Three-dimensional imaging method and apparatus, and three-dimensional imaging device

Abstract

The present application relates to a kind of three-dimensional imaging method and device, and three-dimensional imaging equipment.The three-dimensional imaging method includes: three-dimensional information acquisition step, by depth camera, the three-dimensional image information of three-dimensional shooting area containing detection object is generated by shooting;Mask extraction step, the mask of the detection object is extracted from the three-dimensional image information;Imaging area determination step, according to the mask of the detection object, the imaging area about the detection object is determined;Holographic data acquisition step, by holographic data acquisition device, the holographic data acquisition area containing the detection object is generated by shooting, and holographic data is generated;And image reconstruction step, based on the holographic data, the imaging area is reconstructed to image.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of security inspection, in particular to a three-dimensional imaging method and device using millimeter waves, and a three-dimensional imaging apparatus. BACKGROUND

[0002] Currently, human body security inspection is urgent in specific occasions such as airports and stations.

[0003] Millimeter waves refer to electromagnetic waves with a frequency of 30 GHz to 300 GHz, which are particularly suitable for human body security inspection because they can penetrate clothes and are non-ionizing radiation.

[0004] In recent years, millimeter wave three-dimensional holographic imaging technology has been increasingly applied in the field of personal security inspection, greatly reducing the work burden of security personnel, and can be applied in customs, airports, courts and large security activity sites. It is a safe, civilized and efficient new mode of security inspection.

[0005] It is known that the active millimeter wave human body security inspection imaging technology uses wideband millimeter waves to irradiate the human body, and realizes direct measurement of holographic data through heterodyne mixing technology, and then reconstructs the complex reflectivity image. SUMMARY

[0006] In the image reconstruction process of the previous three-dimensional imaging method using active multi-transmission and multi-reception millimeter waves, the imaging area as the object to be image reconstructed is usually a rectangular frame containing the human body, for example, the range scanned by the antenna array. Within the rectangular frame, in addition to the human body mask (containing the human body and clothes), there are other pixel points. Such other pixel points are useless for the final judgment of whether the person carries dangerous goods. Therefore, in this case, the image reconstruction speed is slow, and all holographic data is needed for each pixel point reconstruction, which requires a huge amount of calculation, so a large amount of time and computing resources are wasted for each image reconstruction.

[0007] The present application provides a three-dimensional imaging system and method capable of reducing the imaging area and reducing the amount of calculation, and a control device.

[0008] According to the first mode of the present application, a three-dimensional imaging method is provided, comprising: a three-dimensional information acquisition step of capturing a three-dimensional capture area containing a detection object by a depth camera to generate three-dimensional image information; a mask extraction step of extracting a mask of the detection object from the three-dimensional image information; an imaging area determination step of determining an imaging area of the detection object according to the mask of the detection object; a holographic data acquisition step of acquiring holographic data of a holographic data acquisition area containing the detection object by a holographic data acquisition device to generate holographic data; and an image reconstruction step of reconstructing an image of the imaging area based on the holographic data.

[0009] In the first method described above, the mask extraction step applies an image segmentation algorithm to the three-dimensional image information to extract the mask of the object to be detected.

[0010] In the first approach described above, the image segmentation algorithm is the DeepLabV3+ segmentation algorithm based on neural networks.

[0011] In the first method described above, the image reconstruction step uses the backpropagation imaging algorithm to perform image reconstruction.

[0012] In the first method described above, the holographic data acquisition device includes a multi-transmitter, multi-receiver antenna array, which includes multiple pairs of transmitting and receiving antennas for transmitting and receiving millimeter waves. The backpropagation imaging algorithm includes: for each pixel in the imaging area, calculating the sum of the distances from that pixel to each pair of transmitting and receiving antennas; based on the holographic data received by the receiving antennas and the sum of the distances, summing the distances according to all transmitting and receiving antennas and the transmission frequency of the millimeter waves to obtain the complex reflectivity of each pixel in the imaging area; calculating the complex reflectivity for all pixels in the imaging area, and forming a reconstructed image based on the complex reflectivity.

[0013] In the first method described above, the backpropagation imaging algorithm includes: using formula (1), calculating the sum of the distances r from each pixel point (x, y, z) in the imaging region to each pair of transmitting and receiving antennas. T，R ,

[0014]

[0015] Based on the holographic data s(x) received by the receiving antenna T ,y T ,x R ,y R, (k), and the sum of distances r T，R Using formula (2), the complex reflectance of the pixel (x, y, z) in the imaging region is calculated.

[0016]

[0017] Among them, (x R y R (x, 0) represents the coordinates of the receiving antenna. T y T (x, y, z) represents the coordinates of the transmitting antenna, (x, y, z) represents the coordinates of a pixel in the imaging region, k is the wavenumber, k = 2π / λ, where λ is the wavelength.

[0018] For all pixels in the imaging area, the complex reflectance is calculated, and a reconstructed image is formed based on the complex reflectance.

[0019] In the first method described above, the holographic data acquisition device includes a multi-transmitter, multi-receiver antenna array, which includes multiple pairs of transmitting and receiving antennas for transmitting and receiving millimeter waves. The backpropagation imaging algorithm is a fast backpropagation imaging algorithm, which includes: calculating the contribution value of each pair of transmitting and receiving antennas to each pixel in the imaging area for each pair of transmitting and receiving antennas; summing the contribution values ​​for all pairs of transmitting and receiving antennas to obtain the complex reflectivity of each pixel in the imaging area; and forming a reconstructed image based on the complex reflectivity.

[0020] In the first method described above, for each pair of transmitting and receiving antennas, calculating the contribution value of the pair of transmitting and receiving antennas to each pixel in the imaging area includes: for each pair of transmitting and receiving antennas, calculating the sum of the distances from each pixel in the imaging area to the pair of transmitting and receiving antennas; and calculating the contribution value based on the sum of the distances and the holographic data received by the receiving antenna.

[0021] In the first method described above, the fast backpropagation imaging algorithm includes: using formula (3), for each pair of transmitting antennas T and receiving antennas R, calculating the sum of distances r from the pixel point (x, y, z) in the imaging region to the pair of transmitting antennas and receiving antennas respectively. T，R ,

[0022]

[0023] Using formula (4), based on the sum of distances r T，R and holographic data s(x) T ,y T ,x R ,y R, Calculate the contribution value P to pixel (x, y, z). T,R ,

[0024]

[0025] Based on contribution value P T,R Using formula (5), the complex reflectance of each pixel in the imaging region can be calculated.

