Deduplication methods and equipment for detection targets

Abstract

This application provides a method and apparatus for deduplicating detected targets. The method includes: acquiring suspicious items from at least two human body images of the same scanned object from different directions; defining suspicious items located at the edge of the human body contour in the first human body image from the at least two different directions as targets to be deduplicated; deduplicating suspicious items in a second human body image from the at least two different directions, based on the targets to be deduplicated, to obtain the number of duplicate suspicious items in the second human body image; and determining the total number of suspicious items carried by the scanned object based on the number of duplicates and the number of suspicious items in the first human body image. This method can effectively identify and remove the same suspicious item repeatedly detected in images from different directions, avoiding duplicate counting and improving security inspection efficiency.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and apparatus for deduplicating detected targets. Background Technology

[0002] Millimeter-wave human body imaging equipment is a security inspection instrument that uses millimeter-wave technology. This equipment can efficiently detect suspicious items hidden under clothing.

[0003] When millimeter-wave body imaging equipment is used in security checks, if the number of suspicious items detected by the equipment exceeds a certain threshold, staff need to be prompted to conduct a full-body manual search of the passenger being inspected. During the identification and detection of suspicious items, some items may be identified multiple times in different body images, leading to an increase in the number of suspicious items detected by the equipment. This increases the probability of staff conducting a full-body manual search of the passenger, thereby reducing security check efficiency. Summary of the Invention

[0004] The purpose of this application is to provide a method for deduplicating detection targets, in order to solve the technical problem that the increased probability of full inspection leads to a decrease in security inspection efficiency.

[0005] Firstly, this application provides a method for deduplicating detected targets, the method comprising:

[0006] Obtain suspicious items from at least two different orientations of the same scanned object;

[0007] The suspicious items located at the boundary of the human body outline in the first human body image of the at least two human body images from different directions are defined as the targets to be deduplicated;

[0008] Based on the target to be deduplicated, the suspicious items in the at least two human images from different directions and in the second human image from a different direction than the first human image are deduplicated to obtain the number of suspicious items in the second human image.

[0009] Based on the recount count and the number of suspicious items in the first human image, the total number of suspicious items carried by the scanned object is determined.

[0010] Optionally, suspicious items located at the boundary of the human body outline in the first human body image among the at least two human body images from different directions are defined as targets to be deduplicated, including:

[0011] When the total number of suspicious items in the human body images from at least two different directions exceeds the re-detection threshold, the suspicious items located at the human body contour boundary in the first human body image from the at least two different directions are defined as targets to be deduplicated.

[0012] Optionally, suspicious items located at the boundary of the human body outline in the first human body image among the at least two human body images from different directions are defined as targets to be deduplicated, including:

[0013] Using human images from at least one orientation, determine the mask image of human body parts;

[0014] Suspicious objects in the first human body image have multiple boundary points. Suspicious objects in the first human body image whose boundary points are located in the background area of ​​the human body part mask are selected as targets for deduplication; and / or

[0015] Based on the human body part mask, the human body contour boundary is determined, and suspicious items in the first human body image whose distance to the human body contour boundary is less than the deduplication distance are identified as deduplication targets.

[0016] Optionally, suspicious items in the first human image whose distance to the human contour boundary is less than the deduplication distance are selected as deduplication targets, including:

[0017] The minimum distance from each boundary point of the suspicious item in the first human body image to each boundary point of the human body contour boundary is determined as the distance from each boundary point of the suspicious item to the human body contour boundary.

[0018] The minimum distance from all boundary points of the suspicious item in the first human body image to the boundary of the human body contour is determined as the distance from the suspicious item to the boundary of the human body contour.

[0019] Based on the distance of each suspicious item in the first human body image to the boundary of the human body contour, suspicious items in the first human body image whose distance to the boundary of the human body contour is less than the deduplication distance are taken as deduplication targets.

[0020] Optionally, based on the target to be deduplicated, suspicious items in the at least two human images from different directions, and in a second human image from a different direction than the first human image, are deduplicated to obtain the recount quantity of suspicious items in the second human image, including:

[0021] One of the first human body image and the second human body image is a frontal image of the human body, and the other is a back image of the human body;

[0022] The deduplicated target in the first human body image is horizontally mirrored and mapped to the second human body image to obtain the mirrored deduplicated target; wherein, the ordinate of the boundary point of the deduplicated target before mirroring is the same as the ordinate of the boundary point of the deduplicated target after mirroring.

[0023] Based on the mirrored target to be deduplicated, remove suspicious items from the second human image that overlap with the mirrored target to be deduplicated, and / or suspicious items whose center point is less than the expected distance from the center point of the mirrored target to be deduplicated;

[0024] The remaining number of suspicious items in the second human body image is taken as the recounted number of suspicious items in the second human body image.

[0025] Optionally, based on the target to be deduplicated, suspicious items in the at least two human images from different directions, and in a second human image from a different direction than the first human image, are deduplicated to obtain the recount quantity of suspicious items in the second human image, including:

[0026] The first human body image is a frontal image and / or a back image of a human body, and the second human body image is a side image of a human body;

[0027] The coordinates of the center position of the suspicious item in the first human body image are mapped to the x and y coordinates in the world coordinate system, and the coordinates of the center position of the human body part corresponding to the center position of the suspicious item are mapped to the z coordinate in the world coordinate system, so as to obtain the first mapped coordinates of the suspicious item in the first human body image.

