Distance determination method and device and electronic equipment

Abstract

The invention discloses a distance determination method and device and electronic equipment, and relates to the technical field of image processing. The method comprises the following steps: acquiring a plurality of image frames and corresponding first central abscissas; performing feature extraction on the target image frame to obtain a detection frame height, a detection frame width and a second central abscissa of a head detection frame in the target image frame; inputting the height of the detection frame into a first distance measurement model, determining first distance information according to the height of the detection frame and a first calibration constant included in the first distance measurement model, inputting the width of the detection frame into a second distance measurement model, and determining second distance information according to the width of the detection frame and a second calibration constant included in the second distance measurement model; and determining the distance between the user corresponding to the target head detection frame and the camera according to the first central abscissa, the second central abscissa, the detection frame width, the first distance information and the second distance information. According to the embodiment of the invention, the accuracy of determining the distance between the person and the camera is improved.

Description

Technical Field

[0001] This application belongs to the field of image processing technology, and in particular relates to a distance determination method, apparatus and electronic device. Background Technology

[0002] Conference equipment typically comes equipped with high-definition cameras and integrates intelligent functions such as speaker tracking, automatic framing, and participant count. The quality of these functions depends heavily on the accurate perception of participants' spatial positions. Therefore, vision-based distance determination is a core perception capability of intelligent video conferencing equipment.

[0003] In existing technologies, when determining the distance to meeting participants, face detection is often performed on video frames to obtain face bounding boxes. The distance between a person and the camera is then determined based on a mapping relationship where the size of the face bounding box and the size of the object's image on the image sensor are inversely proportional to its actual physical distance to the camera. However, different faces have different actual sizes, and the size of the detected face bounding box for the same person can vary significantly depending on their pose. This results in low accuracy in determining the distance between a person and the camera using existing methods. Summary of the Invention

[0004] This application provides a distance determination method, apparatus, and electronic device to address the problem of low accuracy in determining the distance between a person and a camera using existing methods.

[0005] In a first aspect, embodiments of this application provide a distance determination method, which may include: Acquire conference video data, which includes multiple image frames and the first center x-coordinate of each image frame; Feature extraction is performed on the target image frame to obtain information on the head detection box corresponding to at least one user in the target image frame. The information on the head detection box includes the height of the detection box, the width of the detection box, and the second center x-coordinate of the head detection box; wherein, the target image frame is any image frame included in the conference video data; For a target head detection box in a target image frame, the height of the detection box is input into a first ranging model, and a first distance information is determined based on the detection box height and a first calibration constant included in the first ranging model. The width of the detection box is input into a second ranging model, and a second distance information is determined based on the detection box width and a second calibration constant included in the second ranging model. Here, the target head detection box is any head detection box in the target image frame. The distance between the user and the camera corresponding to the target head detection box of the target image frame is determined based on the first center x-coordinate, the second center x-coordinate, the detection box width, the first distance information, and the second distance information.

[0006] Secondly, embodiments of this application provide a distance determining device, which may include: The acquisition module is used to acquire conference video data, which includes multiple image frames and the first center x-coordinate of the image frames; The extraction module is used to extract features from the target image frame to obtain information on the head detection box corresponding to at least one user in the target image frame. The information on the head detection box includes the height of the detection box, the width of the detection box, and the second center horizontal coordinate of the head detection box. The target image frame is any image frame included in the conference video data. The determination module is used to, for the target head detection box of the target image frame, input the height of the detection box into the first ranging model, determine the first distance information based on the detection box height and the first calibration constant included in the first ranging model, input the width of the detection box into the second ranging model, and determine the second distance information based on the detection box width and the second calibration constant included in the second ranging model; wherein, the target head detection box is any head detection box in the target image frame; The determination module is also used to determine the distance between the user and the camera corresponding to the target head detection box of the target image frame based on the first center horizontal coordinate, the second center horizontal coordinate, the detection box width, the first distance information, and the second distance information.

[0007] Thirdly, embodiments of this application provide an electronic device, which includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the distance determination method as described in the first aspect.

[0008] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the distance determination method as described in the first aspect.

[0009] Fifthly, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the distance determination method as described in the first aspect.