[0026]

[0027] Where, r T，R S(x) is the sum of the distances from a pair of transmitting and receiving antennas to a pixel (x, y, z) in the imaging region. T ,y T ,x R ,y R,,k) is holographic data obtained through a pair of transmitting and receiving antennas, where k is the wave number, k=2π / λ, and λ is the wavelength;

[0028] A reconstructed image is formed based on complex reflectance.

[0029] In the first method described above, for each pair of transmitting and receiving antennas, calculating the contribution value of the pair of transmitting and receiving antennas to each pixel in the imaging area includes: among the pixels in the imaging area, grouping the pixels with the same sum of distances to the pair of transmitting and receiving antennas into a group; calculating the contribution value of the pair of transmitting and receiving antennas to any pixel in the group, which is then used as the contribution value of each pixel in that group.

[0030] In the first method described above, for each pair of transmitting and receiving antennas, calculating the contribution of the pair of transmitting and receiving antennas to each pixel in the imaging area includes: calculating the minimum and maximum sum of distances from all pixels in the imaging area to the pair of transmitting and receiving antennas; dividing the distance between the minimum and maximum values ​​into N equal intervals to obtain the equal sums of the N distances (r1, r2, ..., r N ); Calculate the sum of distances from any pixel in the imaging region to a pair of transmit and receive antennas, determine the nearest equal division value to the sum of distances, and use the nearest equal division value as the sum of distances for that pixel; for all pixels in the imaging region, determine the nearest equal division value; use N equal division values ​​as the sum of distances respectively, and calculate the contribution value.

[0031] The first method described above also includes: a key point determination step, which determines the core key points of the detection object based on a deep learning model; and an imaging region adjustment step, which selects the parts of the detection object related to the core key points in the imaging region based on the core key points, thereby adjusting the imaging region.

[0032] The first method mentioned above also includes: a key point determination step, which determines the core key points of the detected object based on a deep learning model; and a detected object pose judgment step, which judges whether the pose of the detected object is correct based on the position of the core key points, and if it is incorrect, an alarm is triggered.

[0033] According to a second aspect of the present invention, a three-dimensional imaging system is provided, characterized in that it includes a depth camera, a holographic data acquisition device, and a control device. The control device includes: a three-dimensional information acquisition module, which controls the depth camera to capture a three-dimensional imaging area containing a detection object and acquire three-dimensional image information; a mask extraction module, which extracts a mask of the detection object from the three-dimensional image information; an imaging area determination module, which determines the imaging area of ​​the detection object based on the mask of the detection object; a holographic data acquisition module, which controls the holographic data acquisition device to acquire data from the holographic data acquisition area containing the detection object and acquire holographic data; and an image reconstruction module, which reconstructs an image of the imaging area based on the holographic data.

[0034] In the second method described above, the holographic data acquisition device includes a multi-transmitter / multi-receiver antenna array, which includes multiple pairs of transmitting antennas and receiving antennas for transmitting and receiving millimeter waves.

[0035] According to a third aspect of the present invention, a three-dimensional imaging device is provided, comprising: a three-dimensional image information acquisition module for acquiring three-dimensional image information of a three-dimensional imaging area containing a detection object captured by a depth camera; a mask extraction module for extracting a mask of the detection object from the three-dimensional image information; an imaging area determination module for determining an imaging area about the detection object based on the mask of the detection object; a holographic data acquisition module for acquiring holographic data of a holographic data acquisition area containing the detection object obtained by a holographic data acquisition device; and an image reconstruction module for reconstructing an image of the imaging area based on the holographic data.

[0036] According to a fourth aspect of the present invention, a computer-readable medium is provided, storing a computer program, characterized in that, when executed by a processor, the computer program implements: a three-dimensional image information acquisition step, acquiring three-dimensional image information obtained by capturing a three-dimensional imaging area containing a detection object through a depth camera; a mask extraction step, extracting a mask of the detection object from the three-dimensional image information; an imaging region determination step, determining an imaging region about the detection object based on the mask of the detection object; a holographic data acquisition step, acquiring holographic data obtained by acquiring data from a holographic data acquisition area containing the detection object through a holographic data acquisition device; and an image reconstruction step, reconstructing an image of the imaging region based on the holographic data.

[0037] According to a fifth aspect of the present invention, a three-dimensional imaging device is provided, comprising a memory, a processor, and program instructions stored in the memory and executable by the processor. The computer device is characterized in that, when the processor executes the program instructions, it implements: a three-dimensional image information acquisition step, acquiring three-dimensional image information obtained by capturing a three-dimensional imaging area containing a detection object using a depth camera; a mask extraction step, extracting a mask of the detection object from the three-dimensional image information; an imaging region determination step, determining an imaging region related to the detection object based on the mask of the detection object; a holographic data acquisition step, acquiring holographic data obtained by acquiring data from a holographic data acquisition area containing the detection object using a holographic data acquisition device; and an image reconstruction step, reconstructing an image of the imaging region based on the holographic data.

[0038] In this invention, a depth camera is used to acquire 3D image information. An image segmentation algorithm is applied to this 3D image information to extract a human body mask, and the imaging region is determined based on this mask. This method can determine a 3D human body mask that conforms to the human body contour, reducing the computational load of image reconstruction. Furthermore, by utilizing a fast backpropagation imaging algorithm to reconstruct the image from this imaging region, the image reconstruction speed can be improved. Attached Figure Description

[0039] Figure 1 This is a schematic diagram illustrating the modular structure of a three-dimensional imaging system according to the first embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram showing an electronically scanned millimeter-wave imaging device;

[0041] Figure 3 This is a schematic diagram illustrating a mechanically scanning millimeter-wave imaging device;

[0042] Figure 4 This is a flowchart illustrating the three-dimensional imaging method according to the first embodiment of the present invention;

[0043] Figure 5 This is a flowchart illustrating an example of the fast backpropagation algorithm;

[0044] Figure 6 This is a schematic diagram used to illustrate the fast backpropagation algorithm;

[0045] Figure 7 This is a flowchart illustrating an example of the fast backpropagation algorithm;

[0046] Figure 8 This is a flowchart illustrating an example of the fast backpropagation algorithm;

[0047] Figure 9 This is a flowchart illustrating a three-dimensional imaging method according to a second embodiment of the present invention. Detailed Implementation

[0048] Hereinafter, with reference to the accompanying drawings, the three-dimensional imaging system and three-dimensional imaging method according to embodiments of the present invention will be specifically described. In the following description, the same or similar components will be labeled with the same or similar reference numerals.

