JP2022552754A - IMAGE DETECTION METHOD AND RELATED DEVICE, DEVICE, STORAGE MEDIUM, AND COMPUTER PROGRAM - Google Patents

Abstract

Embodiments of the present application disclose an image detection method and related devices, equipment, and storage media, and the image detection method includes the steps of: acquiring a first image including a target to be detected; detecting the first image; a step of obtaining a detection result of the image, the detection result including whether or not the detection target in the first image is occluded by a preset object; and a step of executing a predetermined operation that matches the detection result. and, including.

Description

(Cross reference to related applications)
This application is filed based on a Chinese patent application with application number 202011002322.1, dated September 22, 2020, and titled "Image Detection Method and Related Apparatus, Apparatus, and Storage Medium" and claiming priority of the Chinese patent application, which is incorporated herein by reference in its entirety.

TECHNICAL FIELD The present application relates to the technical field of image processing, and more particularly to image detection methods and related devices, equipment, storage media, and computer programs.

Currently, image processing, especially detection and recognition of targets in images, is widely used in various application scenes. Taking faces as an example, the detection and recognition of faces in images is widely used in fields such as finance, customs inspection, government, aviation, electricity, factories, education, and medicine. Conventional technology collects images or video streams that include faces using photographic equipment, automatically detects faces in the images, performs face recognition on the detected faces, and performs corresponding processing based on the recognition results. I do.

Embodiments of the present application provide at least an image detection method and related apparatus, apparatus, storage medium, and computer program.

Embodiments of the present application provide an image detection method. The image detection method includes acquiring a first image including a target to be detected, and detecting the first image to obtain a detection result of the first image, wherein the detection result is the target to be detected in the first image. including the steps of including whether the target is occluded by a preset object; and performing a predetermined operation in accordance with the detection result.

Therefore, by detecting the first image including the target to be detected, obtaining a detection result indicating whether or not the target to be detected is shielded, and then executing a predetermined operation that matches the detection result, the target to be detected is shielded. It is possible to determine whether or not the detection target has been detected, thereby performing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

In some embodiments, the detection result is obtained by detecting the first image with a neural network.

Therefore, detecting by a pre-trained neural network provides more accurate detection results and faster detection speed.

In some embodiments, performing the predetermined operation matching the detection result includes notifying a first reminder if the target to be detected is not obscured by a preset object, wherein the first reminder is , to notify the detection target to be shielded by a preset object.

Therefore, by notifying the first reminder when the target to be detected is not shielded, the situation in which the target to be detected is not shielded by a preset object is notified in a timely manner, thereby allowing the recipient to respond. can take timely action to

In some embodiments, the detection result further includes whether the shielding mode of the preset object of the target to be detected is a predetermined shielding mode, and performing a predetermined operation consistent with the detection result includes: If the target to be detected is occluded by the preset object and the occluder mode is not the predetermined occluder mode, notifying a second reminder, wherein the second reminder adjusts the occluded mode by the preset object. It is intended to notify

Therefore, if the shielding method is not correct, a second reminder is sent to timely adjust the shielding method of the detected target.

In some embodiments, the step of performing the predetermined operation matching the detection result includes removing from the first image at least a first unoccluded portion of the target to be detected if the target to be detected is occluded by a preset object. Extracting features as recognition pending features of the target to be detected; and recognizing the target to be detected using the recognition pending features to obtain recognition results.

Therefore, when the target to be detected is blocked by a preset object, by extracting and recognizing the features of the unshielded portion, recognition based on the local features of the target to be detected is realized, and the local features are Because it is not occluded, it can represent the target to be detected, which guarantees recognition accuracy to some extent.

In some embodiments, extracting from the first image at least a first feature of the unobstructed portion of the target to be detected as a pending recognition feature of the target to be detected comprises: extracting a first feature to obtain a second feature of an occluded portion of the target to be detected; and taking the first feature and the second feature as recognition pending features of the target to be detected.

Therefore, in addition to employing the features of the unshielded portion of the target to be detected, the features of the shielded portion are also employed, thereby further diversifying the features of the target to be detected.

In some embodiments, obtaining the second feature of the occluded portion of the target to be detected includes extracting the feature of the occluded portion from the first image as the second feature, or extracting the predetermined feature of the occluded portion as a second feature, wherein the predetermined features include features obtained based on at least one reference feature, each reference feature being an occluded portion within the reference target where no occluded portion exists. is obtained by extracting from the region corresponding to .

Therefore, as for the method of specifying the features of the shielded portion, the feature of the shielded portion may be directly extracted. Alternatively, the predetermined feature may be obtained as the feature of the obscured portion, this method does not need to perform feature extraction on the obscured portion, saving processing resources to be consumed. , can increase the processing efficiency.

In some embodiments, the step of recognizing the target to be detected using the recognition pending feature and obtaining the recognition result includes combining the recognition pending feature and the preset target if one preset target is included. obtaining a first similarity measure with a pre-stored feature, and determining that if the first similarity measure satisfies a first predetermined condition, the recognition result includes successful identity verification of the target to be detected; If multiple preset targets are included, a second similarity between the feature waiting for recognition and the pre-stored feature of each preset target is obtained respectively, and the recognition result determines the identity of the target to be detected. 2 including identifying as a preset target identity corresponding to a second similarity measure that satisfies a predetermined condition.

Thus, by calculating a first similarity measure of a particular preset target with pre-stored features, or calculating similarity measures of a plurality of preset targets with pre-stored features, Depending on the requirements of the actual scene, the target to be detected can be compared to a specific pre-set target or to a pre-set target in some database.

In some embodiments, the image detection method includes: the first predetermined condition includes that the first similarity is greater than the first similarity threshold; and the second predetermined condition is that the second similarity is greater than the second similarity. including being greater than a threshold.

Therefore, by setting the first similarity thresholds for different scenes, more accurate recognition results can be obtained.

In some embodiments, the image detection method sets the first similarity threshold when the feature to be recognized includes the second feature of the occluded portion of the target to be detected, and the similarity threshold when the feature to be recognized does not include the second feature. and making the second similarity threshold when the feature to be recognized includes the second feature smaller than the second similarity threshold when the feature to be recognized does not include the second feature and at least one of

Therefore, when the second feature is included in the recognition waiting feature, the second feature may differ from the actual feature of the key point of the shielded portion of the target to be detected. can improve the accuracy of

In some embodiments, prior to obtaining a first similarity measure between the features awaiting recognition and the pre-stored features of the preset target, the image detection method further comprises, in response to an account registration request, asking the user: a step of registering an account of; identifying a second image that satisfies a predetermined quality requirement from the second images of at least one frame obtained by photographing the user; extracting the features; associating the features of the given region with the account and storing the features of the given region as pre-stored features of a preset target.

Therefore, by first specifying the second image that satisfies the quality requirements and extracting the features of the predetermined part, more accurate features can be extracted.

In some embodiments, if the target to be detected is occluded by a preset object, before extracting the first feature of at least the unoccluded portion of the target to be detected from the first image, the image detection method further comprises: extracting, as a first image for subsequent feature extraction, a first image that satisfies a predetermined quality requirement from first images of a plurality of frames including a target to be detected; and performing liveness detection on the first image for subsequent feature extraction, and if the detection target is a living body detection result, at least the unshielded portion of the detection target is detected from the first image. extracting one feature and deciding to perform subsequent steps.

Therefore, by performing preprocessing before feature extraction, more accurate features can be extracted and the target to be detected can be recognized only when the target is a living body, thereby improving the safety of recognition. To some extent, counterfeit attacks can be prevented.

