Human shape detection method and device based on artificial intelligence

Abstract

The invention proposes a human shape detection method and device based on artificial intelligence, and the method comprises the steps: obtaining a real-time monitoring image of a monitoring region, extracting a suspected human body contour image from the real-time monitoring image, and enabling the suspected human body contour image to be a contour image of a suspected human shape object matched with the human body shape on the real-time monitoring image, judging whether the human body posture on the suspected human body contour image is matched with a reference human body posture or not, and when the human body posture on the suspected human body contour image is matched with the reference human body posture, extracting a first local image corresponding to the suspected human body contour image from the real-time monitoring image, the first local image is input into a local human shape detection model to execute human body structure recognition to obtain human body structure data of the first local image, the local human shape detection model is an artificial intelligence model with low parameter quantity, and the stability of a monitoring system can be guaranteed while high-precision human shape detection is achieved.

Description

Technical Field

[0001] This invention relates to the field of security monitoring technology, and in particular to a human detection method and device based on artificial intelligence. Background Technology

[0002] Human detection technology identifies human silhouettes in surveillance footage and automatically tracks their posture and position. It uses cameras to capture images in real time, identifies human silhouettes, and then tracks their posture using features such as skeletal structure. This enables applications such as home monitoring, smart home integration, and burglar alarms. With the development of artificial intelligence, high-precision human detection models are now able to accurately identify human structures in complex environments, including densely populated areas.

[0003] Within the existing technological framework, higher detection accuracy means a larger number of parameters for the artificial intelligence model, and thus requires more computing power to operate. Considering factors such as cost, heat dissipation, and size, high-precision human detection models are typically not integrated into surveillance cameras for offline detection. Instead, the surveillance cameras connect to online models, utilizing cloud computing power to complete the human detection task. Because high-definition surveillance generates a massive amount of data, achieving timely human detection in densely monitored areas places extremely high demands on network bandwidth and stability. Network congestion or failures can lead to system malfunctions or even system paralysis. Summary of the Invention

[0004] Based on the above problems, this invention proposes an artificial intelligence-based human detection method and device, which can achieve high-precision human detection while ensuring the stability of the monitoring system.

[0005] In view of this, a first aspect of the present invention proposes an artificial intelligence-based human detection method, comprising: Acquire real-time monitoring images of the monitored area; Extract a suspected human body outline image from the real-time monitoring image. The suspected human body outline image is the outline image of a suspected human-shaped object that matches the shape of a human body in the real-time monitoring image. Determine whether the human pose on the suspected human contour image matches the reference human pose; When the human pose on the suspected human contour image matches the reference human pose, a first local image corresponding to the suspected human contour image is extracted from the real-time monitoring image. The first local image is input into a local human detection model to perform human structure recognition and obtain human structure data of the first local image. The local human detection model is an artificial intelligence model with a low number of parameters.

[0006] Optionally, after determining whether the similarity between the suspected human contour image and the standard human contour image is greater than or equal to a preset similarity threshold, the method further includes: When the human pose on the suspected human contour image does not match the reference human pose, a second local image corresponding to the suspected human contour image is extracted from the real-time monitoring image. The second partial image is transmitted to a cloud server to perform human structure recognition through a cloud-based human detection model in the cloud server. The cloud-based human detection model is an artificial intelligence model with a high number of parameters. Receive human anatomy data from the second local image returned by the cloud server.

[0007] Optionally, the step of extracting a suspected human body contour image from the real-time monitoring image specifically includes: A foreground image is generated from the real-time monitoring image, wherein the foreground image is obtained by performing background recognition and separation on the real-time monitoring image; Generate an edge image of the foreground image, the edge image being composed of the edge lines of the human body and / or objects in the foreground image; A third partial image with a human body shape is cropped from the edge image as the suspected human body contour image.

[0008] Optionally, the step of cropping a third partial image with a human body shape from the edge image as the suspected human body contour image specifically includes: Remove the outlines of foreground objects whose shape and / or size do not match the human body from the edge image; Identify image regions with partial or complete human body shapes on the edge image; A partial image of the image region is extracted as the suspected human body contour image.

[0009] Optionally, before the step of determining whether the human pose on the suspected human contour image matches the reference human pose, the method further includes: Identify the head region, limb region, and torso region of the human body on the suspected human body outline image; Based on the outlines of the head region, the limb region, and the torso region, the skeletal nodes of the human body are marked on the suspected human body outline image. The skeleton nodes are connected according to their positional relationship on the human body to obtain the human body posture skeleton graphic.