[0049] <First Implementation Method>

[0050] Figure 1 This illustrates a three-dimensional imaging system 100 according to a first embodiment of the present invention. For example... Figure 1 As shown, the three-dimensional imaging system 100 includes a millimeter-wave imaging device 10 and a control device 20. The millimeter-wave imaging device 10 includes a depth camera 11 and a holographic data acquisition device 30.

[0051] The millimeter-wave imaging device 10 performs imaging by illuminating a holographic data acquisition area containing the object to be detected with millimeter waves and receiving the reflected millimeter waves. Here, the millimeter-wave imaging device 10 employs active millimeter-wave scanning. The objects to be detected include people, animals, and goods that are subject to security checks. The holographic data acquisition area is the area capable of holographic data acquisition, such as the area covered by an antenna array that transmits and receives millimeter waves.

[0052] Millimeter-wave imaging device 10 utilizes the penetrating power of millimeter waves through ordinary clothing to create an image. Millimeter waves illuminating the human body are reflected after passing through clothing and encountering objects such as the body surface or other concealed objects. By receiving the reflected millimeter waves, holographic data can be acquired. Based on this holographic data, image reconstruction can be performed to determine whether dangerous items such as weapons are concealed on the body surface. The holographic data is a complex signal containing amplitude and phase information.

[0053] The depth camera 11 is a camera capable of detecting the depth distance, or depth information, of the shooting space. In other words, the depth camera 11 can also be called a 3D camera. The depth camera 11 can acquire the distance from a point within a 3D shooting area containing the object to the depth camera 11, thereby obtaining 3D information of the shooting space. The depth camera 11 can be an existing depth camera. Typically, there are three types of depth cameras based on the following principles: structured-light, stereo vision, and time-of-flight (TOF).

[0054] The holographic data acquisition device 30 is used to acquire holographic data. The holographic data acquisition device 30 includes a receiving antenna 12 and a transmitting antenna 13. Specifically, the holographic data acquisition device 30 includes a multi-transmitter multi-receiver antenna array, which includes multiple pairs of transmitting antennas 13 and receiving antennas 12 for transmitting and receiving millimeter waves.

[0055] Here, the number of transmitting antennas 13 and receiving antennas 12 is not necessarily the same. As long as the receiving antenna 12 can receive the millimeter waves transmitted from the transmitting antenna 13, a pair of transmitting and receiving antennas 13 and 12 for transmitting and receiving millimeter waves can be formed. Multiple transmit and receive refers to the combination of multiple pairs of transmitting antennas 13 and receiving antennas 12. For example, when there are 2 transmitting antennas 13 and 3 receiving antennas 12, and all 3 receiving antennas 12 can receive the millimeter waves transmitted by the 2 transmitting antennas 13, there are 6 pairs of transmitting antennas 13 and receiving antennas 12, thereby achieving multiple transmit and receive.

[0056] The transmitting antenna 13 is used to transmit millimeter waves, and the receiving antenna 12 is used to receive reflected millimeter waves. The receiving antenna 12 and the transmitting antenna 13 are arranged in an array capable of multiple transmissions and receptions of millimeter waves. The receiving antenna 12 and the transmitting antenna 13 can use active millimeter waves to scan the holographic data acquisition area, thereby obtaining holographic data.

[0057] The 3D shooting area of ​​the depth camera 11 and the holographic data acquisition area of ​​the holographic data acquisition device 30 can be different, as long as both contain the object to be detected.

[0058] Specifically, for example, the millimeter-wave imaging device 10 can employ electronic scanning (see reference). Figure 2 ) and mechanical scanning type (refer to Figure 3 (Multi-transmitter, multi-receiver millimeter-wave imaging device)

[0059] Figure 2 This is a schematic diagram showing an electronically scanned millimeter-wave imaging device 10'. Figure 2 In this array, the receiving antenna 12' and the transmitting antenna 13' are arranged in a compact, box-shaped multiple-transmit, multiple-receive (MMR) array. This MMR array comprises several sub-box arrays, each containing two sets of orthogonally arranged transceiver arrays. In other words, two pairs of transmitting and receiving antennas are orthogonally arranged. This electronically scanned millimeter-wave imaging device 10' acquires holographic data by rapidly switching the receiving antenna 12' and the transmitting antenna 13'. Specifically, the switching can be performed as follows: any one of the transmitting antennas 13' in the transceiver array sequentially transmits radio frequencies f1, f2…fn in the millimeter-wave band, and all receiving antennas receive the signals. Then, this process is repeated for all transmitting antennas. Additionally, a depth camera 11' is positioned outside the MMR array to acquire depth data.

[0060] Figure 3 This is a schematic diagram illustrating a mechanically scanned millimeter-wave imaging device 10". The millimeter-wave imaging device 10" is a one-dimensional mechanically scanned imaging device with multiple transmitters and receivers. For example... Figure 3As shown, the receiving antenna 12” and the transmitting antenna 13” are arranged in a one-dimensional linear array. Specifically, in the mechanically scanned millimeter-wave imaging device 10”, the aforementioned one-dimensional linear array can move in the vertical direction. For example, an electric motor or the like can be used to move the one-dimensional linear array from top to bottom. During the top-down movement, each transmitting antenna 13” sequentially transmits radio frequencies f1, f2...fn in the millimeter-wave band, and all receiving antennas receive the signals; however, this process is repeated for all transmitting antennas. This process is repeated continuously during the movement by the electric motor or the like until the movement is complete.

[0061] The 11" depth camera is positioned outside the multi-transmitter, multi-receiver one-dimensional linear array to acquire depth data. The placement of the depth camera is not particularly limited, as long as it does not interfere with the transmitting and receiving antennas and can acquire depth information. Although Figure 3 The depth camera 11" is configured with Figure 2 The depth camera 11' can be located in different positions, but it can also be configured in the same position.

[0062] Hereinafter, regardless of the scanning method, the millimeter-wave imaging devices 10' and 10" will be collectively referred to as millimeter-wave imaging device 10, the depth camera 11' and 11" will be collectively referred to as depth camera 11, the receiving antenna 12' and 12" will be collectively referred to as receiving antenna 12, and the transmitting antenna 13' and 13" will be collectively referred to as transmitting antenna 13.

[0063] In the past, the most commonly used technology was one-dimensional single-shot single-receive or quasi-single-shot single-receive linear array mechanical scanning. However, one-dimensional single-shot single-receive scanning methods have problems such as low image quality and large noise impact.