In some embodiments, identifying a first image for subsequent feature extraction from the first images of the plurality of frames containing the target to be detected that satisfies a predetermined quality requirement as the first image for subsequent feature extraction comprises: obtaining a corresponding quality score of the first image of each frame based on the quality factor of the image, the quality factor of the first image being pose information of the target to be detected relative to the imaging device; including at least one of parameter information for representing a target magnitude and luminance information of the first image; selecting the first image satisfying predetermined quality requirements based on the quality score as the first image for subsequent feature extraction; identifying as an image, wherein the selected first image has a higher quality score than the other first images.

Therefore, by identifying images whose quality scores meet the requirements and performing feature extraction, features are extracted that can further represent the detection target.

In some embodiments, the step of preprocessing the first image for subsequent feature extraction includes detecting targets in the first image of targets of interest that meet predetermined extraction requirements if the first image includes a plurality of targets of interest. identifying a region and removing image portions other than the target region in the first image; Rotating the first image so that the tilt angle is less than or equal to the predetermined angle.

Therefore, when a plurality of detection target targets exist in the first image, only the detection target that satisfies the predetermined extraction requirements is specified, and the detection target that does not satisfy the requirement is discarded, thereby eliminating the detection target that does not satisfy the requirement. If the tilt angle of the target to be detected in the first image is greater than the predetermined angle, the position of the target to be detected is corrected to reduce the effect of the tilt of the target to be detected. be.

In some embodiments, the predetermined extraction requirement includes that the area of the corresponding region of the detectable target is greater than the area of the corresponding region of the other detectable target, and the other detectable target is a target other than the detectable target. including.

Therefore, the larger the area of the target to be detected, the more accurate the features to be extracted. Therefore, by selecting the target to be detected having a larger area, a more accurate detection result can be obtained.

In some embodiments, the target to be detected includes a face and the preset object includes a mask.

Therefore, by determining whether the face is wearing a mask and performing corresponding operations, for example, if the face is not wearing a mask or the mask is not applied correctly, a corresponding reminder will be sent to the user. It can be adjusted in a timely manner and performs recognition etc. on the face when the mask is worn on the face.

Embodiments of the present application provide an image sensing device. The image detection device includes an image acquisition module configured to acquire a first image including a target to be detected, and a target detection module configured to detect the first image and obtain a detection result of the first image. wherein the detection result includes whether or not the target to be detected in the first image is occluded by a preset object; and an operation configured to perform a predetermined operation consistent with the detection result. an execution module;

An embodiment of the present application provides an electronic device. The electronic device includes a memory and a processor, and the processor is used to execute program instructions stored in the memory so as to implement the image detection method described above.

Embodiments of the present application provide a computer-readable storage medium having program instructions stored therein which, when executed by a processor, implement the image detection method described above.

An embodiment of the present application provides a computer program product comprising computer readable code for, when executed in an electronic device, causing a processor in the electronic device to perform the image detection method described above.

Embodiments of the present application provide an image detection method and related apparatus, equipment, storage medium, and computer program for detecting a first image containing a target to be detected to obtain whether the target to be detected is occluded, and then , by performing a predetermined operation that matches the detection result, it is possible to determine whether or not the target to be detected is occluded, thereby performing a subsequent predetermined operation that matches the detection result, and shielding the target to be detected in the image Realize flexible processing according to the state.

It should be understood that the above general description and the following detailed description are merely illustrative for interpretation and are not limiting of the present application.

1 is a flowchart of one embodiment of an image detection method according to embodiments of the present application; FIG. 4 is a schematic diagram of a first image in one embodiment of an image detection method according to an embodiment of the present application; FIG. 4 is a schematic diagram of a first image after preprocessing in one embodiment of an image detection method according to an embodiment of the present application; 1 is a configuration diagram of an example of an image detection apparatus according to an example of the present application; FIG. 1 is a configuration diagram of an example of an electronic device according to an example of the present application; FIG. 1 is a configuration diagram of an embodiment of a computer-readable storage medium according to an embodiment of the present application; FIG.

The drawings herein are incorporated into and constitute a part of the specification, and the drawings show embodiments consistent with the present application and are used together with the specification to explain the technical solutions of the present application.

In the following, the solutions of the embodiments of the present application are described in detail with reference to the drawings of the specification.

In the following description, for purposes of explanation rather than limitation, specific details of particular system configurations, interfaces, techniques, etc. are provided to aid in a thorough understanding of the present application.

The term "and/or" herein is only for describing a related relationship of related subjects and indicates that three types of relationships can exist, e.g., A and/or B are Three cases may be shown: A exists alone, A and B exist simultaneously, and B exists alone. Also, the character "/" in this specification generally indicates that the related objects before and after are in an "or" relationship. Also, "plurality" as used herein means two or more. Also, as used herein, the term "at least one" means any one of a plurality or any combination of at least two of a plurality, e.g., at least one of A, B, C may indicate that includes any one or more elements selected from the set consisting of A, B and C.

The image detection methods in embodiments of the present application can be used in devices with image processing capabilities. The device may also have image or video gathering capabilities, for example, the device may include image or video gathering components such as cameras or the like. Alternatively, the device acquires the required video stream or image from the other device or accesses the required video stream or image from the storage resource of the other device by way of data transmission or data exchange with the other device. You can The other device has the function of image acquisition or video acquisition, and has a communication connection with the device, for example, the device performs data transmission or data exchange with other devices in the form of Bluetooth, wireless network, etc. and embodiments of the present application do not limit the communication schemes between the two here, and may include, but are not limited to, the cases listed above. In one embodiment, the device may include a mobile phone, tablet, interactive screen, etc., and is not limited here.

FIG. 1 is a flowchart of one embodiment of an image detection method according to an embodiment of the present application. Here, the image detection method is performed by an electronic device, and the method may include the following steps.

At step S11, a first image including a target to be detected is acquired.

the first image containing the target to be detected may be an initial image containing the target object collected by directly invoking the camera of the device, or may of course be an image acquired from another device; Alternatively, it may be an image that has undergone frame selection, luminance or resolution adjustment, or the like. Detectable targets may include faces, other animal facial parts or extremities, and the like. Therefore, the method of acquiring the first image including the target to be detected is not limited. Other devices refer to devices that cannot operate without relying on their respective central processing units.

In step S12, detecting the first image to obtain a detection result of the first image, the detection result including whether the target to be detected in the first image is blocked by a preset object.

A preset object refers to any object that can shield the target to be detected, such as a visible object such as a mask, a scarf, eyeglasses or an arm, paper, and the like.

It is necessary to detect the first image, and in the detection process, it is necessary to detect whether or not the first image includes an image waiting for detection. determine whether or not A method for determining whether a target to be detected is occluded by a preset object is to first train an occlusion detection model before detecting the first image, and train the first image to an occlusion detection model (e.g., an occlusion detection model). It may be a neural network having a function), thereby grasping whether or not the target to be detected in the first image is blocked by a preset object. Of course, in some embodiments, the method of determining whether a target to be detected is occluded by a preset object includes whether a predetermined detection location on the target to be detected is occluded and whether a predetermined detection position is occluded. It may be determined whether or not the object blocking the position satisfies a preset object condition. A detection result including whether or not the target to be detected is shielded may be obtained by extracting features of the shielding object and a predetermined shielding object and determining the degree of similarity.

In step S13, a predetermined operation matching the detection result is executed.