[0010] This application provides a distance determination method, apparatus, and electronic device. The method first acquires conference video data, including multiple image frames and a first center abscissa of each image frame, providing basic input for subsequent feature extraction and distance calculation. Feature extraction is performed on the target image frame to obtain information about the head detection box corresponding to at least one user in the target image frame. The head detection box information includes the detection box height, detection box width, and a second center abscissa of the head detection box. The target image frame is any image frame included in the conference video data. The size of the head is less affected by posture, its size is more fixed, and its height and width are easier to obtain stably, providing more reliable detection data for the steps. For the target head detection box of the target image frame, the detection box height is input into a first ranging model. First distance information is determined based on the detection box height and a first calibration constant included in the first ranging model. The detection box width is input into a second ranging model. Second distance information is determined based on the detection box width and a second calibration constant included in the second ranging model. The target head detection box is any head detection box in the target image frame. By calculating the distance information using dual models, the basic head size and the deviation value caused by posture can be calculated in parallel. The distance between the user and the camera, corresponding to the target head detection box in the target image frame, is determined based on the first center x-coordinate, the second center x-coordinate, the detection box width, the first distance information, and the second distance information. By fitting the two distance information sets, the deviation caused by posture effects is compensated for on top of the base distance. Therefore, the embodiments of this application improve the accuracy of determining the distance between the person and the camera. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the structure of a distance determination system provided in an embodiment of this application; Figure 2 This is a flowchart illustrating a distance determination method provided in an embodiment of this application; Figure 3 This is a flowchart illustrating a method for determining a third head detection frame according to an embodiment of this application; Figure 4 This is a comparative schematic diagram of the data smoothing effect provided in the embodiments of this application; Figure 5 This is a flowchart illustrating a method for determining longitudinal distance provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of the preset distance fence provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of a distance determining device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0013] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0014] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0015] In intelligent video conferencing, enhancing the sense of presence and interaction efficiency are key objectives. Modern conferencing systems are typically equipped with high-definition cameras and integrate intelligent functions such as speaker tracking, automatic framing, and participant count. However, the quality of these functions highly depends on the accurate perception of participants' spatial positions. Currently, mainstream solutions rely heavily on traditional computer vision techniques for face or human detection. Existing methods are based on a fundamental inference from the pinhole camera model: with a fixed camera focal length, the size of an object's image on the image sensor is inversely proportional to its actual physical distance from the camera. That is, the closer an object is to the camera, the larger its size appears in the image; conversely, the farther away it is, the smaller it appears. Specifically, the steps of existing methods for detection may include: processing video frames using a trained face detection model to identify faces in the image and outputting the bounding box of each face; using the size of the face detection box as a distance proxy feature; and pre-setting a simple mapping relationship to estimate the distance. This relationship is typically established in two ways: a linear inverse proportional model, assuming that the distance d and the width w of the detection box satisfy d=k / w, where k is a constant calibrated experimentally. The lookup table method involves creating a lookup table from "face frame size" to "physical distance" based on pre-collected data. In practical applications, the system matches the detected face frame size against the lookup table, finds the closest size range, and outputs the corresponding distance value. The conferencing system uses this distance information to attempt some basic functions. For example, speaker location: assuming the participant closest to the camera (i.e., with the largest face frame) is the primary speaker, the camera will automatically turn and zoom in on that target. Simple region detection: setting a threshold for face frame size; when a face frame exceeds this threshold, the participant is considered "close" to the conferencing device.

[0016] However, this method assumes that the physical size of all faces is fixed. However, different faces have different actual sizes, and the detected face bounding box size of the same person varies drastically depending on their pose (frontal, profile). The width of a profile face bounding box is much smaller than that of a frontal face, causing the system to incorrectly judge that the participant is farther away, thus triggering camera malfunctions. Furthermore, when multiple people are in the same line of sight, people in the distance may be obscured by those in front, making it impossible to accurately detect their face bounding box size, resulting in missing or incorrect distance information. In addition, existing technologies typically only perform independent frame-by-frame detection without incorporating multi-target tracking technology. This makes it impossible to reliably bind participants to their IDs; when two face bounding boxes are similar in size, the system cannot distinguish whether they are the same person moving or two people swapping positions, thus failing to achieve intelligent interaction based on individual trajectories. Finally, because existing technologies can only provide very coarse and unstable distance estimates and cannot reliably distinguish between different individuals, this method struggles to support complex distance fencing applications (such as accurately determining whether someone enters / leaves a specific area), significantly limiting the potential for improving the intelligence of video conferencing systems.

[0017] To address the problems of existing technologies, embodiments of this application provide a distance determination method, apparatus, and electronic device. The method first acquires conference video data, which includes multiple image frames and a first center abscissa of each image frame, providing basic input for subsequent feature extraction and distance calculation. Feature extraction is performed on the target image frame to obtain information about the head detection box corresponding to at least one user in the target image frame. The head detection box information includes the detection box height, detection box width, and a second center abscissa of the head detection box. The target image frame is any image frame included in the conference video data. The size of the head is less affected by posture, its size is more fixed, and its height and width are easier to obtain stably, providing more reliable detection data for the steps. For the target head detection box of the target image frame, the detection box height is input into a first ranging model. First distance information is determined based on the detection box height and a first calibration constant included in the first ranging model. The detection box width is input into a second ranging model. Second distance information is determined based on the detection box width and a second calibration constant included in the second ranging model. The target head detection box is any head detection box in the target image frame. By calculating the distance information using dual models, the basic head size and the deviation value caused by posture can be calculated in parallel. The distance between the user and the camera, corresponding to the target head detection box in the target image frame, is determined based on the first center x-coordinate, the second center x-coordinate, the detection box width, the first distance information, and the second distance information. By fitting the two distance information sets, the deviation caused by posture effects is compensated for on top of the base distance. Therefore, the embodiments of this application improve the accuracy of determining the distance between the person and the camera.

[0018] The following first introduces a distance determination system 100 provided in this application embodiment, which is based on the Camera Multimedia eXtension (CAMX) framework and the Camera Hardware Interface (CHI) framework, and is used to realize multi-target head detection, tracking, ranging, and intelligent regional interaction under monocular vision. Figure 1 As shown, the system may include: a neural network inference engine 101, a multi-target tracking module 102, a data smoothing module 103, a monocular ranging core module 104, and a distance fence judgment module 105.