[0028] The first mapped coordinates are transformed into the second human body image to obtain the second mapped coordinates;

[0029] If the second mapped coordinates are located inside the suspicious item in the second human body image, then the suspicious item is deduplicated.

[0030] Optionally, mapping the coordinates of the center position of the human body part corresponding to the center position of the suspicious item to the z-coordinate in the world coordinate system includes:

[0031] The z-coordinate of the center point of the human body part on the same horizontal plane as the center of the suspicious item is taken as the z-coordinate in the world coordinate system.

[0032] Optionally, using human images from at least one orientation, a mask map of human body parts is determined, including:

[0033] The background of the human image in at least one direction is segmented using threshold segmentation, region segmentation, edge segmentation, or clustering to obtain the background region and the human target region.

[0034] Joints and / or skeletons are extracted from the target human body region, and the extracted joints and / or skeletons are clustered.

[0035] The target human body region is divided according to the clustering results and the preset human body part labels to obtain the region corresponding to each human body part that matches the human body part label.

[0036] Pixel values ​​are set for the regions corresponding to each human body part and the background region to obtain the human body part mask.

[0037] Secondly, this application also provides a device for detecting target deduplication, the device comprising:

[0038] processor;

[0039] The memory stores computer-readable instructions, which, when executed by the processor, implement a method for deduplicating the detection target in the first aspect.

[0040] Thirdly, this application also provides a millimeter-wave device, including a scanner, a processor, and a memory;

[0041] The scanner is used to scan the object being scanned;

[0042] The processor is configured to define suspicious items located at the human body contour boundary in the first human body image from at least two different orientations as targets to be deduplicated; to deduplicate suspicious items in the second human body image from at least two different orientations, which are in a different orientation than the first human body image, based on the targets to be deduplicated, to obtain the recounted quantity of suspicious items in the second human body image; and to determine the total number of suspicious items carried by the scanned object based on the recounted quantity and the number of suspicious items in the first human body image.

[0043] A memory for storing the total number of suspicious items carried by the scanned object.

[0044] The deduplication method for detecting targets provided in this application removes duplicates of suspicious items located at the human body contour boundary in human body images from different directions. This avoids the repeated detection of suspicious items located at the human body contour boundary in human body images from other perspectives. Since deduplication is performed only on suspicious items close to the human body contour boundary instead of all suspicious items, the computational load of the deduplication operation can be effectively reduced, thus enabling the deduplication process to be completed quickly. Moreover, based on the number of suspicious items in the first human body image and the number of duplicate suspicious items in the second human body image, the total number of suspicious items carried by the scanned object is determined, making the total number of detected suspicious items closer to the total number of suspicious items actually carried by the scanned object. This avoids the problem of increased probability of manual full inspection due to repeated identification and counting of suspicious items, thereby improving security inspection efficiency. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart illustrating a method for deduplicating detection targets provided in an embodiment of this application;

[0047] Figure 2 A schematic diagram of a human body part mask provided in an embodiment of this application;

[0048] Figure 3 A flowchart illustrating a method for determining duplicate targets to be removed, provided in an embodiment of this application;

[0049] Figure 4 A flowchart illustrating a method for deduplicating suspicious items, as provided in this application embodiment;

[0050] Figure 5 This is an internal structural diagram of a device for detecting target deduplication provided in an embodiment of this application. Detailed Implementation

[0051] The present application will be described in detail below with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present application. Any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the protection scope of the present application.

[0052] When millimeter-wave equipment is used in security inspections, if the number of suspicious items detected by the equipment exceeds a certain threshold, it will prompt the operator to conduct a full-body manual inspection of the person being inspected. If suspicious items are identified once from a frontal image and once from a back image, and then the two counts are added together, the number of detected suspicious items will increase, thus increasing the probability of a full manual inspection and consequently reducing the efficiency of security checks.

[0053] like Figure 1 As shown in the figure, this application provides a method for deduplicating detected targets, which specifically includes the following steps:

[0054] Step S101: Obtain suspicious items from human images of the same scanned object from at least two different orientations.

[0055] Images of the same human body from different directions can include, but are not limited to, frontal images of the human body from the frontal view, back images of the human body from the backal view, and side images of the human body from the sideal view. Suspicious objects in the human body images can be knives, liquids, firearms, ammunition, etc. Suspicious objects are detected in human body images from each direction based on intelligent recognition algorithms. For example, the intelligent recognition algorithm can be, but is not limited to, the YOLO algorithm (You Only Look Once) or the SSD algorithm (Single Shot MultiBox Detector).