The detection result indicates that the target to be detected was shielded, but not by a preset object, or that the target to be detected was shielded by a preset object, but the shielding method was set in advance. the target to be detected is obscured by a preset object and the occlusion method is the same as the preset method; or the target to be detected is not occluded at all. In the embodiments of the present application, the situation is considered whether the target to be detected in the first image is occluded by a preset object. Of course, in some embodiments, the preset object can be set as any object, that is, adopting the occlusion detection model, if it can be determined that the target to be detected is occluded, the corresponding predetermined operation to run. The predetermined operation may be any operation related to target detection, such as recognition.

The above solving means detects whether or not the detection target is shielded by detecting a first image including the detection target, and then executes a predetermined operation that matches the detection result, so that the detection target is shielded. It is possible to determine whether or not the detection target has been detected, thereby performing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

In some embodiments, the detection result is obtained by detecting the first image with a neural network. Before detecting the first image, a preset object occlusion model is first trained, so that the preset object occlusion model detects that the target to be detected in the first image is occluded by the preset object. can detect whether or not The preset object may be one, or it may be multiple, for example, two or three different objects, and if there are multiple preset objects, the preset object occlusion model is , it is possible to determine whether or not the target to be detected is shielded by a preset object, and to detect which predetermined object shielded the target to be detected. The target to be detected may be a face, and the preset object may be a mask. Accordingly, the preset object occlusion model is the mask detection model. The mask detection model can detect whether the detected target is wearing a mask and, of course, in some embodiments, if the detected target is wearing a mask, it can also detect whether the masking is correct. . Detecting by a pre-trained neural network provides more accurate detection results and faster detection speed.

If the target to be detected is not blocked by a preset object, a first reminder is sent, and the first reminder is for notifying that the target to be detected is blocked by the preset object. The first reminder may have various reminder forms, including framing the face, and when detecting that the target to be detected is not obstructed by a preset object, the face area is framed. Framed, the face framed at this time may be colored for warning, such as red or yellow, and the first reminder may be a combination of face framed and reminder wording, and the reminder wording is For example, "You are not wearing a mask, so please wear a mask." Of course, the first reminder may be in the form of a reminder voice or a blinking reminder lamp, etc. Obviously, these formats are multiple. They may be used in combination or alone and are not limited here. For example, if the mask detection model detects that no mask is being worn on the face, a first reminder to wear a mask on the face to cover the mouth and nose of the face is issued. By sending a first reminder when the target to be detected is not blocked, timely notification of the situation where the target to be detected is not blocked by a preset object, so that the notified person can take corresponding measures. can be taken in a timely manner.

In some embodiments, the detection result further includes whether the shielding mode of the preset object of the target to be detected is a predetermined shielding mode. When training a neural network called a preset object occlusion model, the training samples may be labeled with a predetermined occlusion scheme, where the predetermined occlusion scheme may be an accurate occlusion scheme, whereby the detection so that when it is detected that the target is blocked by a preset object, the preset object occlusion model can determine whether the occlusion method by the preset object is the predetermined occlusion method; Train a preset object occlusion model. If the target to be detected is blocked by a preset object and the blocking method is not the predetermined blocking method, a second reminder is sent. The second reminder is for notifying to adjust the shielding method with the preset object. For example, the target to be detected is a face, the preset object is a mask, and the predetermined masking method is an accurate masking method. When it detects that a mask is worn on the face, it determines whether the method of wearing the mask is correct. Send a second reminder to adjust. Of course, in some embodiments, the predetermined shielding scheme may vary, such as a precise shielding scheme, a first error shielding scheme, a second error shielding scheme, etc., and may be a predetermined number of targets to be detected. when detecting that the shielding method of the object is the first error shielding method, a reminder corresponding to the first error shielding method is sent, and the shielding method of the preset object of the detection target is the second error shielding method; If it detects that it is the correct shielding method, it will send a reminder corresponding to the second error shielding method, thereby notifying to adjust the shielding method of the target to be detected to the correct shielding method. For example, similarly if the target to be detected is a face and the preset object is a mask, the correct occlusion scheme is that the mask covers both the nose and mouth, and the first false occlusion scheme is that the mask covers the nose. the mask covers the mouth but does not cover the mouth. It is not covered, and the reminder corresponding to the second error masking scheme is to inform the nose of the face to be covered as well. If the shielding method is not correct, send a second reminder to timely adjust the shielding method of the target to be detected. The format of the second reminder is also similar to the format of the first reminder, a combination of face frame and reminder wording, a combination of face frame and reminder voice, or a single reminder wording or a single reminder voice or reminder It may be in the form of blinking a lamp, etc. Of course, if there are a plurality of different predetermined shielding methods, a plurality of reminder messages or voices may be set accordingly, for example, if the predetermined shielding method is the first error shielding method. For example, the reminder text is associated with the first error shielding scheme.

In some embodiments, the target to be detected is recognized when the target to be detected is blocked by a preset object in some business scene. In other business scenes, if the target to be detected is not shielded by a preset object, it will not recognize the target to be detected, such as high-speed trains or aircraft during certain times or special periods. When it is necessary to put on, if it is detected that the face is not wearing a mask, face recognition is not performed on the face, and the face not wearing the mask cannot enter by face recognition. Of course, according to business scene requirements, even if it is detected that the detection target is not shielded by a preset object, the detection target can be recognized. In order to recognize the target to be detected, it is necessary to perform feature extraction on the target to be detected. The image may be identified as the first image for subsequent feature extraction. A method of identifying a first image that satisfies predetermined quality requirements as a first image for subsequent feature extraction is based on a quality factor of the first image of each frame to obtain a corresponding quality score of the first image of each frame. Preferably, the quality factor of the first image is at least one of orientation information of the target to be detected with respect to the imaging device, parameter information for representing the size of the target to be detected in the first image, and luminance information of the first image. including one. The orientation information of the detection target with respect to the imaging device may be angle information of the detection target with respect to the imaging device. The angle information of the target to be detected with respect to the photographing device may be the angle information of the target to be detected with respect to the lens at the time of photographing. For example, a three-dimensional coordinate system is constructed with the lens as the origin, the line connecting the lens and the center of the earth is the X-axis, the line extending just before the lens and perpendicular to the X-axis is the Y-axis, and the X-axis is the X-axis. A line that is both perpendicular to the Y-axis is the Z-axis. The three-dimensional coordinate system is merely for representing the angle between the target to be detected and the imaging device. You may The angle may be divided into angles in the XYZ directions with respect to the lens. For example, when the target to be detected faces the front of the lens, the angles in the XYZ directions are all 0° (degrees), and the side surface of the target to be detected is the first image. When facing the front of the acquisition component, the detected target has an X angle of 90°, a Y angle of 0°, and a Z angle of 0° with respect to the first image acquisition component, where the detected target is rotated about the X axis by 90°, the angle between the target to be detected and the X axis, that is, the angle of the target to be detected with respect to the X direction is 90°. Naturally, the smaller the angle in each direction, the better. The parameter information for representing the size of the target to be detected in the first image includes the size of the area occupied by the target to be detected in the first image, and the size of the area is the size of the area occupied by the target to be detected in the first image. It can be expressed in terms of size. Of course, this assumes that the target to be detected is completely contained within the first image, and if the first image only partially contains the target to be detected, then in the first image of this frame the target to be detected is Lower scores for the size quality factor. The higher the brightness information of the first image, the better, but the closer to the brightness of the natural light at the current time, the better, and the score of this quality factor is accordingly relatively high. The weights of the three quality factors are set according to the relationship of the influence of the three quality factors on the image quality. For example, let the angle weight be 0.4 and the remaining two be 0.3 each, of course this is just an example, the weight between each quality factor may be set as desired, and as many In some embodiments, in addition to these three quality factors, further factors such as the degree of blurring of the first image may be included, and any factor that can affect the image quality can be used to calculate the quality score of the image. . By selecting images whose quality scores satisfy the requirements and performing feature extraction, features are extracted that can further represent the target to be detected. Of course, the setting of the weights may take into consideration the actual requirements for image detection accuracy, the processing capability of the image detection device, the occupation of resources, and the like. For example, in some embodiments, if the processing power of the image detection device is high and the resources occupied are low, the quality score may be calculated considering multiple quality factors, and the processing power of the image detection device is If very low, the quality score may be calculated using several quality factors accordingly, e.g., depending on the time required to calculate each quality factor or the memory space occupied. select. Thus, there is flexibility in choosing how many or which quality factors to use. Of course, in some embodiments, a lower quality score threshold is determined, and if the quality score of a first image falls below the quality score threshold, it is eliminated, and the first image with a quality score greater than the quality score threshold is discarded. may be retained.