[0064] As described above, in the embodiments of the present invention, both electronic scanning and mechanical scanning methods employ a multi-transmitter, multi-receiver array arrangement. Through this arrangement, beneficial effects such as high signal-to-noise ratio and fewer blind spots in the reconstructed image can be achieved.

[0065] The control device 20 controls the millimeter-wave imaging device 10. Specifically, the control device 20 is connected to the depth camera 11 and the holographic data acquisition device 30 included in the millimeter-wave imaging device 10, controls the depth camera 11 to acquire three-dimensional image information, and controls the receiving antenna 12 and transmitting antenna 13 of the holographic data acquisition device 30 to acquire holographic data through millimeter-wave scanning.

[0066] Specifically, the control device 20 includes the following functional modules: a three-dimensional image information acquisition module, which acquires three-dimensional image information of a three-dimensional imaging area containing the detection object captured by a depth camera; a mask extraction module, which extracts the mask of the detection object from the three-dimensional image information; an imaging area determination module, which determines the imaging area of ​​the detection object based on the mask of the detection object; a holographic data acquisition module, which acquires holographic data of a holographic data acquisition area containing the detection object captured by a holographic data acquisition device; and an image reconstruction module, which reconstructs the image of the imaging area based on the holographic data.

[0067] The control device 20 can be implemented in software, in hardware, or in a combination of both.

[0068] The control device 20 may have a processor and a memory.

[0069] The processor may include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor may contain multiple processors or multi-core processors to share some of the processing performed by the control device 20, thereby reducing the processing load of a single processor or single processing unit. Multiple processors or multi-core processors can execute the aforementioned processing in parallel, thereby improving the computing speed.

[0070] Memory includes, for example, ROM (Read Only Memory) which stores programs and various data, and RAM (Random Access Memory) which is used as the working area of ​​the CPU.

[0071] The control device 20 can achieve control by executing a program stored in memory through a processor.

[0072] The control device 20 may also include hardware such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit). Some or all of the processing performed by the control device 20 can be implemented by the FPGA, ASIC, etc.

[0073] The following is for reference Figure 4 The three-dimensional imaging method executed by the control device 20 is described in detail.

[0074] In step S11, the depth camera 11 is controlled to acquire three-dimensional image information. Since the depth camera 11 can obtain the distance (depth) information from a point in the imaging area to the depth camera 11, it can acquire three-dimensional image information of the object in the imaging area.

[0075] In this invention, depth data is obtained by employing a depth camera, thereby enabling the acquisition of three-dimensional image data. Furthermore, by determining the imaging region based on this three-dimensional image information, a three-dimensional spatial region can be used as the imaging area, thus improving the quality of the reconstructed image.

[0076] In step S12, an image segmentation algorithm is applied to the 3D image information acquired by depth camera 11 to extract the human body mask. This human body mask represents the external contour of the human body that is the object of security inspection, including the human body and clothing.

[0077] Here, the image segmentation algorithm is a machine learning-based image segmentation algorithm. Specifically, a neural network can be used.

[0078] In the machine learning-based image segmentation algorithm, a certain amount of security inspection scene data is collected, and human body masks are labeled on the 3D image information as training data. The 3D image information acquired by the depth camera 11 is used as input, and the neural network is used for image segmentation to obtain the learned human body mask.

[0079] By applying a neural network-based image segmentation algorithm, it is possible to effectively identify human masks that closely resemble the actual human body contour from 3D image information.

[0080] Specifically, the image segmentation algorithm can employ the DeepLabV3+ segmentation algorithm. The neural network built upon the DeepLabV3+ segmentation algorithm includes an encoder and a decoder.

[0081] The DeepLabV3+ segmentation algorithm can specifically include:

[0082] The coding layer front end uses dilated convolution to obtain shallow, low-level features, which are then transmitted to the decoder front end.

[0083] The back end of the encoding layer uses VGG-16 (Visual Geometry Group Network) to obtain deep high-level feature information and transmits it to the decoder;

[0084] The decoder connects the features into network layers, and then refines them through a 3×3 feature convolution.

[0085] At the decoder backend, after bilinear 4x sampling, a deep learning-segmented human body mask is obtained.

[0086] By employing the DeepLabV3+ segmentation algorithm, image segmentation with high accuracy can be achieved, and the segmentation effect of human body masks can be improved.

[0087] In step S13, the imaging region is determined based on the human body mask. The imaging region refers to the object area for which image reconstruction (imaging) is to be performed.

[0088] Specifically, a binary segmentation image can be used, setting points inside the human body mask as "1" and points outside as "0". Here, "1" represents points that need to be reconstructed, and "0" represents points that do not need to be reconstructed.

[0089] As described above, in this embodiment, the imaging area is determined based on a human body mask that represents the external contour of the human body.

[0090] However, in previous millimeter-wave imaging devices, the image reconstruction technique typically assumed that the imaging area for image reconstruction was a rectangular box containing the human body. However, pixels within the rectangular box but outside the human body mask do not need to be reconstructed, wasting computational resources and resulting in a slow image reconstruction process.

[0091] In contrast, in this invention, an artificial intelligence-based image segmentation algorithm is used to identify human body masks in a three-dimensional image, and the imaging area is determined based on the human body mask. This avoids image reconstruction of redundant object areas, saves computing resources, and improves the speed of image reconstruction.

[0092] In step S14, an array of multiple transmit and receive antennas is used to scan with active millimeter waves to obtain holographic data.

[0093] The specific method for obtaining holographic data using active millimeter waves can be found in the section on... Figure 2 Explanation of electronic scanning methods and their applications Figure 3 The mechanical scanning method described.

[0094] exist Figure 4 The diagram shows that steps S13 and S14 are performed in parallel, meaning that the determination of the imaging region and the acquisition of holographic data are performed concurrently. However, steps S13 and S14 can also be performed sequentially.

[0095] In step S15, based on the holographic data obtained in step S14, the backpropagation imaging algorithm is used to reconstruct the image for the imaging area determined in step S13.

[0096] The backpropagation imaging algorithm will be explained in detail below.

[0097] (Example 1)

[0098] When reconstructing an image from the imaging region determined in step S13, the backpropagation imaging algorithm may include the following steps:

[0099] First, for each pair of transmitting antennas 13 and receiving antennas 12, the sum of the distances from the pixels in the imaging area to that pair of transmitting antennas 13 and receiving antennas 12 is calculated.