[0010] Optionally, the reference human posture includes one or more of standing, walking, and running postures. Before the step of determining whether the human posture on the suspected human contour image matches the reference human posture, the method further includes: The dimensions of the posture skeleton graphic are corrected; Determine the human height and orientation corresponding to the posture skeleton graphic; Load pre-configured reference human posture data corresponding to the human body height and orientation. The reference human posture data includes the standard node distance range and standard limb angle range under the reference human posture.

[0011] Optionally, the step of determining whether the human pose on the suspected human contour image matches the reference human pose specifically includes: Obtain the node distances and limb angles in the corrected pose skeleton graph; Match the node distances and limb angles in the posture skeleton graphic with the standard node distance ranges and standard limb angle ranges in the reference human posture data; When the node distance and limb angle in the posture skeleton graphic fall within the standard node distance range and standard limb angle range in the reference human posture data, it is determined that the human posture on the suspected human contour image matches the reference human posture.

[0012] Optionally, the output data of the local human detection model and the cloud-based human detection model include human recognition identifiers, as well as the node type code and node coordinates of each skeleton node in the pose skeleton. The node coordinates include the horizontal and vertical coordinates of each skeleton node in the relative coordinate system on the first local image or the second local image.

[0013] Optionally, after the steps of inputting the first partial image into a local human detection model to perform human structure recognition to obtain human structure data of the first partial image, or receiving human structure data of the second partial image returned by the cloud server, the method further includes: The human body structure data is analyzed to obtain the human figure identification identifier, node type code, and node coordinates; Based on the human figure recognition identifier, determine whether the suspected human figure outline image is a human figure outline image; When the suspected human body contour image is a human body contour image, the posture skeleton is drawn on the real-time monitoring image based on the node type encoding and the node coordinates.

[0014] A second aspect of the present invention provides an artificial intelligence-based human detection device, comprising a memory and a processor, wherein the processor executes a computer program stored in the memory to implement the artificial intelligence-based human detection method according to any one of the first aspects of the present invention.

[0015] This invention proposes an artificial intelligence-based human detection method and device. By acquiring real-time monitoring images of a monitored area, a suspected human contour image is extracted from the real-time monitoring image. The suspected human contour image is the contour image of a suspected human object in the real-time monitoring image that matches the shape of a human body. The method determines whether the human posture in the suspected human contour image matches a reference human posture. When the human posture in the suspected human contour image matches the reference human posture, a first partial image corresponding to the suspected human contour image is extracted from the real-time monitoring image. The first partial image is input into a local human detection model to perform human structure recognition and obtain human structure data of the first partial image. The local human detection model is an artificial intelligence model with low parameter count, which can achieve high-precision human detection while ensuring the stability of the monitoring system. Attached Figure Description

[0016] Figure 1 This is a schematic block diagram of a monitoring system provided in one embodiment of the present invention.

[0017] Figure 2 This is a flowchart of an artificial intelligence-based human detection method provided in one embodiment of the present invention. Detailed Implementation

[0018] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0020] In the description of this invention, the term "multiple" refers to two or more. Unless otherwise explicitly defined, the terms "upper," "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. The terms "connect," "install," "fix," etc., should be interpreted broadly. For example, "connect" can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "multiple" means two or more.

[0021] In the description of this specification, the terms "one embodiment," "some implementations," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0022] The following description, with reference to the accompanying drawings, describes an artificial intelligence-based human detection method and apparatus according to some embodiments of the present invention.

[0023] like Figure 1 As shown, the present invention proposes an artificial intelligence-based human detection method and device, which is applied to a monitoring system. The monitoring system includes an artificial intelligence-based human detection device and a cloud server that communicates with the human detection device through a network.

[0024] Specifically, the human detection device is a surveillance camera with human detection function. It includes an image sensing unit for acquiring surveillance images, a storage unit for storing artificial intelligence models and caching surveillance image data, a human detection unit for calling the artificial intelligence models stored in the storage unit to perform human detection, and a processing unit. The processing unit is communicatively connected to the image sensing unit, the storage unit, and the human detection unit. It is used to acquire surveillance image data from the image sensing unit and provide the surveillance image data to the human detection unit to perform human detection. Based on the human detection data output by the human detection unit, including human posture data, it can implement applications such as home care, intelligent linkage, and anti-theft alarm.