In some embodiments, the first image may be preprocessed for subsequent feature extraction before performing feature extraction on the first image for recognition. In the preprocessing method, when a plurality of targets to be detected are included in the first image, a target region in the first image of the target to be detected that satisfies a predetermined extraction requirement is specified, and an image portion other than the target region in the first image is identified. may be removed. The target region here may be a region containing one target to be detected. In other words, when the first image contains a plurality of targets to be detected, the recognition process is limited to some extent in order to recognize only the target regions of the targets to be detected that satisfy the predetermined extraction requirements, rather than recognizing the complete first image. The noise due to other detectable targets in is reduced, thereby reducing the impact of detectable targets that do not meet the requirements on recognition results. The predetermined extraction requirement may be that the area of the corresponding region of the target to be detected is larger than the area of the corresponding region of the other target to be detected, and the other target to be detected includes targets other than the target to be detected. If there are multiple targets to be detected in the first image, the areas occupied by the multiple targets may not match. is selected and recognized. If there are multiple detectable targets with the same area, the detectable target whose center is closer to the center of the first image may be recognized, or in some other embodiments, all detectable targets Of course, the size of the area of all the targets to be detected described in the latter is ranked first, or both of the areas are the predetermined area extraction thresholds. Pointing at a larger detectable target.

In some embodiments, before performing feature extraction on the first image for recognition, the step of preprocessing the first image for subsequent feature extraction includes: is greater than a predetermined angle, and the first image may be rotated so that the tilt angle of the target to be detected is less than or equal to the predetermined angle. In some embodiments, the rotation method may rotate the target of interest, or only the target region containing the target of interest, other than rotating the entire first image, thus repositioning the target of interest. is not limited here. The predetermined angle may be within 0° to 180° clockwise or counterclockwise, and in embodiments herein the predetermined angle is 0°, and in some embodiments the predetermined angle is 30°, 35°, etc. may A method for determining whether or not a target to be detected is inclined at a predetermined angle includes obtaining an included angle between a line connecting a predetermined first key point and a predetermined second key point in the target to be detected and a vertical line, determining whether the included angle is greater than a predetermined angle; if so, the included angle is less than or equal to the predetermined angle and the predetermined first keypoint after rotation is positioned above the predetermined second keypoint; Alternatively, the first image may be rotated, where the top is determined relative to the bottom of the first image. Of course, the tilt angle may also be the tilt angle of the target to be detected with respect to a position in the first image, eg the tilt angle of the target to be detected with respect to the center of the first image. Of course, the predetermined angle here may be set according to the requirements of different scenes, for example, it may be determined according to the area occupied by the area in which the target to be detected is located in the first image. For example, when the area of the region in which the target to be detected is located is larger than the predetermined value of the first area, the predetermined angle may be larger than 30°, and the area of the region in which the target to be detected is located is the predetermined value of the second area. If less than the value, the predetermined angle may be less than 30°. The larger the area of the region in which the target to be detected is located, the larger the area of the target to be detected. Furthermore, since the target to be detected is greatly affected by the tilt angle, the tolerance for the tilt angle of the target to be detected is reduced. Of course, this is only an example, and in some embodiments other correspondences between the area and the predetermined angle are possible, and are not limited here.

For example, FIG. 2 is a schematic diagram of the first image in one example of the image detection method according to the embodiment of the present application, and FIG. It is a schematic diagram of a first image. As shown in FIG. 2, the lower half of the target to be detected 21 in the first image 20 is blocked by the preset object 22, and the target to be detected 21 is clearly tilted to the left, that is, the target to be detected The included angle between the straight line connecting the upper left corner point (first predetermined key point) and the lower left corner point (second predetermined key point) of 21 and the vertical line is 30°. 30°, which is greater than the predetermined angle of 0°. At this time, the first image 20 is rotated to the right, that is, clockwise by 30°, and the first image after rotation is as shown in FIG. The included angle between the straight line connecting the upper left corner point (first predetermined key point) and the lower left corner point (second predetermined key point) of the detection target 21 and the vertical line is 0°, which is equal to the predetermined angle of 0°.

If the inclination angle of the detection target in the first image is greater than a predetermined angle, correcting the position of the detection target makes it possible to detect the detection target in the subsequent process of performing living body detection or target recognition on the detection target. The influence of inclination is reduced.

In some embodiments, before performing feature extraction on the first image and recognizing it, liveness detection is performed on the subsequent first image for feature extraction, and a liveness detection result in which the target to be detected is a living body. , it may be determined to perform the step of extracting from said first image at least a first feature of the unoccluded portion of the target to be detected and subsequent steps. When a plurality of detection target targets exist in the first image, the detection target target having the largest area is selected to perform living body detection. The liveness detection may be performed by inputting a target region corresponding to the target to be detected into a liveness detection model, and the liveness detection model is trained with a plurality of images including the target to be detected obscured by a preset object. It is obtained. By recognizing the target to be detected only when the target to be detected is a living body, the safety of recognition is improved, and counterfeit attacks can be prevented to some extent.

In the process of recognizing the target to be detected, first, at least the first feature of the unoccluded portion of the target to be detected is extracted from the first image as a recognition waiting feature of the target to be detected. A first feature refers to a feature of an unoccluded keypoint within the detected target. Extracting a first feature of the unshielded portion of the detection target from the first image, obtaining a second feature of the shielded portion of the detection target, and using the first feature and the second feature as recognition waiting features of the detection target good too. The second feature here is the keypoint feature of the occluded portion of the detected target. There may be two methods for acquiring the second feature of the shielded portion, one of which is to extract the feature of the shielded portion from the first image as the second feature. That is, although this portion is blocked by a preset object, the second feature of the blocked portion is extracted by the method of extracting the first feature as unoccluded, that is, the detection target is blocked by the preset object. Whether occluded or not, the same processing mechanism is adopted, in other words, occluded by a preset object or not does not affect the feature extraction process. Of course, in some embodiments, even if there is no pre-set object occluding the target, this method can still extract keypoint features within the target. For example, if the target to be detected is a face and the preset object is a mask, the second feature extraction method assumes that the face is not occluded by the mask and extracts the features of each keypoint in the face, i.e. , adopts the same processing mechanism for mask-wearing and non-mask-wearing faces, in other words, wearing a mask does not affect the feature extraction process. Another method is to obtain a predetermined feature of the occluded portion as the second feature, the predetermined feature may be a feature obtained based on at least one reference feature, each reference feature being the occluded It is obtained by extracting from the area corresponding to the occluded part in the reference target where no part exists. That is, before recognizing the first image, the reference features are set in advance for the key points of the occluded portion, that is, the features of the occluded portion are complemented first. For example, a plurality of features of a predetermined portion corresponding to the detection result in which the shielded portion does not exist in the detection target are extracted in advance, and then the average value of the extracted plurality of features is interpolated to determine the portion shielded by the object. As a reference feature, a plurality of features of a predetermined portion corresponding to the detection target in the detection result that is not blocked by a preset object are extracted in advance, and then the average value of the extracted features is interpolated. may be used as a reference feature of a portion occluded by a preset object. For example, a plurality of features of the corresponding predetermined parts of the face not wearing a mask, that is, the features of the parts to be masked such as the nose and mouth are extracted in advance, and the average value of the extracted features is interpolated, A predetermined reference feature of the portion covered by the mask. As for the method of specifying the features of the shielded portion, the features of the shielded portion may be directly extracted. can be improved, or the predetermined feature may be obtained as a feature of the obscured portion, the method does not need to perform feature extraction on the obscured portion, reducing the processing resources consumed, Processing efficiency can be improved.