[0056] Among them, the detected suspicious items have multiple boundary points (e.g., a rectangle) on their boundaries. For example, a suspicious item boundary has 4 boundary points, which can be represented as Cn={Pn1, Pn2, Pn3, Pn4}, Pnm=P(Xnm,Ynm), where P(Xnm,Ynm) are image pixel coordinates, n is the nth suspicious item, and m is the mth boundary point of the nth suspicious item, m∈{1,2,3,4}; the center point On of the suspicious item is: On=( );

[0057] Where (Xn1,Yn1), (Xn2,Yn2), (Xn3,Yn3), and (Xn4,Yn4) are the image pixel coordinates of the four boundary points of the nth suspicious item.

[0058] For example, suspicious items are detected in a frontal image of a human body from a frontal perspective, resulting in a suspicious item set A, where A = {Ca1, Ca2, Ca3, ..., Caj}. This means j suspicious items are detected in the frontal image of a human body, and Caj represents the image pixel coordinates of the j-th suspicious item. Suspicious items are detected in a back-view image of a human body from a rear-view perspective, resulting in a suspicious item set B, where B = {Cb1, Cb2, Cb3, ..., Cbk}. This means k suspicious items are detected in the back-view image of a human body from a rear-view perspective, and Cbk represents the image pixel coordinates of the k-th suspicious item.

[0059] Step S102: Define suspicious items located at the boundary of the human body outline in the first human body image from at least two human body images from different directions as targets to be deduplicated.

[0060] The first human body image can be a frontal human body image from the frontal direction. Considering that when a suspicious item is located at the edge of the body of the object being inspected (i.e. the object being scanned), the suspicious item will be identified once in human body images from different directions, causing the suspicious item located at the edge of the body of the scanned object to be repeatedly identified and counted. Therefore, the suspicious item located at the boundary of the human body outline among the multiple suspicious items in the first human body image is taken as the target to be deduplicated.

[0061] For example, the target to be deduplicated can be a suspicious item in set A that is less than a set distance from the boundary of the human body outline. That is, the suspicious item in set A whose center point is less than a set distance from the boundary of the human body outline is the target to be deduplicated.

[0062] Step S103: Based on the target to be deduplicated, deduplicat suspicious items in at least two human images from different directions and in a second human image from a different direction than the first human image, and obtain the number of duplicate suspicious items in the second human image.

[0063] For example, deduplication of suspicious items in a second human body image can be achieved, but is not limited to, using the following methods: The target to be deduplicated in the first human body image (e.g., a frontal image of a human body) (i.e., a suspicious item located at the boundary of the human body's outline) can be mapped to a second human body image (e.g., a back or side view of the human body) from a different perspective using spatial location mapping rules (e.g., 3D coordinate system transformation or horizontal mirror mapping). All suspicious items in the second human body image are iterated over. If a suspicious item in the second human body image overlaps with the target to be deduplicated, it is determined that the suspicious item is repeatedly identified and counted, and deduplication is required. The number of deduplicated suspicious items in the second human body image is counted as the number of re-counted suspicious items in the second human body image, thereby avoiding the repeated counting of the same suspicious item carried by the scanned object in images from different directions, thus improving the accuracy and efficiency of security checks.

[0064] Step S104: Based on the recount count and the number of suspicious items in the first human body image, determine the total number of suspicious items carried by the scanned object.

[0065] If the first human body image is a frontal view of the human body, and the second human body image is a back view of the human body, the sum of the number of suspicious items in the first human body image and the number of repeated suspicious items in the second human body image is taken as the total number of suspicious items carried by the scanned object. For example, if there are j suspicious items in the first human body image and k suspicious items in the second human body image, with r repeated counts, then the number of repeated suspicious items in the second human body image is kr. Therefore, the total number t of suspicious items carried by the scanned object is: t = j + kr, where j, k, and r are all integers.

[0066] In this embodiment, suspicious items in human images of the same scanned object from different directions are identified. Suspicious items located at the boundary of the human body contour in the first human body image are identified as targets to be deduplicated. All suspicious items in the second human body image, which is in a different direction from the first human body image, are traversed. Suspicious items that are counted repeatedly in the second human body image are identified as the suspicious items that need to be deduplicated. Since these suspicious items located at the boundary of the human body contour will be detected repeatedly in the human body image from another perspective, by performing deduplication processing only on suspicious items close to the boundary of the human body contour (instead of all suspicious items), the amount of computation for deduplication can be reduced, thereby enabling the deduplication process to be completed quickly.

[0067] In an optional embodiment of this application, after step S104, a prompt can be issued when the total quantity t after recounting exceeds a set alarm threshold. The prompt can take the form of a light, a text message on a display screen, or an audio prompt, and is not limited to these methods. The prompt information may include, but is not limited to, messages such as: "Please conduct a manual security check," "Excess quantity of special items," a light, or a ringtone.

[0068] The number of suspicious items in the second human body image after deduplication is counted as the recount count of suspicious items in the second human body image. This avoids the same suspicious item located at the edge of the human body being counted repeatedly in human body images from different directions. Based on the number of suspicious items in the first human body image and the recount count of suspicious items in the second human body image, the total number of suspicious items carried by the scanned object is determined. This makes the total number of detected suspicious items closer to the total number of suspicious items actually carried by the scanned object, avoiding the problem of increased probability of manual full inspection due to repeated identification and counting of suspicious items, thereby improving security inspection efficiency.