[0025] like Figure 2 As shown, the first aspect of the present invention proposes a human detection method based on artificial intelligence, comprising: Acquire real-time monitoring images of the monitored area; Extract a suspected human body outline image from the real-time monitoring image. The suspected human body outline image is the outline image of a suspected human-shaped object that matches the shape of a human body in the real-time monitoring image. Determine whether the human pose on the suspected human contour image matches the reference human pose; When the human pose on the suspected human contour image matches the reference human pose, a first local image corresponding to the suspected human contour image is extracted from the real-time monitoring image. The first local image is input into a local human detection model to perform human structure recognition and obtain human structure data of the first local image. The local human detection model is an artificial intelligence model with a low number of parameters.

[0026] In some implementation methods, a surveillance camera is fixedly installed in a specific location to monitor the area within its field of view and to provide real-time images of the monitored area to the backend for processing, storage, or presentation.

[0027] The suspected human body contour image referred to in this invention is a contour image that matches the shape of a human body. Specifically, it is a line image composed of the edge lines of a person or object with a similar human shape on the real-time monitoring image. Because the shape of the human body contour in the captured image can vary greatly depending on the human body's posture, the image will also show significant differences.

[0028] To extract human contour images from real-time surveillance images, it is usually necessary to perform grayscale processing and edge detection on the real-time surveillance images to generate a binarized line image composed of the edge lines of various objects in the image, including the human body and other objects. In complex environments, it is difficult to accurately extract human contour images from cluttered line images. In the technical solution of this invention, the real-time surveillance image is first converted into an edge image, which is a line image composed of the edge lines of various objects in the image, including the human body and other objects. A suspected human contour image is extracted from the edge image through shape matching, so as to complete the subsequent human body recognition through a human detection model.

[0029] The first partial image is a cropped image directly from the real-time monitoring image, corresponding to the position of the suspected human body outline image. That is, the human body outline in the suspected human body outline image is the outline image of the suspected human body in the first partial image.

[0030] The local human detection model is a low-precision artificial intelligence model with a small number of parameters, typically in the millions or lower. It is an AI model trained using limited sample data in a low-computing-power environment, often referred to as a small model. Its advantages include high recognition rates for human detection in simple environments, low storage space requirements, and low computing power needs. It can be directly deployed on surveillance cameras to perform real-time human detection without relying on backend devices such as cloud servers.

[0031] Optionally, after determining whether the similarity between the suspected human contour image and the standard human contour image is greater than or equal to a preset similarity threshold, the method further includes: When the human pose on the suspected human contour image does not match the reference human pose, a second local image corresponding to the suspected human contour image is extracted from the real-time monitoring image. The second partial image is transmitted to a cloud server to perform human structure recognition through a cloud-based human detection model in the cloud server. The cloud-based human detection model is an artificial intelligence model with a high number of parameters. Receive human anatomy data from the second local image returned by the cloud server.

[0032] Similarly, the second partial image is a partial image directly cropped from the real-time monitoring image, corresponding to the position of the suspected human body outline image. That is, the human body outline in the suspected human body outline image is the outline image of the suspected human body in the second partial image.

[0033] Because the amount of sample data used to train the local human detection model is relatively small, the low-parameter local human detection model can only recognize relatively simple human images. For special human poses, or in densely populated or complex environments, it has a low recognition rate. The cloud-based human detection model, on the other hand, is a high-precision artificial intelligence model with a large number of parameters, typically in the hundreds of millions. The cloud-based human detection model is trained using massive amounts of sample data and is often referred to as a large model. It can accurately recognize humans in various special poses and complex environments, achieving a higher recognition rate and wider applicability compared to the local human detection model.

[0034] Optionally, the step of extracting a suspected human body contour image from the real-time monitoring image specifically includes: A foreground image is generated from the real-time monitoring image, wherein the foreground image is obtained by performing background recognition and separation on the real-time monitoring image; Generate an edge image of the foreground image, the edge image being composed of the edge lines of the human body and / or objects in the foreground image; A third partial image with a human body shape is cropped from the edge image as the suspected human body contour image.