In some embodiments, after obtaining recognition-pending features of a target to be detected, the recognition-pending features are used to recognize the target to be detected. Scenes of recognition can be divided into 1:1 scenes and 1:N scenes, where 1:1 refers to the comparison between two features, and 1:N refers to the comparison between one feature and multiple features. refers to a comparison with In a 1:1 scene, i.e., when one preset target is included, obtain a first similarity between the feature to be recognized and the pre-stored feature of the preset target; If the degree satisfies a first predetermined condition, determine that the recognition result includes successful identity authentication of the target to be detected. The first predetermined condition may be that the first similarity is greater than a first similarity threshold. A first similarity threshold when the feature to be recognized includes the second feature of the occluded portion of the target to be detected is set smaller than the first similarity threshold when the feature to be recognized does not include the second feature. When the recognition waiting features include the second feature, the second feature may differ from the actual feature of the key point of the shielded portion of the target to be detected. can improve sexuality. If the features pending recognition do not include the second feature, the first similarity threshold may be determined and selected based on the ratio of the number of occluded keypoints to the total number of keypoints of the target to be detected. . For example, if the keypoints of the occluded portion are one-third of the total number of keypoints of the detected target, the first similarity threshold is one-third of the similarity threshold for recognizing the unoccluded detected target. may be determined. At this time, when the second feature is included in the recognition-waiting features, the first similarity threshold is set to be 0.1 smaller than the first similarity threshold when the recognition-waiting features do not include the second feature, or other It may be set to a smaller value, and is not limited here. For example, the similarity threshold for recognizing unoccluded targets of interest may be between 0.6 and 1. Of course, this is only an example, and in some embodiments, the first similarity threshold when the feature to be recognized includes a second feature of an occluded portion of the target to be detected is such that the feature to be recognized includes the second feature. It may be equal to the first similarity threshold if it is not included, and if the feature waiting to be recognized includes the second feature, the above method may be determined according to the actual situation.

In some embodiments, before obtaining a first degree of similarity between the awaiting recognition feature and the pre-stored features of the preset goal, the user account and the pre-stored features of the preset goal are first associate. Its embodiments are as follows. Register an account for a user in response to an account registration request. The account here may be any electronic payment account, and any application capable of target recognition may respond to an account registration request to register the user's account. The user can register with the mobile phone number in the corresponding application, and after the registration is successful, the user will get the information such as user name, password and so on. A second image that satisfies a predetermined quality requirement is specified from the second images of at least one frame obtained by photographing the user, and features of a predetermined part of the user are extracted from the specified second images. The predetermined portion here is the same as the predetermined portion of the target to be detected. The step of selecting the second image meeting the predetermined quality requirement is the same as the step of selecting the first image meeting the predetermined quality requirement above. Finally, the feature of the site is associated with the account, and the feature of the site is saved as a pre-stored feature of a preset target. That is, the predetermined site of the user is a preset target.

In a 1:N scene, i.e., when a plurality of preset targets are included, a second similarity between the recognition-waiting feature and the pre-stored feature of each preset target is obtained, respectively, and the recognition result is obtained. includes identifying the identity of the target to be detected as the identity of the preset target corresponding to the second similarity that satisfies the second predetermined condition. The second predetermined condition may be that the second similarity is greater than a second similarity threshold. In general, satisfying the second predetermined condition referred to here means not only being greater than the second similarity threshold, but in many cases, selecting the maximum value from all the second similarities, i.e., selecting the maximum value. , select the preset target identity corresponding to the largest second similarity as the identity of the target to be detected. A second similarity threshold when the features to be recognized include the second features is set smaller than a second similarity threshold when the features to be recognized do not include the second features. When the recognition waiting features include the second feature, the second feature may differ from the actual feature of the key point of the shielded portion of the target to be detected. can improve sexuality. The method for determining the second similarity threshold is the same as the method for determining the first similarity threshold.

1:N can be a scene involving multiple faces, for example, a scene in which face recognition gates are implemented at the doorway of an office building or company, and in such a scene, within the confines of the building or company, It is necessary to register each person to try and form a face library, when the registered person appears in front of the gate, the camera on the gate will detect the face, snap it, and snap the face to those in the face library, if the comparison is successful, open the gate, and when an unregistered person appears at the gate, the comparison is unsuccessful and the gate does not respond.

The above solving means detects whether or not the detection target is shielded by detecting a first image including the detection target, and then executes a predetermined operation that matches the detection result, so that the detection target is shielded. It is possible to determine whether or not the object has been detected, thereby executing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

Based on the above embodiments, embodiments of the present application further provide an image detection method. The image detection method uses the model detection ability based on the deep learning algorithm, starting from the face verification scene, realizing the image detection flow in the 1:1 and 1:N scenes, and detecting whether the mask is worn or not. We provide a method for inspecting and embodiments of face comparison and face search in the process when wearing a mask.

(1) Face verification scenes mainly include two scenes: 1:1 and 1:N.

For example, 1:1 in the payment scene means that 1:1 verification is performed between the face photo snapped in real time and the library image bound to the member, and if it is confirmed that it is the same user, authentication is successful. do. Also, for example, 1:N scenes often involve face retrieval, such as scenes in which face recognition gates are implemented at the entrances and exits of office buildings or companies. , it is necessary to register each user who wants to enter and exit, forming a face library. When the user appears in front of the gate, the camera on the gate detects the face, snaps it, compares the snapped face with the images in the face library, and if the comparison succeeds, opens the gate and the comparison takes place. If it fails, the gate does not respond, i.e. the gate remains closed.

In some embodiments, live crime prevention is necessary in both 1:1 and 1:N scenes. For example, it is necessary to prevent a dummy attack using a photograph (a photograph obtained by photographing a person, an electronic composite photograph, etc.), a mask, or the like.

(2) The image detection method for a 1:1 scene may be implemented as follows.

First, the user has an account, eg an electronic payment account. A user registers with a mobile phone number in the application system corresponding to said electronic payment account, and after successful registration, the user obtains information such as user name, password, etc., that is, obtains an electronic payment account. Then, the application system induces the user to perform a face binding operation by some event, and if the user passes the biometric authentication, performs face recognition, and if the quality of the face in the video meets the requirements of face collection. , to collect the user's facial image. A frame of the highest quality from the shooting process may be selected, and the quality criteria include one or more of the following dimensions: face angle, lighting intensity, face size, and the like.