[0100] Next, based on the sum of the distances and the holographic data acquired in step S14, the complex reflectance of the pixel is obtained by summing the values ​​of all transmitting antennas 13 and receiving antennas 12, as well as all transmission frequencies in the millimeter-wave band. Finally, the complex reflectance is calculated for all pixels in the imaging area, and a reconstructed image is formed based on the complex reflectance. Summing the values ​​of all transmitting antennas 13 and receiving antennas 12, as well as all transmission frequencies in the millimeter-wave band, can be achieved by iterating through all transmitting antennas 13 and receiving antennas 12, summing the values ​​of all transmitting antennas 13 and receiving antennas according to the horizontal and vertical axes, and summing the values ​​of each frequency point in the millimeter-wave band.

[0101] Specifically, backpropagation imaging algorithms can include:

[0102] Using formula (1), calculate the sum of distances r from each pixel (x, y, z) in the imaging region to each pair of transmitting and receiving antennas. T，R ,

[0103]

[0104] Based on the holographic data s(x) received by the receiving antenna T ,y T ,x R ,y R, (k), and the sum of distances r T，R Using formula (2), the complex reflectance of the pixel (x, y, z) in the imaging region is calculated.

[0105]

[0106] Among them, (x R y R (x, 0) represents the coordinates of receiving antenna 12. T y T (x, y, z) represents the coordinates of the transmitting antenna 13, (x, y, z) represents the coordinates of the pixel in the imaging region, k is the wave number, k = 2π / λ, where λ is the wavelength, k is a parameter representing the transmission frequency of the millimeter wave, and j is an imaginary number.

[0107] In formula (2), for holographic data s(x) T ,y T ,xR ,y R, Apply phase compensation exp(jkr) T,R It iterates through all receiving antennas and transmitting antennas, as well as all transmission frequencies in the millimeter-wave band, that is, it performs a 5-fold summation according to the horizontal and vertical coordinates of the receiving antenna, the horizontal and vertical coordinates of the transmitting antenna 13, and the wave number.

[0108] For all pixels in the imaging area, repeat the above steps, that is, traverse all pixels, calculate the complex reflectance, and form a reconstructed image based on the complex reflectance.

[0109] Previous technologies did not provide corresponding multi-scan and multi-receive image reconstruction algorithms. Even when a multi-scan and multi-receive scanning method was used, a single-scan and single-receive reconstruction algorithm was still employed. As a result, the advantages of multi-scan and multi-receive arrays, such as high signal-to-noise ratio and fewer blind spots, could not be realized.

[0110] As described above, in the embodiments of the present invention, a multi-scan, multi-receiver image reconstruction algorithm that matches the scanning mode of a multi-scan, multi-receiver array is provided, thereby demonstrating the advantages of the multi-scan, multi-receiver array in reconstructing images, such as high signal-to-noise ratio and fewer blind spots.

[0111] (Example 2)

[0112] The backpropagation imaging algorithm in Example 2 is the fast backpropagation imaging algorithm.

[0113] The fast backpropagation imaging algorithm may include the following steps:

[0114] For each pair of transmitting and receiving antennas, calculate the sum of the distances from each pixel in the imaging region to the transmitting and receiving antennas.

[0115] The contribution value is calculated based on the sum of the distances to the holographic data received by the receiving antenna.

[0116] For all transmitting and receiving antennas, the complex reflectivity of each pixel in the imaging region is obtained by summing the contribution values ​​across all transmitting and receiving antennas. Specifically, this summation can be performed separately for the x-coordinate and y-coordinate of each transmitting and receiving antenna.

[0117] Specifically, such as Figure 5 As shown, the fast backpropagation imaging algorithm may include:

[0118] In step S51, using formula (3), for each pair of transmitting antennas T and receiving antennas R, the sum of distances r from the pixel point (x, y, z) in the imaging region to the pair of transmitting antennas and receiving antennas is calculated. T，R ,

[0119]

[0120] In step S52, based on formula (4), the sum of distances r T，R and holographic data s(x) T ,y T ,x R ,y R, Calculate the contribution value P to pixel (x, y, z). T,R ,

[0121]

[0122] Specifically, based on the sum of distances r T，R The phase compensation value exp(jkr) is calculated from the wavenumber k. T,R ), for holographic data s(x) T ,y T ,x R ,y R, Apply phase compensation exp(jkr) T,R ), calculate the contribution value P. T,R .

[0123] In step S53, based on the contribution value P T,R Using formula (5), the complex reflectance of the pixels in the imaging region can be calculated.

[0124]

[0125] Where, r T，R Let s(x) be the sum of the distances from the transmitting antenna and the receiving antenna to the pixel (x,y,z) in the imaging region. T ,y T ,x R ,y R, ,k) is the holographic data received through a pair of transmitting and receiving antennas, where k is the wave number, k = 2π / λ, where λ is the wavelength, k is a parameter representing the transmission frequency of the millimeter wave, and j is an imaginary number.

[0126] The contribution of holographic data formed by a pair of transmitting and receiving antennas to different pixels is the sum of the distances r. T，R Relevant. For example Figure 6 As shown, when the sum of distances r T，R At the same time, these pixels are distributed on an ellipsoid, the two foci of which are the transmitting antenna (x T y T ,0) and receiving antenna (x) R y R ,0). If r is changed T，RIf this happens, the major axis of the ellipsoid will also change.

[0127] When reconstructing an image using the fast backpropagation imaging algorithm described above, the contribution value P of pixels located on the same ellipsoid is... T,R same.

[0128] Therefore, based on this principle, the pixels in the imaging region can be divided according to an ellipsoid. In other words, the sum of distances r in the imaging region... T，R Identical pixels, i.e., pixels located on the same ellipsoid, are grouped together. For each group of pixels, the contribution value P is calculated only once. T,R Based on this contribution value, the complex reflectance of this group of pixels is calculated, thereby achieving image reconstruction.

[0129] In other words, such as Figure 7 As shown, in Figure 5 Before step S52, step S54 may also be included. In step S54, the sum of distances r calculated by formula (3) is first... T，R All pixels in the imaging area are grouped, and in step S52, the calculation of formula (4) is performed only once for the same group of pixels. In other words, the contribution value of any pixel in the group is calculated and used as the contribution value of all pixels in the group.

[0130] If we explain the process of grouping pixels in more detail, the fast backpropagation imaging algorithm can be described as follows: Figure 8 As shown.

[0131] In step S151, the sum of the distances r from each pixel in the imaging region to the pair of transmit and receive antennas is calculated. T，R minimum value r min and maximum value r max .