[0035] In some implementations, the background image is an environmental image of the monitored area. Since surveillance cameras typically monitor and capture images of the monitored area continuously around the clock, they can compare the content of surveillance footage captured at different times to identify unchanging environmental elements within the monitored area, thereby extracting the background image containing these environmental elements. The foreground image is an image obtained by removing the environmental elements from the background image from the real-time monitoring image. After removing the environmental elements, the edge image generated from the foreground image contains fewer interfering factors, resulting in a higher accuracy rate for extracting human contour images.

[0036] Preferably, in order to facilitate the subsequent localization of the human body, especially the skeletal nodes, in the image, the foreground image and the edge image generated based on the foreground image are both images with the same resolution and size as the real-time monitoring image.

[0037] Optionally, the step of cropping a third partial image with a human body shape from the edge image as the suspected human body contour image specifically includes: Remove the outlines of foreground objects whose shape and / or size do not match the human body from the edge image; Identify image regions with partial or complete human body shapes on the edge image; A partial image of the image region is extracted as the suspected human body contour image.

[0038] The edge image may contain outlines of foreground objects, such as cars and animals, that are significantly different from the human body in shape and size. By removing the outlines of these foreground objects from the edge image, the interference of non-human foreground objects on human contour recognition can be further reduced.

[0039] Furthermore, in the step of removing the outlines of foreground objects whose shape and / or size do not match the human body on the edge image, the distance between each foreground object and the surrounding static environmental elements can be calculated on the real-time monitoring image, thereby calculating the size of each foreground object based on the size of the environmental elements.

[0040] In the step of extracting a third partial image with a human body shape from the edge image as the suspected human body contour image, the suspected human body contour image is the original image extracted from the real-time monitoring image, without involving image rotation and scaling processing. Therefore, the pixel coordinates on the suspected human body contour image have a one-to-one correspondence with the pixel coordinates on the real-time monitoring image.

[0041] Optionally, before the step of determining whether the human pose on the suspected human contour image matches the reference human pose, the method further includes: Identify the head region, limb region, and torso region of the human body on the suspected human body outline image; Based on the outlines of the head region, the limb region, and the torso region, the skeletal nodes of the human body are marked on the suspected human body outline image. The skeleton nodes are connected according to their positional relationship on the human body to obtain the human body posture skeleton graphic.

[0042] In some implementations, shape matching can be used to identify the head, limbs, and torso regions of a human body in the suspected human silhouette image. This implementation is suitable for situations where the region in the real-time monitoring image corresponding to the suspected human silhouette image does not have a clear face, such as when the back or side of the human head is facing the monitoring camera, or when the face is blurred due to the human body being in rapid motion or being at a distance.

[0043] In other implementations, face recognition can be performed in the region of the real-time monitoring image corresponding to the suspected human body contour image to locate the head region. After determining the head region, the limb and torso regions on the suspected human body contour image can be quickly located based on adjacency relationships. This implementation is suitable for situations where the real-time monitoring image clearly captures a face in the region corresponding to the suspected human body contour image.

[0044] The skeletal nodes include, but are not limited to, head nodes, neck nodes, shoulder nodes, elbow nodes, wrist nodes, hip nodes, knee nodes, ankle nodes, and foot nodes. Each of the shoulder, elbow, wrist, hip, knee, ankle, and foot nodes includes a left-side node and a right-side node, such as a left shoulder node and a right shoulder node. In some embodiments, depending on the specific implementation scenario, several spinal nodes may also be configured in the torso region.

[0045] Optionally, the reference human posture includes one or more of standing, walking, and running postures. Before the step of determining whether the human posture on the suspected human contour image matches the reference human posture, the method further includes: The dimensions of the posture skeleton graphic are corrected; Determine the human height and orientation corresponding to the posture skeleton graphic; Load pre-configured reference human posture data corresponding to the human body height and orientation. The reference human posture data includes the standard node distance range and standard limb angle range under the reference human posture.

[0046] In the step of correcting the size of the posture skeleton graphic, since the same object will have different sizes on the real-time monitoring image at different relative distances from the monitoring camera, it is necessary to map the foreground object corresponding to the posture skeleton to a reference distance to correct its size for matching the reference human posture.