The collected facial images of the user may be associated with the account, specifically with the identifier of the account, and may be comparison images in the face library. When the user pays online, the payment stage is entered after the order is confirmed, and if the user performs face image binding at this time and chooses face payment, the face snap shooting stage is entered. When the face is detected, the wearing of the mask is checked (that is, the face attribute is detected). If the mask is worn, the subsequent flow is continued. The user may be notified to wear a mask.

If the user is wearing a mask, perform face detection, select one frame with the highest quality within this period of the imaging process, and align the face to the image of this selected frame. (for example, if the face has a certain tilt angle, the face position is corrected), and facial feature values are extracted based on the face image after alignment.

There are two methods for extracting and comparing facial feature values. The first is to extract facial feature values that include masked parts, for example, facial features without a mask in images in the face library. Let A be the value, and perform feature extraction on the face wearing the mask in a similar manner to extract the feature value A1. Compare the feature vectors of the feature values A for the cases. The second is to extract the face feature values of the visible part above from the mask, assuming that there are 128 keypoints in every face and 64 keypoints above the mask, Compare the feature values extracted from the 64 keypoints of the visible part above with the feature values extracted from the corresponding 64 keypoints in the face library images. Further, the generated feature value is compared with the feature value of the image in the face library of the user, and if the comparison threshold is exceeded (for example, the comparison threshold is 0.8, and the same user), the comparison is successful and the face verification flow ends.

(3) The image detection method for a 1:N scene may be implemented as follows.

First, in an office building staff management business system, input a photo of each user who has the authority to enter the building (to form a comparison image of the face library in the passage scene of the building), and at this time, the image in the face library and the There is a one-to-one correspondence with the user of the building. The images in the face library are then bound and associated with the gate's face recognition device so that the gate's face recognition device can read the images in the face library. When the staff is about to pass through the gate, the face recognition device on the gate detects face information and enters the face snap shooting state (at this time, the shooting module in the face recognition device may always be activated, and the face is tracked, and the face frame always tracks the movement of the face).

Subsequently, mask wearing inspection (ie, face attribute detection) is performed at the time of face detection, and operation is performed according to the inspection result.

There are also two methods for extracting and searching facial feature values in a 1:N scene. Assuming that there are five face feature values B, C, D, and E that are not wearing a mask, feature extraction is performed on the face wearing a mask in a similar manner to obtain A1, B1, and C1 , D1 and E1, and searching for the feature value A1 when the mask is worn. A value A can be retrieved. The second is to extract the face feature values of the upper visible part from the mask, and the part of generating the feature values is the same as the method for the 1:1 scene, so please refer to the above description. After the feature values are generated, a facial feature search (1:N search) is performed on the generated feature values with the images of the face libraries of all users in all buildings, and there are feature values that exceed the comparison threshold. If so, consider the search successful. If the search is successful, the face recognition device opens and transmits a signal, the gate opens, and the face verification flow ends.

In the embodiment of the present application, the above method checks whether the user is wearing a mask and notifies a reminder before face recognition. It can be completed, greatly improving the efficiency of passage and reducing the risk of removing the mask.

The execution subject of the image detection method may be an image detection device. Digital Assistant: abbreviated as PDA), a portable device, a computing device, an in-vehicle device, a wearable device, or other terminal device or other processing device. In some possible embodiments, the image detection method may be implemented by a processor invoking computer readable instructions stored in memory.

FIG. 4 is a configuration diagram of an embodiment of an image detection device according to an embodiment of the present application. The image detection device 40 includes an image acquisition module 41 , a target detection module 42 and an operation execution module 43 . The image acquisition module 41 is configured to acquire a first image including the target to be detected, the target detection module 42 is configured to detect the first image and obtain a detection result of the first image, the detection result is a first image. The operation execution module 43 is configured to execute a predetermined operation that matches the detection result, including whether the target to be detected in one image is blocked by a preset object.

The above solving means detects whether or not the detection target is shielded by detecting a first image including the detection target, and then executes a predetermined operation that matches the detection result, so that the detection target is shielded. It is possible to determine whether or not the object has been detected, thereby executing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

In some embodiments, the detection result is obtained by target detection module 42 detecting the first image with a neural network.

The above solution detects by a pre-trained neural network, resulting in more accurate detection results and faster detection speed.

In some embodiments, the operation executing module 43 performing the predetermined operation that matches the detection result includes sending a first reminder if the target to be detected is not obstructed by a preset object. , the first reminder is for notifying the target to be detected to be shielded by a preset object.

The above solution notifies the first reminder when the target to be detected is not occluded, thereby timely notifying the situation that the target to be detected is not occluded by a preset object, thereby providing a notification the person can take corresponding measures in a timely manner.

In some embodiments, the detection result further includes whether the shielding method of the preset object of the target to be detected is a predetermined shielding method, and the operation execution module 43 performs a predetermined operation that matches the detection result. The executing step includes notifying a second reminder when the target to be detected is blocked by a preset object and the blocking method is not the predetermined blocking method, wherein the second reminder is performed by the preset object It is for notifying to adjust the shielding method.

The above solution sends a second reminder to timely adjust the shielding method of the detected target when the shielding method is not accurate.

In some embodiments, the step of the operation execution module 43 performing the predetermined operation that matches the detection result includes removing at least the undetected target from the first image when the target to be detected is blocked by a preset object. The method includes extracting a first feature of the occluded portion as a recognition-waiting feature of the detection target, and recognizing the detection target using the recognition-waiting feature to obtain a recognition result.

The above-described solution means realizes recognition based on local features of the detection target by extracting and recognizing the features of the unshielded portion when the detection target is blocked by a preset object. Since the local features are not occluded, they can represent the target to be detected, which guarantees recognition accuracy to some extent.

In some embodiments, the step of extracting from the first image by the operation execution module 43 at least the first feature of the unoccluded portion of the target to be detected as a pending recognition feature of the target to be detected includes: extracting a first feature of the unshielded portion of the target to obtain a second feature of the shielded portion of the detection target; and using the first feature and the second feature as recognition waiting features of the detection target. include.

In addition to adopting the features of the unshielded portion of the target to be detected, the above solution also adopts the features of the shielded portion, thereby making it possible to further diversify the features of the target to be detected.

In some embodiments, the step of obtaining the second feature of the occluded portion of the target to be detected by the operation execution module 43 includes extracting the feature of the occluded portion from the first image as the second feature; obtaining predetermined features of the occluded portion as second features, the predetermined features including features obtained based on at least one reference feature, each reference feature within a reference target where no occluded portion is present; is obtained by extracting from the region corresponding to the shielded portion of .

With regard to the method of specifying the features of the shielded portion, the above solution means may directly extract the features of the shielded portion. The recognition accuracy can be improved, or the predetermined feature can be obtained as the feature of the occluded part, this method does not need to perform feature extraction on the occluded part, which consumes processing resources. can be saved and processing efficiency can be improved.

In some embodiments, the step of recognizing the target to be detected using the recognition pending feature by the operation execution module 43 and obtaining the recognition result includes using the recognition pending feature and the pre-recognition target when one preset target is included. Obtaining a first degree of similarity between a set target and a pre-stored feature, and if the first degree of similarity satisfies a first predetermined condition, the recognition result includes successful identity authentication of the target to be detected. determining, and if a plurality of preset targets are included, respectively obtaining a second similarity between the feature waiting for recognition and the pre-stored feature of each preset target, and the recognition result is the target to be detected; as a preset target identity corresponding to a second similarity that satisfies a second predetermined condition.

