner in which each module performs operations has been described in detail in the method embodiments, which will not be repeated here.
[0093] In an exemplary embodiment, a single-view human body measurement device includes a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the single-view human body measurement method based on RGBD in the above-mentioned embodiments is implemented.
[0094] In an exemplary embodiment, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the single-view human body measurement method in the above-mentioned embodiments.
[0095] The beneficial effects of the present application are as follows: the present application determines the position of each part of the human body in the pictures at each angle by acquiring depth maps and color maps at multiple angles, extracts key information of each measurement position for modeling, and acquires the shape information of the cross section of each measurement position according to the model obtained by modeling, thereby obtaining the parameters of each measurement position. While maintaining a high measurement accuracy, the parameters of the human body's neck circumference, waist circumference, hip circumference, wrist circumference and arm span can be automatically obtained based on a single perspective, and the cost is relatively low, the operation is relatively simple, and it is convenient for users to operate.
[0096] The above contents are only preferred exemplary embodiments of the present invention and are not intended to limit the implementation scheme of the present invention. A person skilled in the art can easily make corresponding changes or modifications based on the main concept and spirit of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope required by the claims.
Claims
1. A single-view human body measurement method based on RGBD, characterized in that: The human body measurement method comprises: Image acquisition: A depth image and a color image of a human body to be measured in a set direction are acquired through a depth camera and an RGB camera respectively; Positioning the measurement position: determining a set part of the human body in the depth map and the color map of the set direction as the measurement position; Measurement position modeling: Modeling the measurement position according to a preset model and information of a depth map and a color map corresponding to the measurement position; Obtaining human body parameters: obtaining human body parameters at the measurement position according to the modeling result; Before the step of acquiring the image, a step of performing stereo calibration and correction on the image is also included, and the step of performing stereo calibration and correction on the image includes: Correcting image distortion caused by the lenses of the depth camera and the RGB camera; Establishing a depth map and a color map with mutual mapping relationship; The step of locating the measurement position includes: Roughly segmenting the human body region in the depth map; mapping the roughly segmented depth map to the color map; finely segmenting the mapped color map; and determining the measurement position of the human body region according to the finely segmented color map; The step of performing human body segmentation on the depth map and the color map includes: The depth map is roughly segmented by using a region growing and threshold segmentation algorithm to obtain a rough human body region; Mapping the roughly segmented human body region to the color image; marking the foreground and background of the mapped color image to obtain a mixed Gaussian model of the foreground and background; segmenting the foreground and background using a maximum flow algorithm to obtain the finely segmented human body region; The step of measuring position modeling includes: The result obtained by the rough segmentation is mapped onto the color image, and the area mapped onto the color image is marked as the foreground. Then, the pixels of the foreground are divided into a set number of groups by using the kmeans algorithm, and a mixed Gaussian model is obtained for each group. The result obtained by the rough segmentation is expanded to a set size and then mapped to the color image, and the area not mapped to the color image is marked as the background. The pixels of the background are also divided into a set number of groups by the kmeans algorithm, and a mixed Gaussian model is obtained for each group; The probability that each pixel belongs to the Gaussian model of the foreground and the probability that each pixel belongs to the Gaussian model of the background are calculated, and the pixel is classified into the foreground or the background according to the probability.
2. The single-view human body measurement method according to claim 1, characterized in that: The step of measuring position modeling includes: Acquiring and establishing a preset model according to the depth information of the set position of the measurement position; The preset model is extracted and modified according to the key sparse points of the measurement position to obtain a human body model of the measurement position.
3. The single-view human body measurement method according to claim 1, characterized in that: The step of obtaining human body parameters includes: Acquiring depth information of the midpoint of each of the measurement positions to establish a diamond model; The key sparse points of each measurement position are extracted, and the diamond model is corrected by using the key sparse points through binomial interpolation to form a human body model of the measurement position. The expression of the human body model is: Among them, x is the point on the horizontal coordinate of the diamond model, and y is the point on the vertical coordinate of the diamond model.
4. The single-view human body measurement method according to claim 2, characterized in that: The step of obtaining human body parameters includes: By integrating the curve of the human body model, the circumference of the human body model can be obtained. The curve integral expression is:
5. A single-view human body measurement device based on RGBD, characterized in that: The device comprises: An image acquisition module is used to acquire a depth image and a color image of a human body to be measured in a set direction through a depth camera and an RGB camera respectively; A positioning measurement position module, used to determine a set part of the human body in the depth map and the color map of the set direction as the measurement position; A measurement position modeling module, used to model the measurement position according to a preset model and information of a depth map of the measurement position; A human body parameter acquisition module, used for acquiring the human body parameters of the measurement position according to the modeling result; The image acquisition module is further used to perform stereo calibration and correction on the image, and the step of performing stereo calibration and correction on the image includes: Correcting image distortion caused by the lenses of the depth camera and the RGB camera; Establishing a depth map and a color map with mutual mapping relationship; The positioning and measuring position module is also used for: Roughly segmenting the human body region in the depth map; mapping the roughly segmented depth map to the color map; finely segmenting the mapped color map; and determining the measurement position of the human body region according to the finely segmented color map; The step of performing human body segmentation on the depth map and the color map includes: The depth map is roughly segmented by using a region growing and threshold segmentation algorithm to obtain a rough human body region; Mapping the roughly segmented human body region to the color image; marking the foreground and background of the mapped color image to obtain a mixed Gaussian model of the foreground and background; segmenting the foreground and background using a maximum flow algorithm to obtain the finely segmented human body region; The measurement location modeling module is further used for: The result obtained by the rough segmentation is mapped onto the color image, and the area mapped onto the color image is marked as the foreground. Then, the pixels of the foreground are divided into a set number of groups by using the kmeans algorithm, and a mixed Gaussian model is obtained for each group. The result obtained by the rough segmentation is expanded to a set size and then mapped to the color image, and the area not mapped to the color image is marked as the background. The pixels of the background are also divided into a set number of groups by the kmeans algorithm, and a mixed Gaussian model is obtained for each group; The probability that each pixel belongs to the Gaussian model of the foreground and the probability that each pixel belongs to the Gaussian model of the background are calculated, and the pixel is classified into the foreground or the background according to the probability.
6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the human body measurement method according to any one of claims 1 to 4 is implemented.
Citation Information
Patent Citations
Human body posture estimation method
CN105389569A
Method for establishing human body model base and system thereof
CN106952335A
Image-based human body measuring method and system
CN107680130A
Human body three-dimensional reconstruction method and device based on RGBD single-view images
CN110335343A
Imaging a Body
US20170273639A1