[0069] In an optional embodiment of this application, before step S102, it can be determined whether the total number of suspicious items in at least two different human images exceeds the re-detection threshold. If the total number exceeds the reconstruction threshold, step S102 is executed, that is, suspicious items located at the edge of the human body contour in the first human image of at least two different human images are defined as targets to be deduplicated. In this case, the number of deduplications can be reduced.

[0070] Specifically, deduplication is triggered only when the total number of suspicious items detected in at least two different orientations (e.g., front and back) of the same scanned object exceeds a preset re-inspection threshold (i.e., the sum of the number of suspicious items in each image before deduplication). If the total number of detected suspicious items does not reach the re-inspection threshold, no deduplication is required. Considering that a full manual inspection will not be triggered or the probability of a full manual inspection will not increase when the total number of detected suspicious items does not reach the re-inspection threshold, deduplication is not performed on suspicious items. This improves the efficiency of security checks on suspicious items without excessively increasing the computational load.

[0071] In one embodiment, step S102 defines suspicious items located at the boundary of the human body outline in the first human body image among at least two human body images from different directions as targets to be deduplicated. This can be achieved, but is not limited to, through the following steps:

[0072] Step S201: Using human body images from at least one direction, determine a mask image of human body parts.

[0073] The size of the human body part mask is the same as the size of the human body image. The human body part mask has pixel values ​​of different sizes. The pixel value is used to represent the region to which the current pixel belongs. The region to which the current pixel belongs is one of the regions corresponding to each human body part and the background region.

[0074] For example, as shown in 2, a frontal image of the human body is used to determine a mask image of the human body parts. Figure 2 In the image 'a', there is a frontal image of a human body, in which two suspicious items have been detected. Figure 2 Figure b shows a human body part mask. Different pixel values ​​(0 to 14) in the mask represent the region to which the current pixel belongs, such as whether it belongs to the area corresponding to the human body part or the background area. The human body parts corresponding to different pixel values ​​are as follows: 1-Head, 2-Tortoise, 3-Left upper arm, 4-Right upper arm, 5-Left lower arm, 6-Right lower arm, 7-Left hand, 8-Right hand, 9-Left thigh, 10-Right thigh, 11-Left calf, 12-Right calf, 13-Left leg, 14-Right foot. The background area is represented by 0.

[0075] Step S202: Suspicious items in the first human body image have multiple boundary points. Suspicious items in the first human body image whose boundary points are located in the background area of ​​the human body part mask are taken as deduplication targets.

[0076] Assuming the frontal image is used as the first human body image, for suspicious items carried by the scanned object in the frontal image, if at least one boundary point of the suspicious item is not located in the area corresponding to the human body part in the human body part mask, but in the background area of ​​the human body part mask, then it is determined that the suspicious item is located at the boundary of the human body contour and needs to be deduplicated. In this way, by determining that at least one boundary point of the suspicious item is located in the background area, the target to be deduplicated can be determined simply and accurately and quickly located from multiple suspicious items in the first human body image.

[0077] In this embodiment of the application, in addition to finding the deduplicated target from multiple suspicious items in the first human body image through step S202, step S203 can also be used, or a combination of steps S202 and S203 can be used to find the deduplicated target from multiple suspicious items in the first human body image.

[0078] Step S203: Based on the human body part mask map, determine the human body contour boundary, and take suspicious items in the first human body image whose distance to the human body contour boundary is less than the deduplication distance as the deduplication target.

[0079] For example, the human body contour boundary can be determined based on the various human body part regions and the background region in the human body part mask image. For instance, the regions corresponding to each human body part in the human body part mask image can be combined, and the human body contour boundary can be determined based on the combined region and the background region.

[0080] For example, the boundaries of each human body part region (foreground) in the human body part mask image can be determined by the foreground and background segmentation algorithm as the human body contour boundary.

[0081] For example, assuming the suspicious item has a cuboid shape, the deduplication distance could be one-third of the shortest boundary of the suspicious item, or one-half of the longest boundary. If the suspicious item has a circular shape, the deduplication distance could be one-fifth of its diameter. The deduplication distance can be an empirical value, or it can be configured and adjusted by the user through an interactive interface.

[0082] If the distance between a suspicious item carried by the scanned object and the boundary of the human body contour is less than the deduplication distance in the human body part mask image, it can be inferred that the suspicious item is located at the boundary of the human body contour, which will result in duplicate identification and counting. Therefore, it needs to be regarded as a target to be deduplicated.

[0083] In one embodiment, determining a human body part mask using a human body image from at least one orientation may include:

[0084] Threshold segmentation, region segmentation, edge segmentation, or clustering are used to segment the background of a human image in at least one direction, resulting in a background region and a human target region.

[0085] Extract joints and / or skeletons from the target area of ​​the human body, and cluster the extracted joints and / or skeletons;

[0086] The target human body region is divided according to the clustering results and the preset human body part labels, and the regions corresponding to each human body part that matches the human body part labels are obtained.

[0087] Pixel values ​​are set for the regions corresponding to each human body part and the background region to obtain the human body part mask.