The solution computes a first similarity measure of a particular preset target with pre-stored features, or computes a similarity measure of multiple preset targets with pre-stored features. Thus, the target to be detected can be compared with a specific pre-set target or with a pre-set target in some database according to the requirements of the actual scene.

In some embodiments, the first predetermined condition includes a first similarity greater than a first similarity threshold and the second predetermined condition includes a second similarity greater than a second similarity threshold.

According to the above solution, a more accurate recognition result can be obtained by setting the first similarity threshold in different scenes.

In some embodiments, the first similarity threshold when the feature to be recognized includes the second feature of the occluded portion of the target to be detected is the first similarity threshold when the feature to be recognized does not include the second feature. smaller, making the second similarity threshold when the features to be recognized include the second features less than the second similarity threshold when the features to be recognized do not include the second features.

In the above solving means, when the second feature is included in the recognition waiting features, the second feature may be different from the actual feature of the key point of the shielded portion of the target to be detected. can improve recognition accuracy.

In some embodiments, image sensing device 40 further includes a pre-stored module (not shown). Before the operation execution module 43 obtains the first similarity between the recognition pending feature and the pre-stored feature of the preset target, the pre-storage module registers the user's account in response to the account registration request. identifying a second image that satisfies a predetermined quality requirement from at least one frame of a second image obtained by photographing the user; extracting features of a predetermined part of the user from the identified second image; It is configured to associate the feature of the given body part with an account and store the feature of the given body part as a pre-stored feature of a preset goal.

The above solution first identifies the second image that satisfies the quality requirements and extracts the features of the predetermined part, thereby extracting more accurate features.

In some embodiments, if the target to be detected is occluded by a preset object, before extracting the first feature of at least the unoccluded portion of the target to be detected from the first image, the operation performing module 43 further: identifying a first image that satisfies a predetermined quality requirement as a first image for subsequent feature extraction from among first images of a plurality of frames including a target to be detected; and performing liveness detection on the first image for subsequent feature extraction, and if the detection target is a living body detection result, at least the unshielded portion of the detection target is detected from the first image. extracting one feature and deciding to perform subsequent steps.

The above solution means extracts more accurate features by first performing preprocessing before feature extraction, and recognizes the target to be detected only when the target to be detected is a living body, thereby increasing the safety of recognition. It can improve and, to some extent, prevent counterfeit attacks.

In some embodiments, the operation execution module 43 identifying, from among the first images of the plurality of frames containing the detected target, the first image that satisfies a predetermined quality requirement as the first image for subsequent feature extraction comprises: obtaining a corresponding quality score of the first image of each frame based on the quality factor of the first image of each frame, wherein the quality factor of the first image is the orientation information of the target to be detected with respect to the imaging device; including at least one of parameter information representing the size of a target to be detected in one image and luminance information of the first image; identifying a first image for feature extraction, wherein the selected first image has a higher quality score than the other first images.

The above solution identifies images whose quality scores meet the requirements and performs feature extraction to extract features that can further represent the target to be detected.

In some embodiments, the step of operation execution module 43 preprocessing the first image for subsequent feature extraction includes identifying targets of interest that meet predetermined extraction requirements if the first image includes multiple targets of interest. identifying a target region in the first image and removing image portions other than the target region in the first image; and/or detecting that the tilt angle of the detected target in the first image is greater than a predetermined angle and rotating the first image until the tilt angle of the target to be detected is less than a predetermined angle.

When a plurality of detection target targets exist in the first image, the above-described solution means identifies only the detection target that satisfies a predetermined extraction requirement, and discards the detection target that does not satisfy the requirement, thereby detecting the detection target that does not satisfy the requirement. The influence of the target on the recognition result is reduced, and if the inclination angle of the detection target in the first image is larger than a predetermined angle, the position of the detection target is corrected to reduce the influence of the inclination of the detection target. is reduced.

In some embodiments, the predetermined extraction requirement includes that the area of the corresponding region of the detectable target is greater than the area of the corresponding region of the other detectable target, and the other detectable target is a target other than the detectable target. including.

In the above solution, the larger the area of the target to be detected, the more accurate the extracted features. Therefore, by selecting the target to be detected having a larger area, a more accurate detection result can be obtained.

In some embodiments, the target to be detected includes a face and the preset object includes a mask.

The above solution is to determine whether the face is wearing a mask, perform the corresponding operation, for example, if the face is not wearing the mask or the mask is not applied correctly, the corresponding reminder is sent. , the user can timely adjust, perform recognition, etc. on the face if the mask is worn on the face.

The above solving means detects whether or not the detection target is shielded by detecting a first image including the detection target, and then executes a predetermined operation that matches the detection result, so that the detection target is shielded. It is possible to determine whether or not the object has been detected, thereby executing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

FIG. 5 is a configuration diagram of an example of an electronic device according to an example of the present application. Electronic device 50 includes memory 51 and processor 52, and processor 52 is used to execute program instructions stored in memory 51 to implement the steps in any of the image detection method embodiments described above. In one implementation scene, the electronic device 50 may include, but is not limited to, a microcomputer, a server, and the electronic device 50 may include, but is not limited to, portable devices such as laptops and tablets. .

Processor 52 is used to control itself and memory 51 to implement the steps in any of the image detection method embodiments described above. The processor 52 may be called a central processing unit (abbreviated as CPU). Processor 52 may be an integrated circuit chip having signal processing capabilities. The processor 52 is a common processor, a digital signal processor (abbreviated as DSP), an application specific integrated circuit (abbreviated as ASIC), and a field-programmable gate array (abbreviated as FPGA). or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. The common processor may be a microprocessor or any common processor or the like. Processor 52 may also be co-embodied by an integrated circuit chip.

The above solving means detects whether or not the detection target is shielded by detecting a first image including the detection target, and then executes a predetermined operation that matches the detection result, so that the detection target is shielded. It is possible to determine whether or not the object has been detected, thereby executing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

FIG. 6 is a configuration diagram of one embodiment of a computer-readable storage medium according to an embodiment of the present application. A computer-readable storage medium 60 stores processor-executable program instructions 61, which are used to implement steps in any of the image detection method embodiments described above.

The above solving means detects whether or not the detection target is shielded by detecting a first image including the detection target, and then executes a predetermined operation that matches the detection result, so that the detection target is shielded. It is possible to determine whether or not the object has been detected, thereby executing a subsequent predetermined operation that matches the detection result, and to realize flexible processing according to the occlusion state of the target to be detected in the image.

In some embodiments, embodiments of the present application provide a computer program product comprising computer readable code for causing a processor in an electronic device to perform the above method when executed in the electronic device.

In some embodiments, the functions or modules included in the apparatus provided by the embodiments of the present application are used to perform the methods described in the above method embodiments, and for the sake of simplicity, the specific For embodiments, reference may be made to the method example description above.

The above description of each embodiment tends to emphasize the differences between each embodiment, and for the sake of simplicity, identical or similar parts thereof may be referenced to each other.