[0019] The neural network inference engine 101 includes the Snapdragon Neural Processing Engine (SNPE) and a head detection model. The Snapdragon SNPE is responsible for efficiently running deep learning models (such as the lightweight YOLOX). The head detection model, based on a deep learning model, identifies and locates the user's head from the original image, outputting key information such as the detection box (coordinates, width, and height) for each head, providing basic features for subsequent tracking and ranging. The multi-target tracking module 102, based on the ByteTrack algorithm, assigns a unique ID to each detected head, achieving cross-frame identity continuity and preventing the same user from being misidentified as different targets in different frames, thus solving tracking stability issues in scenarios such as target occlusion and brief disappearance. The data smoothing module 103 uses a Kalman filter to filter noise and smooth the head position and size data output by the multi-target tracking module 102, providing a more stable input for the monocular ranging core module 104. The monocular ranging core module 104 includes a height ranging model, a width ranging model, and an intelligent decision-maker. In frontal poses, it uses the height of the head detection box (a direct mapping of the head's actual physical height, without projection distortion) to calculate the longitudinal distance using a linear model. In side-face poses, it uses the width of the head detection box (although distorted by pose, the pattern can be fitted) to compensate for pose errors using a non-linear model to calculate the longitudinal distance. Based on a threshold for the detection box width, it selects or fuses the results of the two models to output the final longitudinal distance; simultaneously, it combines the horizontal coordinates of the image frame center and the head detection box center to calculate the lateral distance, achieving 3D spatial positioning. The distance fence judgment module 105 configures and executes region filtering logic according to application scenario requirements.

[0020] The distance determination method provided in the embodiments of this application is described below.

[0021] Figure 2 A flowchart illustrating a distance determination method according to an embodiment of this application is shown. Figure 2 As shown, the method may include the following steps: S201 to S204.

[0022] S201, Obtain conference video data, which includes multiple image frames and the first center x-coordinate of each image frame.

[0023] Among them, the conference video data is a set of video frame data continuously collected by a monocular camera; the image frame is the basic unit that makes up the video, which is a single static image; the first center horizontal coordinate is the pixel coordinate of the midpoint of the image frame in the horizontal direction.

[0024] In some embodiments, conference video data, i.e., camera video streams, are obtained through the Camera Multimedia eXtension (CAMX) pipeline and the Camera Hardware Interface (CHI) pipeline.

[0025] In some embodiments, the lower left corner of the image frame can be used as the origin of the pixel coordinate system, with the positive direction of the x-axis being horizontal to the right and the positive direction of the y-axis being vertical upward.

[0026] In one example, the resolution of the image frame is 1152×640 pixels, and the corresponding first center x-coordinate is 1152 / 2=576 pixels.

[0027] This application embodiment obtains conference video data to provide basic data for subsequent distance determination, and obtains the first center x-coordinate to provide a reference position for spatial positioning.

[0028] In some embodiments, when acquiring conference video data, a confirmation request is sent to the target user to confirm that the conference video data will be used to extract user features such as head and facial features for identification purposes, including head information and distance to the camera. Upon receiving the user's confirmation, the conference video data is acquired, thereby ensuring the security of the acquired conference video data. The target user can be a pre-defined user with permission to access the conference video data, or a user appearing in the conference video data; the settings can be configured according to actual needs.

[0029] S202, perform feature extraction on the target image frame to obtain information on the head detection box corresponding to at least one user in the target image frame. The information on the head detection box includes the height of the detection box, the width of the detection box, and the second center horizontal coordinate of the head detection box. The target image frame is any image frame included in the conference video data.

[0030] Wherein, the target image frame is any image frame selected from the conference video data; the head detection box is a rectangular box extracted from the image frame to select the user's head region; the second center x-coordinate is the horizontal coordinate of the geometric center of the head detection box in the pixel coordinate system of the image frame; the detection box height is the vertical pixel length of the head detection box; and the detection box width is the horizontal pixel length of the head detection box.

[0031] In some embodiments, when the information of the head detection box corresponding to at least one user in the target image frame is obtained, a unique and stable tracking ID is assigned to each user. Newly appearing users will be assigned a new ID, and the IDs of missing users will be temporarily retained to prevent them from reappearing after a brief occlusion.

[0032] In some embodiments, when performing feature extraction on a target image frame, the target image frame can be input into a pre-trained target detection model; the target detection model identifies all user heads in the target image frame, generates a rectangular detection box for each user's head, and outputs the height, width, and second center x-coordinate of each head detection box.

[0033] This application embodiment obtains head detection bounding box information by extracting features from the target image frame. Compared to the face, the head is less affected by pose, which can provide more reliable detection data for subsequent steps.

[0034] In some embodiments, when extracting features from a target image frame, multi-scale feature extraction can be performed on the target image frame, and head detection boxes can be output at each scale. The intersection-union ratio (IU) of the head detection boxes extracted at each scale can be calculated, and head detection boxes with IU greater than a set threshold can be identified as head detection boxes of the same user. The average information of head detection boxes of the same user at multiple scales can be taken to obtain the user's head detection box, or the confidence of the head detection boxes at each scale can be obtained, and the head detection box at the scale with the highest confidence can be selected as the head detection box corresponding to the user.