[0132] Generally speaking, the minimum value r min for The maximum value is the sum of the distances from a pair of transmitting and receiving antennas to each vertex of the imaging region.

[0133] In step S152, as Figure 5 As shown, the minimum value r min and maximum value r max The distance r between d Divide the distance dr into N equal parts, thus obtaining the sum of the N distances (r1, r2, ..., r). N ). Where, dr=(r max -r min ) / N, r1=r min, r2=r1+dr, r3=r2+dr,…r N =r max N is an integer greater than 1, which can be arbitrarily set according to the calculation precision and the size of the imaging area.

[0134] In step S153, the sum of the distances from any pixel in the imaging region to the pair of transmitting and receiving antennas is calculated using the above formula (3). Here, Figure 8 The processing in step S153 is equivalent to Figure 7 The processing of step S51 in the process.

[0135] In step S154, from the equal division values ​​(r1, r2, ..., r N Determine the nearest equal division value (e.g., r) to the sum of the distances. i Here, Figure 8 The processing in steps S151-S152 and S154 is equivalent to Figure 7 The processing of step S54 in the process.

[0136] In step S155, it is determined whether the operation of step S153 has been completed for all pixels in the imaging area. If yes, proceed to step S156; otherwise, return to step S153.

[0137] In step S156, the N equal division values ​​(r1, r2, ..., r) are respectively... N ) as the sum of distances r T，R Based on this equal score and the holographic data s(x) received through the pair of transceiver antennas T ,y T ,x R ,y R, Using formula (4) above, the contribution value of the pair of transceiver antennas is calculated. Here, Figure 8 The processing in step S156 is equivalent to Figure 7 The processing of step S52 in the process.

[0138] In step S157, it is determined whether steps S151 to S156 have been completed for all transmitting and receiving antennas. If yes, proceed to step S158; otherwise, return to step S151 and perform calculations for the next pair of transmitting and receiving antennas.

[0139] In step S158, using formula (5) above, the contribution values ​​are summed for all pairs of transmitting and receiving antennas to obtain the complex reflectivity. Here, Figure 8 The processing in step S158 is equivalent to Figure 7 The processing of step S53 in the process.

[0140] Through this operation, pixels within the imaging area that are the same distance from each other or are adjacent are assigned the same equal score. In other words, as... Figure 6 As shown, through equal division values ​​(r1, r2, ..., r N This divides all pixels in the imaging region into N ellipsoids, or N groups. For each pixel located on an ellipsoid (e.g., r...), the total number of pixels is 1. i The contribution value can be calculated once for each pixel on the graph.

[0141] As described above, by dividing all pixels in the imaging region into N groups, and assigning equal division values ​​r1, r2, ..., r to each of these N groups of pixels, N The contribution value is calculated as the sum of distances, so only N contribution values ​​need to be calculated.

[0142] Therefore, compared with Example 1, it is not necessary to sum the distances of all pixels separately, which greatly speeds up the reconstruction.

[0143] exist Figure 8 Although the example shows the case where steps S151-S152 are executed first, followed by steps S153-S154—that is, first determining the points where the minimum and maximum values ​​of the distance sum might occur, calculating the minimum and maximum values, and then calculating the distance sum for all pixels and grouping them—it is not limited to this. Alternatively, step S153 can be executed first to calculate the distance sum for all pixels in the imaging area, and then the minimum and maximum values ​​can be determined from this (i.e., executing step S151). Then, a process equivalent to... Figure 7 S54 steps Figure 8 Steps S151 to S152 and S154 in the text.

[0144] If we calculate the sum of distances to all pixels and then select the minimum and maximum values, a large number of distance sums would need to be stored in memory for a long time until the minimum and maximum values ​​of all distance sums are obtained and the grouping is complete. However, when using methods such as... Figure 8 In the process described above, the calculation of the minimum and maximum values ​​of the sum of distances is independent of the calculation of the sum of distances of individual pixels. They can be calculated directly based on geometric relationships, and the minimum and maximum values ​​can be obtained with less computation. Moreover, each time a sum of distances is calculated, its nearest equal division value is determined, and subsequent calculations use this equal division value. There is no need to store the previously calculated sum of distances for a long time, which can save storage space and computer resources.

[0145] (Second Implementation)

[0146] The second embodiment differs from the first embodiment in that the control device 20 further includes a key point determination module and an imaging area adjustment module. The key point determination module determines the key points of the human body based on three-dimensional image information, and the imaging area adjustment module adjusts the imaging area based on the core key points. In other words, the three-dimensional imaging method executed in the control device 20 differs from that in the first embodiment.

[0147] refer to Figure 7 This describes the three-dimensional imaging method involved in the second embodiment. Figure 7 Steps S11 to S15 in the middle Figure 4 Steps S11 to S15 are the same. The difference is that after step S13, there are also steps S16 and S17.

[0148] In step S16, based on the 3D image information acquired by depth camera 11, the core key points of the human body in the image are determined using a deep learning model. Here, core key points are key points on the human body contour, such as the top of the head, neck, and major joints of the limbs. There are no particular restrictions on the deep learning model used to determine the core key points; existing models can be used. Examples include the Deep Pose model, the HeatMap model, and the CPM (Convolutional Pose Machine) model.

[0149] In step S17, based on the core key points determined in step S16, human body parts related to each core key point in the imaging area are selected, thereby adjusting the imaging area determined in step S13.

[0150] As mentioned above, the key points can represent parts of the human body such as the top of the head, neck, and major joints of the limbs. The imaging area determined in step S13 is the area corresponding to the human body mask, that is, the imaging area at this time is the area that includes the entire human body outline. In step S17, a portion of the imaging area that includes the entire human body outline is selected and retained, while the other portion is removed, so that the imaging area only includes a portion of the human body outline.

[0151] For example, depending on the actual application scenario of the equipment, if it is believed that items are difficult to conceal on the head and neck, and therefore do not need to be included in security checks, these areas can be excluded from the imaging area based on key points. For instance, the human head can be removed from the imaging area determined in step S13. This adjusts the imaging area to determine the final imaging area. However, in special regions such as Arab countries, headscarves may be worn, and items can be concealed there; in such cases, the head also needs to be included in the imaging area. Furthermore, the imaging of private parts of the human body is controversial; therefore, these parts can also be excluded from the imaging area.

[0152] Human body parts related to each core key point can be pre-stored in the 3D imaging system, thereby automatically adjusting the imaging area, or they can be manually selected later as needed.