[0088] Specifically, threshold segmentation, region segmentation, edge segmentation, or clustering methods are used to perform background segmentation on human images (such as frontal or back views) from at least one direction, dividing the image into background and human target regions. Joints and / or skeletons are extracted from the human target regions to identify key structural features of the human body. Cluster analysis is then performed on the extracted joints and / or skeletons to determine the relative positional relationships of different parts of the human body. Based on the clustering results and predefined human body part labels (such as head, torso, limbs, etc.), the human target regions are precisely divided to obtain regions corresponding to each human body part that match the predefined labels. Pixel values ​​are assigned to each divided human body part's corresponding region and the background region, thereby generating a human body part mask map that clearly identifies the regions corresponding to each human body part and the background region.

[0089] In one embodiment of this application, such as Figure 3 As shown, in step S203, suspicious items in the first human image whose distance to the human contour boundary is less than the deduplication distance are taken as the deduplication targets. This can be achieved as follows:

[0090] Step S301: Determine the minimum distance from each boundary point of the suspicious item in the first human body image to each boundary point of the human body contour boundary, and use this distance as the distance from each boundary point of the suspicious item to the human body contour boundary.

[0091] In S301, the distance from each boundary point of the suspicious item in the first human body image to each boundary point of the human body contour boundary can be determined, and the minimum distance among them is taken as the distance from each boundary point to the human body contour boundary.

[0092] For example, the boundary of a suspicious item is a geometric shape (e.g., a rectangle) enclosed by four points. The set of boundary points for the nth suspicious item Cn in the first human image is {Pn1, Pn2, Pn3, Pn4}, where Pnm = P(Xnm, Ynm), P(Xnm, Ynm) are image pixel coordinates, n is the nth suspicious item, and m is the mth boundary point of the nth suspicious item, m ∈ {1, 2, 3, 4}. The human body contour boundary has several boundary points, Z is a set of several boundary points, Z = {Pz1, Pz2, Pz3, ...}, for example, Pz1 represents the image pixel coordinates of the first boundary point of the human body contour boundary.

[0093] For each boundary point of the nth suspicious item, such as Pn1, calculate its Euclidean distance to each boundary point (Pz1, Pz2, Pz3, ...) on the human body contour boundary. This yields a set of distances from boundary point Pn1 to each boundary point (Pz1, Pz2, Pz3, ...) on the human body contour boundary. The minimum value from this set is selected as the distance from boundary point Pn1 to the human body contour boundary. Similarly, the distances from each boundary point (Pn1, Pn2, Pn3, and Pn4) of the nth suspicious item to the human body contour boundary can be obtained.

[0094] Step S302: Determine the minimum distance from all boundary points of the suspicious item in the first human body image to the boundary of the human body contour, and use this distance as the distance from the suspicious item to the boundary of the human body contour.

[0095] For example, for the nth suspicious item, the minimum distance among all the boundary points (Pn1, Pn2, Pn3 and Pn4) of the nth suspicious item to the boundary of the human body contour is taken as the distance from the nth suspicious item to the boundary of the human body contour. The distance from the suspicious item to the boundary of the human body contour can be denoted as distance Lcz.

[0096] Step S303: Based on the distance from each suspicious item in the first human body image to the human body outline, suspicious items in the first human body image whose distance to the human body outline is less than the deduplication distance are taken as deduplication targets.

[0097] Specifically, based on the distances of n suspicious items to the human body boundary in the first human body image, i.e., n distances Lcz, the distances Lcz that are less than the deduplication distance are determined from the n distances Lcz, and the suspicious items corresponding to them are marked as targets to be deduplicated.

[0098] In this embodiment of the application, by calculating the minimum distance from each suspicious item in the first human body image to the human body contour boundary, the suspicious items located at the edge of the human body in the first human body image are accurately identified. These suspicious items will be repeatedly detected in the human body image from another perspective. By taking these suspicious items in the first human body image as the deduplication targets, the amount of computation for deduplication can be effectively reduced.

[0099] In one embodiment, such as Figure 4 As shown, step S103 involves deduplicating suspicious items in at least two human images from different directions and in a second human image from a different direction than the first human image, based on the target to be deduplicated, to obtain the recount quantity of suspicious items in the second human image. This specifically includes the following steps:

[0100] Step S401: One of the first human body image and the second human body image is a frontal image of the human body, and the other is a back image of the human body.

[0101] Step S402: Horizontally mirror the target to be deduplicated in the first human body image and map it to the second human body image to obtain the mirrored target to be deduplicated; wherein, the ordinate of the boundary point of the target to be deduplicated before mirroring is the same as the ordinate of the boundary point of the target to be deduplicated after mirroring.

[0102] Specifically, horizontal mirroring refers to transforming each boundary point of the target to be deduplicated into the second human image symmetrically from left to right, using the vertical midline of the first human image as the axis of symmetry. The boundary points of the target to be deduplicated before and after mirroring satisfy the following conditions: the ordinates of the boundary points before and after mirroring remain unchanged; the sum of the x-coordinates of the boundary points before and after mirroring equals the width of the first human image.

[0103] For example, assuming Pa(Xa, Ya) represents the pixel coordinates of the boundary point to be deduplicated before mirroring, and Pb(Xb, Yb) represents the pixel coordinates of the boundary point to be deduplicated after mirroring, and the width of the first human image is W, then their coordinates have the following relationship: Xb=W-Xa, Yb=Ya.