[0035] This application embodiment reduces the problem of missed detection of small heads for distant users and misjudgment of large head edges for near users by extracting head detection boxes of target image frames at multiple scales.

[0036] S203, for the target head detection box of the target image frame, the height of the detection box is input into the first ranging model, and the first distance information is determined according to the height of the detection box and the first calibration constant included in the first ranging model. The width of the detection box is input into the second ranging model, and the second distance information is determined according to the width of the detection box and the second calibration constant included in the second ranging model. Wherein, the target head detection box is any head detection box in the target image frame.

[0037] Wherein, the target head detection box is any head detection box in the target image frame whose distance needs to be calculated; the first ranging model is a ranging model that estimates the distance by the height of the detection box; the first calibration constant is a preset constant used to calculate the distance by the height of the detection box; the second ranging model is a ranging model that estimates the distance by the width of the detection box; the second calibration constant is a preset constant used to calculate the distance by the width of the detection box.

[0038] Because a single ranging model is easily affected by user posture limitations, the accuracy of the determined distance is relatively low. In this embodiment, the detection box height (representing head height) and detection box width (representing head width) are input into a first ranging model and a second ranging model, respectively. The two models are adapted to different head postures, and the resulting first and second distance information are also adapted to different head postures, providing multi-posture reference distance information for subsequent distance calculations.

[0039] In some embodiments, multiple first and second calibration constants can be preset for different scenarios (e.g., small conference rooms and large conference rooms), with the first and second calibration constants differing for each scenario. By adapting the first and second calibration constants to the specific scenario, the model can be made more closely resemble the actual usage environment, reducing ranging errors caused by scenario differences.

[0040] S204, determine the distance between the user and the camera corresponding to the target head detection box of the target image frame based on the first center horizontal coordinate, the second center horizontal coordinate, the detection box width, the first distance information and the second distance information.

[0041] The distance between the user and the camera is defined as a three-dimensional spatial distance, including both longitudinal and lateral distances. The longitudinal distance represents the distance between the user and the camera, while the lateral distance is the horizontal offset of the user relative to the center line of the camera's optical axis.

[0042] This application embodiment acquires conference video data to provide basic input for subsequent feature extraction and distance calculation. Feature extraction is performed on the target image frame to obtain head detection box information. The head size is less affected by posture, its size is more fixed, and its height and width are easier to obtain stably, providing more reliable detection data for the steps. By calculating distance information through a dual-model approach, the basic head size and the deviation value caused by posture can be calculated in parallel. Finally, by fitting the two distance information, the deviation value caused by posture is compensated for on the basis of the basic distance, thereby improving the accuracy of determining the distance between the person and the camera.

[0043] In some embodiments, feature extraction of a target image frame to obtain information about a head detection box corresponding to at least one user in the target image frame may include: Feature extraction is performed on the target image frame to obtain the information of the first head detection box of the target image frame; wherein, the first head detection box is the head bounding box obtained independently based on the target image frame; Obtain the information of the second head detection box of the image frame preceding the target image frame, and determine the information of the third head detection box of the user corresponding to the first head detection box in the information of the second head detection box; wherein, the image frame preceding the target image frame is the frame before the target image frame, for example, the target frame is frame 100 and the previous frame is frame 99; the second head detection box is the head detection box in the image frame preceding the target image frame; the third head detection box is the head detection box in the second head detection box that belongs to the same user as the first head detection box of the target image frame.

[0044] The target head detection box information corresponding to the user is determined based on the information of the first head detection box and the third head detection box; the target head detection box is the optimized head box obtained by fusing the target head detection box and the third head detection box of the previous frame.

[0045] Detecting a single image frame may result in jitter errors in the first head detection frame due to camera noise or changes in lighting. Directly using this information for distance determination can lead to significant errors in the distance results.

[0046] In this embodiment, user identity is associated with the first head detection box of the target image frame and the second head detection box of the previous image frame. The information of the first head detection box and the second head detection box is fused to obtain the third head detection box, which can smooth out jitter error and improve the accuracy of the information of the target synchronization detection box.

[0047] In some embodiments, determining the target head detection box information corresponding to the user based on the information of the first head detection box and the information of the third head detection box may include: averaging the information of the first head detection box and the third head detection box to obtain the target avatar detection box information.

[0048] In some embodiments, determining the target head detection box information corresponding to the user based on the information of the first head detection box and the information of the third head detection box may include: obtaining the confidence scores of the first head detection box and the third head detection box; and using the confidence scores as weight parameters to perform weighted fusion of the first head detection box and the third head detection box to obtain the target avatar detection box information.

[0049] In some embodiments, such as Figure 3 As shown, determining the information of the third head detection box of the user corresponding to the first head detection box in the information of the second head detection box may include: S301 and S302.

[0050] S301, determine the intersection-over-interference ratio and motion information of the first head detection frame and the second head detection frame based on the information of the first head detection frame and the second head detection frame.

[0051] Intersection over Union (IoU) is an indicator used to measure the degree of overlap between two head detection boxes; a larger value indicates a higher degree of overlap. Motion information describes the motion state of the head detection boxes between frames and may include information such as motion direction and motion speed.