[0153] In the second embodiment, by adjusting the imaging area according to the core key points of the human body, the imaging area can be made more reasonable, the image reconstruction range can be further reduced, and the image reconstruction calculation speed can be further improved.

[0154] (Modified Example)

[0155] Based on the first embodiment described above, the control device 20 may further include a key point determination module and a posture judgment module. Regarding the key point determination module, please refer to the second embodiment described above. Alternatively, based on the second embodiment described above, the control device 20 may also further include a posture judgment module.

[0156] The posture judgment module determines the position of the core key points identified by the key point determination module to judge whether the person's posture is correct. If the posture is incorrect, an alarm will be triggered.

[0157] This posture judgment and processing can be done in Figure 4 After step S13, or in Figure 9 It is performed after step S17 and before step S15.

[0158] If a person's posture is incorrect, it may be difficult to accurately locate items. By adding a posture judgment module, the posture of people undergoing security checks can be corrected.

[0159] In addition, since this posture determination process is performed before image reconstruction in step S15, it is possible to avoid generating invalid security inspection images due to incorrect posture before image reconstruction, thus avoiding a waste of computing resources.

[0160] Furthermore, in this invention, for example, when the control device 20 includes multiple processors, or includes a multi-core processor, during image reconstruction, each core of the multi-core GPU can perform image reconstruction calculations for one pixel independently. Specifically, in Figure 8 In the example shown, the nearest equal division value can be determined by calculating the sum of the distances to each pixel using each core of the GPU, i.e., by executing steps S153 to S154 as described above. In this case, reconstruction of thousands of pixels can be performed simultaneously, thereby improving the computational speed.

[0161] Furthermore, when the control device 20 includes an FPGA (Field Programmable Gate Array), some of the processing performed by the control device 20 can be implemented by the FPGA. For example, the FPGA can control the holographic data acquisition device in a millimeter-wave imaging device, thereby controlling the transmission and data acquisition of millimeter waves. Specifically, after acquiring holographic data from a pair of transmitting and receiving antennas, preprocessing operations can be performed on it. The preprocessed holographic data is then sent to other processors, such as a CPU, for further processing, while the FPGA controls the next pair of transmitting antennas and the receiving antennas to acquire holographic data. For example, the preprocessing could be: for the pair of transmitting and receiving antennas, the FPGA first calculates the contribution value of the pair of transmitting and receiving antennas to the pixel. Then, the calculated contribution value is sent to other processors, such as a CPU, for subsequent image reconstruction operations. At this time, the FPGA then controls the acquisition of holographic data from the next pair of transmitting and receiving antennas. As described above, through the division of labor between the FPGA and processors such as the CPU, not only can the workload of the CPU be reduced, but the image reconstruction time can also be reduced.

[0162] While the embodiments and specific examples of the present invention have been described above in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and all such modifications and variations fall within the scope defined by the claims.

Claims

1. A three-dimensional imaging method, comprising: The three-dimensional information acquisition step involves capturing a three-dimensional imaging area containing the detected object using a depth camera to generate three-dimensional image information. The mask extraction step involves applying an image segmentation algorithm to the three-dimensional image information to extract the mask of the detection object. The image segmentation algorithm is a machine learning-based image segmentation algorithm trained by labeling human body masks in a certain number of three-dimensional image information of security inspection scenarios as training data. The imaging region determination step involves determining the imaging region about the detection object based on the mask of the detection object, wherein the imaging region is the region to be reconstructed. The holographic data acquisition step involves using a holographic data acquisition device to acquire data from a holographic data acquisition area containing the detection object, thereby generating holographic data; and The image reconstruction step involves reconstructing the image of the imaging region based on the holographic data.

2. The three-dimensional imaging method as described in claim 1, wherein, The image segmentation algorithm is the DeepLabV3+ segmentation algorithm based on neural networks.

3. The three-dimensional imaging method as described in claim 1, wherein, The image reconstruction step uses the backpropagation imaging algorithm to reconstruct the image.

4. The three-dimensional imaging method as described in claim 3, wherein, The holographic data acquisition device includes a multi-transmitter, multi-receiver antenna array, which comprises multiple pairs of transmitting antennas and receiving antennas for transmitting and receiving millimeter waves. The backpropagation imaging algorithm includes: For each pixel in the imaging area, calculate the sum of the distances from that pixel to each pair of transmitting and receiving antennas; Based on the sum of the holographic data received by the receiving antenna and the distance, the complex reflectance of each pixel in the imaging area is obtained by summing the data according to all transmitting and receiving antennas and the transmission frequency of the millimeter wave. The complex reflectance is calculated for all pixels in the imaging area, and a reconstructed image is formed based on the complex reflectance.

5. The three-dimensional imaging method as described in claim 4, wherein, The backpropagation imaging algorithm includes: Using formula (1), calculate the sum of distances r from each pixel (x, y, z) in the imaging region to each pair of transmitting and receiving antennas. T，R , Based on the holographic data s(x) received by the receiving antenna T ,y T ,x R ,y R, k), and the sum of the distances r T，R Using formula (2), the complex reflectance of the pixel (x, y, z) in the imaging region is calculated. Among them, (x R y R (x, 0) represents the coordinates of the receiving antenna. T y T (x, y, z) represents the coordinates of the transmitting antenna, (x, y, z) represents the coordinates of a pixel in the imaging region, k is the wave number, k = 2π / λ, where λ is the wavelength, and j is an imaginary number. For all pixels in the imaging area, the complex reflectance is calculated, and a reconstructed image is formed based on the complex reflectance.

6. The three-dimensional imaging method as described in claim 3, wherein, The holographic data acquisition device includes a multi-transmitter, multi-receiver antenna array, which comprises multiple pairs of transmitting antennas and receiving antennas for transmitting and receiving millimeter waves. The backpropagation imaging algorithm is a fast backpropagation imaging algorithm, including: For each pair of transmitting and receiving antennas, calculate the contribution value of each pair of transmitting and receiving antennas to each pixel in the imaging area; For all transmitting and receiving antennas, based on the contribution values, the complex reflectance of each pixel in the imaging region is obtained by summing the values ​​for all transmitting and receiving antennas. A reconstructed image is formed based on the complex reflectance.

7. The three-dimensional imaging method as described in claim 6, wherein, For each pair of transmitting and receiving antennas, the contribution value of the pair of transmitting and receiving antennas to each pixel in the imaging region is calculated, including: For the pair of transmitting antennas and receiving antennas, calculate the sum of the distances from each pixel in the imaging area to the pair of transmitting antennas and the receiving antennas; The contribution value is calculated based on the sum of the distances and the holographic data received by the receiving antenna.