[0104] Step S403: Based on the mirrored target to be deduplicated, remove suspicious items in the second human image that overlap with the mirrored target to be deduplicated, and / or suspicious items whose center point is less than the expected distance from the center point of the mirrored target to be deduplicated;

[0105] Specifically, based on the mirrored target to be deduplicated, all suspicious items in the second human image are compared one by one to determine whether there are any suspicious items that overlap with or are too close to the mirrored target. Here, a suspicious item overlapping with the mirrored target is defined as one where, after the mirroring operation, the overlapping area of ​​two or more suspicious items exceeds an area overlap threshold (e.g., 90% or 80%, depending on the detection requirements).

[0106] For example, if a suspicious item in the second human image overlaps with the boundary of the mirrored target to be deduplicated by a region larger than a certain area, or if the distance between the center point of the suspicious item and the center point of the mirrored target to be deduplicated is less than a preset expected distance threshold, then these two targets are considered to be duplicate detections of the same suspicious item in images from different directions. The suspicious items in the second human image that meet the above conditions are deduplicated, thereby preventing the same suspicious item located at the edge of the human body from being counted repeatedly in human images from different directions. Further optionally, if there are suspicious items that both overlap with the mirrored target to be deduplicated and are too close to the mirrored target to be deduplicated, these suspicious items can be simultaneously identified as duplicate detections. The values ​​of the various thresholds mentioned in this embodiment can be flexibly configured according to detection requirements.

[0107] Step S404: The remaining number of suspicious items in the second human body image is used as the recounted number of suspicious items in the second human body image.

[0108] In this embodiment of the application, the target to be deduplicated in the frontal image of the human body is mapped to the corresponding position in the back image of the human body by mirror mapping, or the target to be deduplicated in the back image of the human body is mapped to the corresponding position in the frontal image of the human body, so as to perform matching and deduplication. This can simply and efficiently solve the problem of repeated counting of the same suspicious item located at the edge of the human body under multiple perspectives.

[0109] In one embodiment, based on the target to be deduplicated, suspicious items are deduplicated from at least two human images in different orientations and from a second human image in an orientation different from the first human image, to obtain the recount count of suspicious items in the second human image, including:

[0110] The first human body image is a frontal image and / or a back image of the human body, and the second human body image is a side image of the human body;

[0111] The coordinates of the center position of the suspicious item in the first human body image are mapped to the x and y coordinates in the world coordinate system, and the coordinates of the center position of the human body part corresponding to the center position of the suspicious item are mapped to the z coordinate in the world coordinate system, so as to obtain the first mapped coordinates of the suspicious item in the first human body image.

[0112] The first mapped coordinates are transformed into the second human body image to obtain the second mapped coordinates;

[0113] If the second mapped coordinates are located inside the suspicious item in the second human body image, then the suspicious item is deduplicated.

[0114] For example, the first human body image can be a frontal or back view of the human body, and the second human body image is a side view of the human body. Suspicious items identified in the first human body image are processed by mapping the image coordinates of the center position of the suspicious item to the world coordinate system, corresponding to the x and y coordinates of the world coordinate system, respectively. Simultaneously, the image coordinates of the center position of the human body part containing the suspicious item are mapped to the z coordinate of the world coordinate system, thus obtaining the first mapped coordinates (x, y, z) of the suspicious item in the world coordinate system. Based on the coordinate transformation relationship between the world coordinate system and the second human body image, the first mapped coordinates (x, y, z) are transformed into the image coordinate system of the second human body image to obtain the second mapped coordinates. It is determined whether the second mapped coordinates are located inside the bounding box of a suspicious item in the second human body image. If the second mapped coordinates fall within the bounding box of a suspicious item, it indicates that the suspicious item overlaps or is at a similar height to the suspicious item in the first human body image, constituting a duplicate detection of the same item. Therefore, the suspicious item in the second human body image is deduplicated. In this embodiment, this method can effectively avoid the same suspicious item being counted repeatedly in images from different perspectives, improving the accuracy and reliability of security inspection results.

[0115] In one embodiment, mapping the coordinates of the center position of the human body part corresponding to the center position of the suspicious item to the z-coordinate in the world coordinate system specifically includes:

[0116] The z-coordinate of the center point of the human body part on the same horizontal plane as the center of the suspicious item is used as the z-coordinate in the world coordinate system.

[0117] For example, to determine the spatial location of a suspicious item in the world coordinate system, the center position of the suspicious item in a first human image (such as a frontal or back view of the human body) is first identified, and its image coordinates are mapped to x and y coordinates in the world coordinate system. Using the horizontal plane where the center position of the suspicious item is located, the center point of the human body part on that plane (e.g., the horizontal center point of the waist, chest, or abdomen) is found, and the z coordinate of this center point is used as the z coordinate output in the world coordinate system. In this embodiment, using human body parts as a reference when determining the spatial location of a suspicious item can reduce fluctuations in the estimation of the suspicious item's location.