[0052] In some embodiments, determining the intersection-over-union ratio (IoU) of the first head detection frame and the second head detection frame based on the information of the first head detection frame and the second head detection frame may include: Obtain the coordinates of the first head detection box and the second head detection box, and calculate the area of ​​the intersection region and the area of ​​the union region of the first head detection box and the second head detection box based on the coordinates of the first head detection box and the second head detection box. The ratio of the intersection region area to the union region area is determined as the intersection-union ratio of the first head detection box and the second head detection box.

[0053] In some embodiments, determining the motion information of the first head detection frame and the second head detection frame based on the information of the first head detection frame and the second head detection frame may include: Obtain the center coordinates and frame interval of the first head detection box and the second head detection box, and calculate the displacement of the first head detection box in the horizontal and vertical directions based on the center coordinates of the first head detection box and the second head detection box. The motion direction is calculated based on the displacement of the first head detection box in the horizontal and vertical directions, and the motion speed is calculated based on the displacement in the horizontal and vertical directions and the frame interval between the first head detection box and the second head detection box. The motion direction and speed of the second head detection box are obtained. The motion direction and speed of the second head detection box are pre-calculated and can be obtained directly, and the calculation process is the same as that of the motion direction and speed of the first head detection box.

[0054] S302, if the cross-union ratio exceeds a set threshold and the motion information is the same, determine the information of the second head detection box as the information of the third head detection box of the user corresponding to the first head detection box.

[0055] In some embodiments, the threshold for the intersection-over-union ratio (IoU) can be dynamically set based on the user density in the scene. For example, in high-density scenes, where the distance between users is small, the IoU threshold is lowered to allow users to have a higher degree of overlap; in low-density scenes, where the distance between users is large, the IoU threshold is raised to prevent the problem of false association of user head detection boxes.

[0056] In some embodiments, identical motion information can mean that the first head detection frame and the second head detection frame move in the same direction and their speeds differ within a set range.

[0057] This application's embodiments ensure that the positions of the first and second head detection boxes are close by determining the intersection-union ratio (IU), and ensure the continuity of movement between the first and second head detection boxes by determining motion information. Using dual-dimensional features reduces the risk of misassociating adjacent users or users with similar motion states by mitigating the tendency of a single feature.

[0058] In some embodiments, determining the target head detection box information corresponding to the user based on the information of the first head detection box and the information of the third head detection box may include: The information of the third head detection box is input into the head detection box prediction model. Based on the information of the third head detection box and the model parameters of the head detection box prediction model, the information of the predicted head detection box of the user in the image frame is determined. The information of the predicted head detection box is the predicted head detection box corresponding to the first head detection box in the target image frame output by the head detection box prediction model. The target head detection box information corresponding to the user is determined based on the information of the first head detection box, the information of the predicted head detection box, and the preset weight parameters.

[0059] In some embodiments, a short-time trajectory queue can be set for the ID of the user corresponding to each first head detection box, which contains head detection boxes of multiple consecutive image frames corresponding to the user ID. When making predictions, the head detection boxes of image frames in the short-time trajectory queue that precede the target image frame can be input into the head detection box prediction model.

[0060] In some embodiments, the head detection box prediction model can be a Kalman filter, and the model parameters of the head detection box prediction model can be a state transition matrix and process noise covariance.

[0061] In one example, such as Figure 4 As shown, the position of the avatar frame obtained by the existing technology shows a large jump, while the data smoothing effect of the avatar frame position predicted by Kalman filtering is smoother and more stable compared to the existing technology.

[0062] This application's embodiments predict the theoretical position and size of the head detection box of the target image frame based on the historical motion pattern information of the third head detection box of the previous image frame, and combine the predicted head detection box with the first head detection box. By correcting the information of the first head detection box with the information of the predicted head detection box, the information of the first head detection box of the acquired target image frame is prevented from deviating from the actual value due to detection errors.

[0063] In some embodiments, the second calibration constant includes a fitting coefficient, an exponential coefficient, and an offset coefficient. Determining the first distance information based on the detection frame height and the first calibration constant included in the first ranging model may include: The first distance information is determined based on the detection frame height and the first calibration constant included in the first ranging model. The first distance information is the ratio of the first calibration constant to the detection frame height. The first calibration constant is a preset constant that fuses the average physical height of the human head and the camera focal length, and is obtained through calibration in a real scene. The first ranging model is a linear model. The first distance information is an estimated value of the longitudinal distance calculated by the first ranging model. Determining the second distance information based on the detection frame width and the second calibration constant included in the second ranging model can include: Calculate the exponent of the detection box width to obtain the result of the exponentiation; The ratio of the fitting coefficient to the result of the exponentiation is added to the offset coefficient to obtain the second distance information; wherein, the second ranging model is a nonlinear model; the second distance information is the longitudinal distance estimate calculated by the second ranging model, which is used to compensate for the error caused by head turning.

[0064] In this embodiment, the detection box height is input into a linear model. Since the physical height of the head is relatively constant and unaffected by left or right head turning, the distance is calculated using the inverse relationship between head height and a calibration constant. This calculation is highly accurate and requires no complex computation. The detection box width is input into a nonlinear model. When the head turns, the nonlinear model can fit the nonlinear relationship between the head turning and the resulting change in detection box width, effectively compensating for the distortion of the detection box width caused by changes in user posture.