It should be understood that in some of the embodiments provided by the present application, the disclosed method and apparatus may be embodied in other forms. For example, the embodiments of the apparatus described above are merely exemplary, for example, the division into modules or units is merely the division of logical functions, and may actually be divided in other forms when implemented, such as multiple units or components of may be combined or integrated into another system, or some features may be omitted or not performed. On the other hand, mutual couplings or direct couplings or communication connections shown or described may be indirect couplings or communication connections through some interface, device or unit, and may be electrical, mechanical or otherwise.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, may physically exist independently, or may be integrated into one unit by two or more. good too. The above integrated units may be implemented in the form of hardware or in the form of software functional units. When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored on a computer readable storage medium. Based on this view, the technical solutions of the embodiments of the present application may be implemented in the form of software products, or the part that contributes to the prior art, or all or part of the technical solutions may be implemented in the form of software products. The computer software product can be stored in a storage medium and executed in a computer device (which may be a personal computer, a server, a network device, etc.) or a processor to perform all of the methods of each embodiment according to the examples of the present application. Or execute some steps. The storage medium can store program codes such as USB memory, mobile hard disk, read-only memory (abbreviated as ROM), random access memory (abbreviated as RAM), magnetic disk or optical disk. Including a variety of media.

Embodiments of the present application provide an image detection method and associated apparatus, apparatus, storage medium, and computer program, the method comprising acquiring a first image including a target to be detected; detecting the first image; obtaining a detection result of the first image, wherein the detection result includes whether or not a target to be detected in the first image is occluded by a preset object; and performing a predetermined operation to. According to the image detection method provided by the embodiments of the present application, it is possible to determine whether the target to be detected is occluded, thereby performing a subsequent predetermined operation according to the detection result, and detecting in the image Realize flexible processing according to the occlusion state of the target.

Claims (20)

acquiring a first image including a target to be detected;
detecting the first image to obtain a detection result of the first image, wherein the detection result includes whether or not a target to be detected in the first image is blocked by a preset object; a step;
and performing a predetermined operation that matches the detection result. 2. The method of claim 1, wherein the detection result is obtained by detecting the first image with a neural network. The step of executing a predetermined operation that matches the detection result includes:
If the target to be detected is not occluded by the preset object, sending a first reminder, wherein the first reminder is to occlude the target to be detected by the preset object. for the purpose of notifying
3. A method according to claim 1 or 2. The detection result includes whether or not a method of shielding a preset object of the target to be detected is a predetermined shield method, and the step of performing a predetermined operation that matches the detection result includes:
sending a second reminder when the target to be detected is blocked by the preset object and the blocking method is not the predetermined blocking method, wherein the second reminder is the preset object; 4. The method according to any one of claims 1 to 3, wherein the method is for notifying to adjust the shielding scheme of the . The step of executing a predetermined operation that matches the detection result includes:
extracting at least a first feature of an unoccluded portion of the detection target from the first image as a recognition waiting feature of the detection target when the detection target is blocked by the preset object; ,
5. The method of any one of claims 1 to 4, comprising recognizing the detected target using the pending recognition feature to obtain a recognition result. The step of extracting from the first image at least a first feature of an unoccluded portion of the target to be detected as a recognition pending feature of the target to be detected,
extracting a first feature of an unoccluded portion of the target to be detected from the first image and obtaining a second feature of an occluded portion of the target to be detected;
6. The method of claim 5, comprising making the first and second features a pending recognition feature of the target to be detected. The step of obtaining a second feature of an occluded portion of the target to be detected includes:
extracting the feature of the shielded portion from the first image as the second feature; or
obtaining predetermined features of the shielded portion as the second features, the predetermined features including features obtained based on at least one reference feature, each of the reference features corresponding to the shielded portion; is extracted from a region corresponding to the occluded portion in a reference target in which the is absent. The step of recognizing the detection target using the recognition pending feature to obtain a recognition result includes:
if one preset target is included, obtain a first similarity between the recognition pending feature and a pre-stored feature of the preset target, wherein the first similarity satisfies a first predetermined condition; if so, determining that the recognition result includes successful identity authentication of the target to be detected;
when a plurality of the preset targets are included, obtaining a second similarity between the recognition-waiting feature and a pre-stored feature of each of the preset targets, and obtaining the recognition result as the target to be detected; as a preset target identity corresponding to a second similarity that satisfies a second predetermined condition. A method according to any one of the first predetermined condition includes that the first similarity is greater than a first similarity threshold;
9. The method of claim 8, wherein the second predetermined condition includes at least one of: the second similarity measure is greater than a second similarity threshold value. The first similarity threshold when the feature awaiting recognition includes the second feature of the shielded portion of the target to be detected is less than the first similarity threshold when the feature awaiting recognition does not include the second feature. to be smaller and
making the second similarity threshold when the feature awaiting recognition includes the second feature smaller than the second similarity threshold when the feature awaiting recognition does not include the second feature. 10. The method of claim 9, comprising at least one. Before obtaining a first similarity between the awaiting recognition features and pre-stored features of the preset target, further comprising:
registering an account for the user in response to the account registration request;
Identifying the second image that satisfies a predetermined quality requirement from at least one frame of the second image obtained by photographing the user, and extracting features of a predetermined part of the user from the identified second image. a step;
associating the predetermined body features with the account and storing the predetermined site features as pre-stored features of the preset goal. the method of. When the target to be detected is occluded by the preset object, before extracting a first feature of at least an unoccluded portion of the target to be detected from the first image, further:
identifying, from among the first images of the plurality of frames containing the detected target, the first image that satisfies predetermined quality requirements as the first image for subsequent feature extraction;
preprocessing the first image for subsequent feature extraction;
Living body detection is performed on the first image for subsequent feature extraction, and when the detection target is a living body detection result, at least the first unshielded portion of the detection target is detected from the first image. extracting features and deciding to perform subsequent steps;
12. A method according to any one of claims 5-11. identifying, from among the first images of the plurality of frames containing the detected target, the first image that satisfies a predetermined quality requirement as the first image for subsequent feature extraction;
obtaining a corresponding quality score for the first image of each frame based on a quality factor of the first image of each frame, wherein the quality factor of the first image is an orientation of the target to be detected with respect to an imaging device; at least one of: information, parameter information representing the size of a target to be detected in said first image, and luminance information of said first image;
identifying, based on the quality scores, the first images that meet predetermined quality requirements as the first images for subsequent feature extraction, wherein the quality scores of the selected first images are equal to the quality scores of the other first images; a higher than one image quality score. Preprocessing the first image for subsequent feature extraction comprises:
when the first image includes a plurality of the detection targets, specifying a target region in the first image of the detection target satisfying a predetermined extraction requirement, and extracting an image portion other than the target region in the first image; removing and/or
detecting that the tilt angle of the detection target in the first image is greater than a predetermined angle, and rotating the first image so that the tilt angle of the detection target is equal to or less than the predetermined angle. 14. The method according to Item 12 or 13. The predetermined extraction requirement includes that the area of the corresponding region of the detection target is larger than the area of the corresponding region of another detection target, and the other detection target includes a target other than the detection target. 15. The method of claim 14. 11. The method of claim 10, wherein the target to be detected comprises a face and the preset object comprises a mask. an image acquisition module configured to acquire a first image including a target to be detected;
A target detection module configured to detect the first image and obtain a detection result of the first image, wherein the detection result is an object whose target to be detected in the first image is a preset object. a target detection module including occluded or not;
an operation execution module configured to execute a predetermined operation matching the detection result. 17. An electronic device comprising a memory and a processor, said processor being used to execute program instructions stored in said memory to implement the method of any one of claims 1 to 16. 17. A computer readable storage medium having program instructions stored thereon, said program instructions causing a processor to perform the method of any one of claims 1 to 16. 17. A computer program product comprising computer readable code for, when run on an electronic device, causing a processor in said electronic device to perform the method of any one of claims 1 to 16.

Images