[0047] In the technical solution of this invention, the reference human posture is a posture in which the head, limbs, and torso can be clearly identified in the monitoring image, such as a standing posture, a walking posture, or a running posture. The standard node distance range is the normal distance range between each adjacent skeletal node under the reference human posture, such as the shoulder width range, arm length range, etc. The standard limb angle range is the angle range between the lines connecting adjacent skeletal nodes under the reference human posture, such as the angle range between the line connecting the head and neck node and the line connecting the shoulder nodes, the angle range between the line connecting the knee node and the ankle node and the line connecting the ankle node and the foot node, etc.

[0048] In some embodiments of the present invention, human posture data corresponding to different heights and orientations are pre-configured. This data can be configured in the form of a parameter relationship table or a relationship function, depending on the specific implementation requirements. The human orientation includes frontal orientation, side orientation, and back orientation.

[0049] Optionally, the step of determining whether the human pose on the suspected human contour image matches the reference human pose specifically includes: Obtain the node distances and limb angles in the corrected pose skeleton graph; Match the node distances and limb angles in the posture skeleton graphic with the standard node distance ranges and standard limb angle ranges in the reference human posture data; When the node distance and limb angle in the posture skeleton graphic fall within the standard node distance range and standard limb angle range in the reference human posture data, it is determined that the human posture on the suspected human contour image matches the reference human posture.

[0050] For non-reference human postures, such as static postures like sitting or lying down, or dynamic postures like sit-ups or push-ups, human contour recognition in monitoring images is difficult, resulting in low recognition rates using low-precision models. Therefore, in the technical solution of this invention, a local human detection model is used to identify local images corresponding to suspected human contour images that match the reference human posture, while local images corresponding to suspected human contour images that do not match the reference human posture are uploaded to a cloud-based human detection model on a cloud server for recognition.

[0051] Optionally, the output data of the local human detection model and the cloud-based human detection model include human recognition identifiers, as well as the node type code and node coordinates of each skeleton node in the pose skeleton. The node coordinates include the horizontal and vertical coordinates of each skeleton node in the relative coordinate system on the first local image or the second local image.

[0052] The human figure recognition identifier is a symbol used to mark whether the first or second partial image corresponding to the suspected human figure contour image is a human figure image. For example, identifier 0 can be used to represent a non-human figure image, and identifier 1 can be used to represent a human figure image. The node type encoding is a coding symbol used to represent skeleton nodes at different positions. To facilitate quantization, different numerical symbols can be used to represent skeleton nodes at different positions.

[0053] In some embodiments of the present invention, the relative coordinate system on the first local image or the second local image can be an orthogonal plane coordinate system with the head node as the origin, the width direction on the first local image or the second local image as the X-axis direction, and the height direction on the first local image or the second local image as the Y-axis direction, wherein the head node can be the geometric center point of the contour line of the head region.

[0054] Optionally, after the steps of inputting the first partial image into a local human detection model to perform human structure recognition to obtain human structure data of the first partial image, or receiving human structure data of the second partial image returned by the cloud server, the method further includes: The human body structure data is analyzed to obtain the human figure identification identifier, node type code, and node coordinates; Based on the human figure recognition identifier, determine whether the suspected human figure outline image is a human figure outline image; When the suspected human body contour image is a human body contour image, the posture skeleton is drawn on the real-time monitoring image based on the node type encoding and the node coordinates.

[0055] A second aspect of the present invention provides an artificial intelligence-based human detection device, comprising a memory and a processor, wherein the processor executes a computer program stored in the memory to implement the artificial intelligence-based human detection method according to any one of the first aspects of the present invention.

[0056] The artificial intelligence-based human detection device can be a surveillance camera, or a computer device connected to the surveillance camera via wired or wireless means, or other dedicated or general-purpose image processing devices connected to the surveillance camera via wired or wireless means.

[0057] 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 limitation, 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 said element.