[0065] In some embodiments, the first ranging model is a ranging model based on the principle of a pinhole camera, and the formula for determining the first distance information based on the detection frame height and the first calibration constant included in the first ranging model can be: d_h=(H_real*f) / h_pixel Where d_h is the first distance information, H_real is the actual physical average height of the human head, f is the camera focal length, and h_pixel is the detection box height.

[0066] In some embodiments, the second ranging model is a power function fitting model, and the formula for determining the second distance information based on the detection box width and the second calibration constant included in the second ranging model can be: d_w = a / (w_pixel^n) + b Where d_w represents the second distance information, a is the fitting coefficient, w_pixel is the detection box width, n is the exponential coefficient, and b is the offset coefficient. a, n, and b are parameters obtained by fitting a large amount of actual data, where a=250, n=1.38, and b=0.85.

[0067] In some embodiments, determining the distance between the user and the camera corresponding to the target head detection box of the target image frame based on the first center x-coordinate, the second center x-coordinate, the detection box width, the first distance information, and the second distance information may include: The longitudinal distance between the user and the camera corresponding to the target head detection frame is determined based on the detection frame width, the first distance information, and the second distance information; where the longitudinal distance is the distance between the user and the camera along the optical axis of the camera. The lateral distance between the user and the camera corresponding to the target head detection box is determined based on the vertical distance, the first center x-coordinate, and the second center x-coordinate; where the lateral distance is the horizontal offset distance of the user relative to the center line of the camera's optical axis.

[0068] This application embodiment calculates the longitudinal and lateral distances between the user and the camera. The longitudinal distance determines the distance between the user and the camera, and the lateral distance supplements the left and right distances between the user and the camera, thereby achieving accurate three-dimensional positioning of the user in three-dimensional space.

[0069] In some embodiments, such as Figure 5 As shown, determining the longitudinal distance between the user and the camera corresponding to the target head detection frame based on the detection frame width, the first distance information, and the second distance information can include: S501 to S503.

[0070] S501, when the detection frame width is not less than the first set threshold, determine the longitudinal distance between the user and the camera corresponding to the target head detection frame based on the first center horizontal coordinate, the second center horizontal coordinate, and the second distance information.

[0071] The first set threshold is the critical width value for determining whether the head is close to a frontal facial posture.

[0072] In one example, when the detection box width w_pixel is not less than the first set threshold W_threshold_high (e.g., the detection box width is ≥70 pixels), it is considered that the head is facing the camera at a near front / back position. At this time, the head width information is reliable, and the second distance information d_w calculated by the second ranging model is preferred as the longitudinal distance d_final.

[0073] When the width of the detection frame is not less than the first set threshold, it indicates that when the current user is in the frontal face posture, the width of the detection frame in the frontal face is a direct mapping of the true width of the head. Therefore, in the embodiments of the present application, the longitudinal distance is determined by the first distance information determined by the height of the detection frame to adapt to the frontal face posture of the user.

[0074] S502. When the width of the detection frame does not exceed the second set threshold, determine the longitudinal distance between the user corresponding to the target head detection frame and the camera according to the first central abscissa, the second central abscissa, and the first distance information.

[0075] Among them, the second set threshold is a critical width value for determining that the head is close to the side face posture, and the second set threshold is less than the first set threshold.

[0076] In an example, when the width w_pixel of the detection frame does not exceed the second set threshold W_threshold_low (for example, the width of the detection frame ≤ 50 pixels), it is considered that the human head is significantly lateral, and the width information is unreliable. The first distance information d_h of the height model is preferentially used as the longitudinal distance d_final.

[0077] When the width of the detection frame does not exceed the second set threshold, it indicates that the current user is in the side face posture, and the head height information is less affected by the posture. Therefore, in the embodiments of the present application, the longitudinal distance is determined by the first distance information determined by the height of the detection frame to adapt to the side face posture of the user.

[0078] S503. When the width of the detection frame is greater than the first set threshold and less than the second set threshold, determine the longitudinal distance between the user corresponding to the target head detection frame and the camera according to the first central abscissa, the second central abscissa, the first distance information, the second distance information, and the preset weight parameter.

[0079] In an example, when the width is in the middle range (for example, 50 < w_pixel < 70), a weighted fusion strategy is adopted, such as d_final = α * d_w + (1 - α) * d_h, where the weight α can vary linearly or non-linearly according to the width of the detection frame.

[0080] When the width of the detection frame is greater than the first set threshold and less than the second set threshold, it indicates that the user is in the semi-side face posture. In the embodiments of the present application, the first distance information and the second distance information are used simultaneously. The first distance information retains the basic accuracy in the frontal face, and the second distance information compensates for the posture distortion. The finally obtained longitudinal distance takes into account both stability and anti-pose interference ability.

[0081] This application embodiment compares the detection box width with a set threshold to classify user poses into frontal, side, and half-side poses, and selects different adaptive calculation methods according to different user poses, thereby improving the accuracy of determining the longitudinal distance for different poses.