8. The three-dimensional imaging method as described in claim 7, wherein, The fast backpropagation imaging algorithm includes: Using formula (3), for each pair of transmitting antennas and receiving antennas, calculate the sum of distances r from the pixel point (x, y, z) in the imaging region to the pair of transmitting antennas and receiving antennas, respectively. T，R , Using formula (4), based on the sum of the distances r T，R and the holographic data s(x) T ,y T ,x R ,y R ,k), calculate the contribution value P to the pixel (x, y, z). T,R , Based on the contribution value P T,R Using formula (5), the complex reflectance of each pixel in the imaging region is calculated. Where, r T，R S(x) is the sum of the distances from the pair of transmitting antennas and the receiving antennas to the pixel (x, y, z) in the imaging region. T ,y T ,x R ,y R ,k) is holographic data obtained through a pair of transmitting and receiving antennas, where k is the wave number, k = 2π / λ, where λ is the wavelength, and j is the imaginary number; A reconstructed image is formed based on the complex reflectance.

9. The three-dimensional imaging method as described in claim 6, wherein, For each pair of transmitting and receiving antennas, the contribution value of the pair of transmitting and receiving antennas to each pixel in the imaging region is calculated, including: Among the pixels in the imaging area, pixels whose sum of distances to the pair of transmitting antennas and the receiving antennas are equal are grouped together; Calculate the contribution value of the pair of transmitting antennas and receiving antennas to any pixel in the group, and use it as the contribution value of each pixel in the group.

10. The three-dimensional imaging method as described in claim 6, wherein, For each pair of transmitting and receiving antennas, the contribution value of the pair of transmitting and receiving antennas to each pixel in the imaging region is calculated, including: Calculate the minimum and maximum sum of distances from all pixels in the imaging region to the pair of transmitting antennas and the receiving antenna; Divide the distance between the minimum and maximum values ​​into N equal intervals to obtain the sum of the N distances (r1, r2, ..., r). N ); Calculate the sum of the distances from any pixel in the imaging region to the pair of transmitting antennas and the receiving antenna, determine the nearest equal division value to the sum of the distances, and use the nearest equal division value as the sum of the distances of the pixel; For all pixels in the imaging area, determine the nearest equal division value; The contribution value is calculated by summing the N equal divisions as the distances.

11. The three-dimensional imaging method as described in claim 1, further comprising: The key point identification process, based on a deep learning model, identifies the core key points of the object being detected. The imaging area adjustment step involves selecting, based on the core key point, parts of the detection object related to the core key point within the imaging area, thereby adjusting the imaging area.

12. The three-dimensional imaging method as described in claim 1, further comprising: The key point identification process, based on a deep learning model, identifies the core key points of the object being detected. The object posture judgment step involves determining whether the object's posture is correct based on the location of the core key points. If it is incorrect, an alarm is triggered.

13. A three-dimensional imaging system, characterized in that, It includes a depth camera, a holographic data acquisition device, and a control device, wherein the control device includes: The 3D information acquisition module controls the depth camera to capture a 3D imaging area containing the detected object, thereby acquiring 3D image information. The mask extraction module applies an image segmentation algorithm to the three-dimensional image information to extract the mask of the detection object. The image segmentation algorithm is a machine learning-based image segmentation algorithm trained by labeling human body masks in a certain number of three-dimensional image information of security inspection scenarios as training data. An imaging region determination module determines an imaging region about the detected object based on the mask of the detected object, wherein the imaging region is the region to be reconstructed. The holographic data acquisition module controls the holographic data acquisition device to acquire data from the holographic data acquisition area containing the detection object, thereby obtaining holographic data; and The image reconstruction module reconstructs the image of the imaging area based on the holographic data.

14. The three-dimensional imaging system as described in claim 13, wherein, The holographic data acquisition device includes a multi-transmitter, multi-receiver antenna array, which includes multiple pairs of transmitting antennas and receiving antennas for transmitting and receiving millimeter waves.

15. A three-dimensional imaging device, comprising: The 3D image information acquisition module acquires 3D image information of the 3D shooting area containing the detection object captured by the depth camera; The mask extraction module applies an image segmentation algorithm to the three-dimensional image information to extract the mask of the detection object. The image segmentation algorithm is a machine learning-based image segmentation algorithm trained by labeling human body masks in a certain number of three-dimensional image information of security inspection scenarios as training data. An imaging region determination module determines an imaging region about the detected object based on the mask of the detected object, wherein the imaging region is the region to be reconstructed. The holographic data acquisition module acquires holographic data obtained by acquiring data from a holographic data acquisition area containing the detection object using a holographic data acquisition device; and The image reconstruction module reconstructs the image of the imaging area based on the holographic data.

16. A computer-readable medium storing a computer program, characterized in that, The computer program is executed by the processor to achieve the following: The three-dimensional image information acquisition step involves acquiring three-dimensional image information of the three-dimensional shooting area containing the detected object, captured by a depth camera. The mask extraction step involves applying an image segmentation algorithm to the three-dimensional image information to extract the mask of the detection object. The image segmentation algorithm is a machine learning-based image segmentation algorithm trained by labeling human body masks in a certain number of three-dimensional image information of security inspection scenarios as training data. The imaging region determination step involves determining the imaging region about the detection object based on the mask of the detection object, wherein the imaging region is the region to be reconstructed. The holographic data acquisition step involves acquiring holographic data by using a holographic data acquisition device to acquire data from a holographic data acquisition area containing the detection object; and The image reconstruction step involves reconstructing the image of the imaging region based on the holographic data.

17. A three-dimensional imaging device, comprising a memory, a processor, and program instructions stored in the memory and executable by the processor, characterized in that, when the processor executes the program instructions, it implements: The three-dimensional image information acquisition step involves acquiring three-dimensional image information of the three-dimensional shooting area containing the detected object, captured by a depth camera. The mask extraction step involves applying an image segmentation algorithm to the three-dimensional image information to extract the mask of the detection object. The image segmentation algorithm is a machine learning-based image segmentation algorithm trained by labeling human body masks in a certain number of three-dimensional image information of security inspection scenarios as training data. The imaging region determination step involves determining the imaging region about the detection object based on the mask of the detection object, wherein the imaging region is the region to be reconstructed. The holographic data acquisition step involves acquiring holographic data by using a holographic data acquisition device to acquire data from a holographic data acquisition area containing the detection object; and The image reconstruction step involves reconstructing the image of the imaging region based on the holographic data.

Images