[0118] In an optional embodiment of this application, after estimating the location and deduplicating suspicious items, a prompt can be issued when the total number t after recounting exceeds a set alarm threshold. The prompt can take the form of an indicator light, a text message on a display screen, or an audio prompt; there are no limitations on this. The prompt information may include, but is not limited to, messages such as: "Please proceed with manual security check," "Excess quantity of special items," or a light or bell ringing. By only issuing a prompt when the total number of deduplicated suspicious items exceeds the alarm threshold, the false alarm rate and the frequency of manual security checks can be reduced, thus improving security check efficiency.

[0119] Based on the same concept, embodiments of this application also provide a millimeter-wave device, including a scanner, a processor, and a memory;

[0120] A scanner is used to scan objects.

[0121] The processor is configured to define suspicious items located at the human body contour boundary in a first human body image from at least two different orientations as targets to be deduplicated; to deduplicate suspicious items in a second human body image from at least two different orientations, which are in orientations different from the first human body image, based on the targets to be deduplicated, to obtain the recounted number of suspicious items in the second human body image; and to determine the total number of suspicious items carried by the scanned object based on the recounted number and the number of suspicious items in the first human body image.

[0122] The memory is used to store the total number of suspicious items carried by the scanned object.

[0123] Based on the same inventive concept, embodiments of this application also provide a device for detecting duplicate targets. Figure 5 This is a schematic diagram of the structure of the device for detecting duplicate targets provided in the embodiments of this application, as shown below. Figure 5 As shown, exemplarily, a device 50 for detecting target deduplication may include a first processor 51.

[0124] For example, the device 50 for detecting target deduplication may also include a memory 52 and a transceiver 53.

[0125] The first processor 51, memory 52, and transceiver 53 can be connected via a communication bus.

[0126] The following is combined Figure 5 The following example illustrates the various components of the device 50 used for detecting duplicate targets:

[0127] The device 50 for detecting duplicate targets may include the following components: the first processor 51 may be a single processor or a collective term for multiple processing elements. For example, the first processor 51 may be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement the embodiments of this application, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0128] For example, the first processor 51 can perform various functions of the device 50 for detecting target deduplication by running or executing software programs stored in memory 52 and calling values ​​stored in memory 52.

[0129] In a specific implementation, as one example, the first processor 51 may include one or more CPUs, for example... Figure 5 CPU0 and CPU1 are shown in the diagram.

[0130] As an optional embodiment, the device 50 for detecting target deduplication may also include multiple processors, such as... Figure 5 The first processor 51 and the second processor 54 are shown. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process values ​​(such as computer program instructions).

[0131] The memory 52 is used to store the software program that executes the solution of the present invention, and is controlled by the first processor 51 to execute it. The specific implementation method can be referred to the above-described embodiment of the target deduplication method, which will not be repeated here.

[0132] For example, memory 52 may be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having an instruction or value structure and accessible by a computer, but not limited thereto. Memory 52 may be integrated with the first processor 51 or may exist independently and be connected via an interface circuit of the device 50 for detecting target deduplication. Figure 5 (Not shown in the image) is coupled to the first processor 51, but this embodiment does not specifically limit this.

[0133] Transceiver 53 is used to communicate with network devices or with terminal devices.

[0134] For example, transceiver 53 may include a receiver and a transmitter. Figure 5 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0135] For example, transceiver 53 can be integrated with the first processor 51 or exist independently, and can be connected to the interface circuit of the device 50 for detecting target deduplication. Figure 5 (Not shown in the image) is coupled to the first processor 51, but this embodiment does not specifically limit this.

[0136] For example, the device 50 for detecting target deduplication may also include a display 55 for displaying rendering results. The display 55 may be a liquid crystal display or an electronic ink display. This application embodiment does not specifically limit this.

[0137] Furthermore, the technical effects of the device 50 for detecting target deduplication can be referred to the technical effects of the method for detecting target deduplication in the above method embodiments, and will not be repeated here.

[0138] Based on the same inventive concept, this application also provides an electronic device, including: at least one memory and at least one processor, wherein the at least one memory stores executable code, and the at least one processor is used to execute the executable code in the at least one memory to implement the above-mentioned detection target deduplication method.

[0139] It should be understood that the processor in the embodiments of this application can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0140] It should also be understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0141] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or value center to another website, computer, server, or value center via infrared, microwave, or other means. A computer-readable storage medium can be any available medium that a computer can access or a value storage device such as a server or value center that contains one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), solid-state drives, etc.

[0142] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0143] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0144] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0145] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0146] The above are merely preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A method for deduplicating detected targets, characterized in that, The method includes: Obtain suspicious items from at least two different orientations of the same scanned object; The suspicious items located at the boundary of the human body outline in the first human body image of the at least two human body images from different directions are defined as the targets to be deduplicated; Based on the target to be deduplicated, the suspicious items in the at least two human images from different directions and in the second human image from a different direction than the first human image are deduplicated to obtain the number of suspicious items in the second human image. Based on the recount count and the number of suspicious items in the first human image, the total number of suspicious items carried by the scanned object is determined.