[0082] In some embodiments, the formula for calculating the lateral distance can be: Lateral=(x_center-x_image_center)*d_final / f Where Lateral is the horizontal distance, x_center is the x-coordinate of the center of the target head detection box, x_image_center is the x-coordinate of the image center, d_final is the vertical distance, and f is the equivalent pixel focal length of the camera.

[0083] In some embodiments, the method may further include: The system acquires the area parameters of a preset distance fence, which include a first vertical distance, a second vertical distance, a first horizontal distance, and a second horizontal distance; wherein the first vertical distance is less than the second vertical distance, and the first horizontal distance is less than the second horizontal distance; the preset distance fence is a virtual three-dimensional area defined within the camera's field of view, used to identify whether the user has entered the target area; If the vertical distance between the user and the camera corresponding to the head detection frame is between the first vertical distance and the second vertical distance, and the horizontal distance is between the first horizontal distance and the second horizontal distance, it is determined that the user is within the preset distance fence, and the operation corresponding to the preset distance fence is executed.

[0084] This application embodiment achieves intelligent triggering of intelligent operations by setting a preset distance fence and binding the distance fence with different operations, which can meet the actual needs of meeting scenarios.

[0085] In one example, such as Figure 6 As shown, the larger square area represents the preset distance fence, and the smaller square area represents the conference table. The horizontal line segment closer to the monocular camera represents the first vertical distance, and the horizontal line segment farther from the monocular camera represents the second vertical distance. The vertical line segment to the left of the monocular camera represents the first horizontal distance, and the vertical line segment to the right of the monocular camera represents the second horizontal distance. Four users A, B, C, and D were identified within the preset distance fence, with distances to the monocular camera of 1.2m, 1.8m, 2.8m, and 3.5m, respectively. Users E, F, G, and H, located at the boundary and outside the preset distance fence, were not identified.

[0086] In some embodiments, multiple distance fences can be set in the same scene, and each distance fence is bound to a different operation.

[0087] In the embodiments of the present application, by setting multiple fences bound to different operations, multi-region differential intelligent interaction can be achieved.

[0088] In one example, taking a video conferencing machine equipped with a Qualcomm Snapdragon QCM6490 chip and an IMX678 camera as an example, its specific implementation parameters are as follows: Detection model: Use the lightweight YOLOX-S model, adjust the input resolution to 1152x640, the model file is yolox_s_1152_640_quantize.dlc, and run it on the DSP using the SNPE v2.33 version.

[0089] Tracker parameters: ByteTrack tracking algorithm, and the number of frames for retaining the track is 30 frames.

[0090] Kalman filter parameters: The process noise covariance matrix Q is set to diag(1,1,1,1,0.01,0.01), and the measurement noise covariance matrix R is set to diag(1,1).

[0091] Range measurement model parameters: Height model constant C: Measured by calibration to be 113.5. Width model parameters: a = 250, n = 1.38, b = 0.85. Intelligent decision threshold: When the pixel width w of the head box ≥ 65, the width model result is adopted; when w ≤ 55, the height model result is adopted; when 55 < w < 65, the weighted average d_final = 0.7 * d_w + 0.3 * d_h is adopted.

[0092] Camera internal parameters: Assume the camera focal length f = 1000 pixels.

[0093] Distance fence example: The preset main speaker area fence has the first longitudinal distance Near = 1.5m, the second longitudinal distance Far = 4.0m, the first lateral distance Left = -2.0m, and the second lateral distance Right = 2.0m.

[0094] In one example, the overall process of intelligently performing operations during a video conference by applying the distance determination method of the embodiments of the present application may include: (1) The conferencing machine starts, the system is initialized, the model is loaded, and a "main speaker area" fence (such as an area 1.5 meters to 3 meters from the camera and -2 meters to 2 meters in width) is preset.

[0095] (2) The camera captures the conference room image, the system detects multiple participants, and performs tracking and range measurement.

[0096] (3) When a participant walks into the "main speaker area", the system determines its entry event based on its ID and coordinates.

[0097] (4) The system automatically controls the camera to focus on and zoom in on the participant corresponding to the ID, and increases the audio gain to achieve the "speaker automatic tracking" function.

[0098] (5) When the participant leaves the area, the system triggers a departure event, and the camera can switch to a wide-angle view or track the next person to enter the area.

[0099] In summary, the embodiments of this application, by outputting stable and accurate absolute physical coordinates for each participant, enable the conferencing system to define and identify complex virtual areas. For example, a "presentation area" can be set up, triggering an automatic close-up only when a participant enters this specific physical area; or a "proximity wake-up area" can be set up to achieve contactless screen wake-up, supporting complex distance fence applications. Furthermore, because each target is assigned a persistent and unique tracking ID, the system can accurately track the movement trajectory of individual participants. This enables advanced functions, such as accurately counting the number of people entering / leaving the meeting room (avoiding double counting), identifying who is walking towards the whiteboard, and audio beamforming based on individual location, greatly improving the intelligence level and interactive experience of video conferencing, thereby achieving precise individualized behavioral analysis.