[0058] As described above, these embodiments of the present invention do not exhaustively cover all details, nor do they limit the invention to the specific embodiments described. Clearly, many modifications and variations can be made based on the above description. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to effectively utilize the invention and its modifications. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for humanoid detection based on artificial intelligence, characterized in that, The method comprises the following steps: acquiring a real-time monitoring image of a monitoring area; extracting a suspected human contour image from the real-time monitoring image, the suspected human contour image being a contour image of a suspected human-shaped object on the real-time monitoring image that matches a human shape; judging whether a human posture on the suspected human contour image matches a reference human posture; when the human posture on the suspected human contour image matches the reference human posture, extracting a first local image corresponding to the suspected human contour image from the real-time monitoring image; inputting the first local image into a local human detection model to perform human structure recognition to obtain human structure data of the first local image, the local human detection model being an artificial intelligence model with low parameter quantity. 2.The artificial intelligence-based human shape detection method of claim 1, wherein, After the step of judging whether the similarity of the suspected human contour image and the standard human contour image is greater than or equal to a preset similarity threshold, the method further comprises the following steps: when the human posture on the suspected human contour image does not match the reference human posture, extracting a second local image corresponding to the suspected human contour image from the real-time monitoring image; transmitting the second local image to a cloud server to perform human structure recognition through a cloud human detection model in the cloud server, the cloud human detection model being an artificial intelligence model with high parameter quantity; receiving human structure data of the second local image returned by the cloud server. 3.The AI-based human shape detection method of claim 2, wherein, The step of extracting a suspected human contour image from the real-time monitoring image specifically comprises the following steps: generating a foreground image of the real-time monitoring image, the foreground image being obtained after performing background recognition and separation on the real-time monitoring image; generating an edge image of the foreground image, the edge image being composed of edge lines of a human body and / or an object on the foreground image; cutting a third local image with a human shape from the edge image as the suspected human contour image. 4.The AI-based human shape detection method of claim 3, wherein, The step of cutting a third local image with a human shape from the edge image as the suspected human contour image specifically comprises the following steps: removing contour lines of foreground objects on the edge image that do not match a human body in shape and / or size; identifying an image region with a partial or complete human shape on the edge image; cutting a local image of the image region as the suspected human contour image. 5.The AI-based human shape detection method of claim 4, wherein, Before the step of judging whether the human posture on the suspected human contour image matches the reference human posture, the method further comprises the following steps: identifying a head region, a limb region and a torso region of a human body on the suspected human contour image; annotating a skeleton node of a human body on the suspected human contour image according to contour lines of the head region, the limb region and the torso region; connecting the skeleton nodes according to their positional relationships on a human body to obtain a posture skeleton graph of the human body. 6.The AI-based human shape detection method of claim 5, wherein, The reference human posture comprises one or more of a standing posture, a walking posture and a running posture, and before the step of judging whether the human posture on the suspected human contour image matches the reference human posture, the method further comprises the following steps: correcting a size of the posture skeleton graph; determine a height and an orientation of the human body corresponding to the posture skeleton graph; load pre-configured reference human posture data corresponding to the height and the orientation of the human body, the reference human posture data including a standard node distance range and a standard limb angle range in the reference human posture. 7.The AI-based human shape detection method of claim 6, wherein, The step of determining whether the posture of the human body in the suspected human contour image matches the reference human posture specifically includes: obtaining the node distance and the limb angle in the corrected posture skeleton graph; matching the node distance and the limb angle in the posture skeleton graph with the standard node distance range and the standard limb angle range in the reference human posture data; when the node distance and the limb angle in the posture skeleton graph fall within the standard node distance range and the standard limb angle range in the reference human posture data, it is determined that the posture of the human body in the suspected human contour image matches the reference human posture. 8.The artificial intelligence-based human shape detection method of any one of claims 2-7, wherein, The output data of the local human detection model and the cloud human detection model include a human recognition identifier, a node type code of each node in the posture skeleton, and node coordinates, the node coordinates including horizontal and vertical coordinates of each node in a relative coordinate system of the first local image or the second local image. 9.The AI-based human shape detection method of claim 8, wherein, After the step of inputting the first local image into the local human detection model to perform human structure recognition to obtain human structure data of the first local image, or receiving human structure data of the second local image returned by the cloud server, the method further includes: analyzing the human structure data to obtain the human recognition identifier, the node type code, and the node coordinates; determining whether the suspected human contour image is a human contour image according to the human recognition identifier; when the suspected human contour image is a human contour image, drawing the posture skeleton on the real-time monitoring image based on the node type code and the node coordinates.

10. An artificial intelligence-based human shape detection device, characterized by comprising: The apparatus includes a memory and a processor, the processor executing a computer program stored in the memory to implement the human detection method based on artificial intelligence. The apparatus includes a memory and a processor, the processor executing a computer program stored in the memory to implement the human detection method based on artificial intelligence.

Images