2. The method for deduplicating detected targets according to claim 1, characterized in that, In the first human image of the at least two human images from different directions, suspicious items located at the boundary of the human body outline are defined as targets to be deduplicated, including: When the total number of suspicious items in the human body images from at least two different directions exceeds the re-detection threshold, the suspicious items located at the human body contour boundary in the first human body image from the at least two different directions are defined as targets to be deduplicated.

3. The method for deduplicating detected targets according to claim 1, characterized in that, In the first human image of the at least two human images from different directions, suspicious items located at the boundary of the human body outline are defined as targets to be deduplicated, including: Using human images from at least one orientation, determine the mask image of human body parts; Suspicious objects in the first human body image have multiple boundary points. Suspicious objects in the first human body image whose boundary points are located in the background area of ​​the human body part mask are selected as targets for deduplication; and / or Based on the human body part mask, the human body contour boundary is determined, and suspicious items in the first human body image whose distance to the human body contour boundary is less than the deduplication distance are identified as deduplication targets.

4. The method for deduplicating detected targets according to claim 3, characterized in that, Suspicious items in the first human image whose distance to the human contour boundary is less than the deduplication distance are identified as targets to be deduplicated, including: The minimum distance from each boundary point of the suspicious item in the first human body image to each boundary point of the human body contour boundary is determined as the distance from each boundary point of the suspicious item to the human body contour boundary. The minimum distance from all boundary points of the suspicious item in the first human body image to the boundary of the human body contour is determined as the distance from the suspicious item to the boundary of the human body contour. Based on the distance of each suspicious item in the first human body image to the boundary of the human body contour, suspicious items in the first human body image whose distance to the boundary of the human body contour is less than the deduplication distance are taken as deduplication targets.

5. The method for deduplicating detected targets according to claim 1, characterized in that, Based on the target to be deduplicated, suspicious items are deduplicated from the at least two human images from different directions, and from the second human image from a different direction than the first human image, to obtain the recount quantity of suspicious items in the second human image, including: One of the first human body image and the second human body image is a frontal image of the human body, and the other is a back image of the human body; The deduplicated target in the first human body image is horizontally mirrored and mapped to the second human body image to obtain the mirrored deduplicated target; wherein, the ordinate of the boundary point of the deduplicated target before mirroring is the same as the ordinate of the boundary point of the deduplicated target after mirroring. Based on the mirrored target to be deduplicated, remove suspicious items from the second human image that overlap with the mirrored target to be deduplicated, and / or suspicious items whose center point is less than the expected distance from the center point of the mirrored target to be deduplicated; The remaining number of suspicious items in the second human body image is taken as the recounted number of suspicious items in the second human body image.

6. The method for deduplicating detected targets according to claim 1, characterized in that, Based on the target to be deduplicated, suspicious items are deduplicated from the at least two human images from different directions, and from the second human image from a different direction than the first human image, to obtain the recount quantity of suspicious items in the second human image, including: The first human body image is a frontal image and / or a back image of a human body, and the second human body image is a side image of a human body; The coordinates of the center position of the suspicious item in the first human body image are mapped to the x and y coordinates in the world coordinate system, and the coordinates of the center position of the human body part corresponding to the center position of the suspicious item are mapped to the z coordinate in the world coordinate system, so as to obtain the first mapped coordinates of the suspicious item in the first human body image. The first mapped coordinates are transformed into the second human body image to obtain the second mapped coordinates; If the second mapped coordinates are located inside the suspicious item in the second human body image, then the suspicious item is deduplicated.

7. The method for deduplicating detected targets according to claim 6, characterized in that, The step of mapping the coordinates of the center position of the human body part corresponding to the center position of the suspicious item to the z-coordinate in the world coordinate system includes: The z-coordinate of the center point of the human body part on the same horizontal plane as the center of the suspicious item is taken as the z-coordinate in the world coordinate system.

8. The method for deduplicating detected targets according to claim 3, characterized in that, Using human images from at least one orientation, determine a mask image for human body parts, including: The background of the human image in at least one direction is segmented using threshold segmentation, region segmentation, edge segmentation, or clustering to obtain the background region and the human target region. Joints and / or skeletons are extracted from the target human body region, and the extracted joints and / or skeletons are clustered. The target human body region is divided according to the clustering results and the preset human body part labels to obtain the region corresponding to each human body part that matches the human body part label. Pixel values ​​are set for the regions corresponding to each human body part and the background region to obtain the human body part mask.

9. A device for detecting duplicate targets, characterized in that, The device for detecting target deduplication includes: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the method for deduplication of detected targets as described in any one of claims 1 to 8.

10. A millimeter-wave device, characterized in that, Includes scanner, processor, and memory; The scanner is used to scan the object being scanned; The processor is used to define suspicious items located at the boundary of the human body outline in the first human body image of at least two human body images from different directions as targets to be deduplicated; Used to remove duplicates of suspicious items in at least two human images from different directions and in a second human image from a different direction than the first human image, based on the target to be deduplicated, to obtain the recounted quantity of suspicious items in the second human image; used to determine the total number of suspicious items carried by the scanned object based on the recounted quantity and the number of suspicious items in the first human image; A memory for storing the total number of suspicious items carried by the scanned object.

Images