[0100] Furthermore, this application's embodiment is based on the monocular camera already standard on video conferencing machines, eliminating the need to integrate expensive dedicated hardware such as depth sensors, binocular cameras, or LiDAR. This minimizes material costs and allows for the deployment of advanced intelligent features in mainstream products. Moreover, the entire technical solution is deeply optimized and embedded in the Qualcomm platform's CAMX / CHI underlying framework, fully utilizing its DSP / NPU or GPU for neural network inference and parallel computing. This ensures real-time frame rates (e.g., 30fps) even on resource-constrained mobile computing platforms, meeting the real-time requirements of video conferencing.

[0101] Figure 7 This application illustrates a distance determination device 700 according to an embodiment of the present application. The device may include: The acquisition module 701 is used to acquire conference video data, which includes multiple image frames and the first center x-coordinate of the image frames. The extraction module 702 is used to extract features from the target image frame to obtain information of the head detection box corresponding to at least one user in the target image frame. The information of the head detection box includes the height of the detection box, the width of the detection box, and the second center horizontal coordinate of the head detection box. The target image frame is any image frame included in the conference video data. The determination module 703 is used to, for the target head detection box of the target image frame, input the height of the detection box into the first ranging model, determine the first distance information based on the detection box height and the first calibration constant included in the first ranging model, input the width of the detection box into the second ranging model, and determine the second distance information based on the detection box width and the second calibration constant included in the second ranging model; wherein, the target head detection box is any head detection box in the target image frame; The determination module 703 is also used to determine the distance between the user and the camera corresponding to the target head detection box of the target image frame based on the first center horizontal coordinate, the second center horizontal coordinate, the detection box width, the first distance information, and the second distance information.

[0102] In some embodiments, the extraction module 702 is further configured to perform feature extraction on the target image frame to obtain information of the first head detection box of the target image frame; The determining module 703 is further configured to obtain information of the second head detection box of the image frame preceding the target image frame, and determine information of the third head detection box of the user corresponding to the first head detection box in the information of the second head detection box. The determination module 703 is also used to determine the information of the target head detection box corresponding to the user based on the information of the first head detection box and the information of the third head detection box.

[0103] In some embodiments, the determining module 703 is further configured to determine the intersection-over-interference ratio and motion information of the first head detection frame and the second head detection frame based on the information of the first head detection frame and the second head detection frame. The determination module 703 is further configured to determine the information of the second head detection box as the information of the third head detection box of the user corresponding to the first head detection box when the cross-union ratio exceeds a set threshold and the motion information is the same.

[0104] In some embodiments, the determining module 703 is further configured to input the information of the third head detection box into the head detection box prediction model, and determine the information of the predicted head detection box of the user in the image frame based on the information of the third head detection box and the model parameters of the head detection box prediction model. The determination module 703 is also used to determine the information of the target head detection box corresponding to the user based on the information of the first head detection box, the information of the predicted head detection box, and the preset weight parameters.

[0105] In some embodiments, the distance determining device 700 may further include: The determining module 703 is further configured to determine first distance information based on the detection frame height and the first calibration constant included in the first ranging model, wherein the first distance information is the ratio of the first calibration constant to the detection frame height; The calculation module is used to calculate the exponent of the detection box width to obtain the exponentiation result; The calculation module is also used to add the ratio of the fitting coefficient to the result of the exponentiation operation to the offset coefficient to obtain the second distance information.

[0106] In some embodiments, the determining module 703 is further configured to determine the longitudinal distance between the user and the camera corresponding to the target head detection frame based on the detection frame width, the first distance information, and the second distance information; The determination module 703 is also used to determine the lateral distance between the user and the camera corresponding to the target head detection box based on the vertical distance, the first center horizontal coordinate, and the second center horizontal coordinate.

[0107] In some embodiments, the determining module 703 is further configured to determine the longitudinal distance between the user and the camera corresponding to the target head detection frame based on the first center horizontal coordinate, the second center horizontal coordinate, and the second distance information, provided that the detection frame width is not less than the first set threshold. The determining module 703 is also used to determine the longitudinal distance between the user and the camera corresponding to the target head detection frame based on the first center horizontal coordinate, the second center horizontal coordinate and the first distance information, provided that the detection frame width does not exceed the second set threshold. When the width of the detection box is greater than the first set threshold and less than the second set threshold, the vertical distance between the user and the camera corresponding to the target head detection box is determined based on the first center horizontal coordinate, the second center horizontal coordinate, the first distance information, the second distance information, and the preset weight parameters.

[0108] In some embodiments, the distance determining device 700 may further include: The acquisition module 701 is also used to acquire the area parameters of the preset distance fence, including a first longitudinal distance, a second longitudinal distance, a first lateral distance, and a second lateral distance; wherein, the first longitudinal distance is less than the second longitudinal distance, and the first lateral distance is less than the second lateral distance; The execution module is used to determine that the user is within a preset distance fence when the vertical distance between the user and the camera corresponding to the head detection frame is between the first vertical distance and the second vertical distance, and the horizontal distance is between the first horizontal distance and the second horizontal distance, and then executes the operation corresponding to the preset distance fence.

[0109] Figure 7 The various modules in the device shown can achieve Figure 2 The various steps involved, and the corresponding technical effects achieved, will not be elaborated upon here for the sake of brevity.

[0110] Figure 8 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0111] Electronic devices may include a processor 801 and a memory 802 storing computer program instructions.

[0112] Specifically, the processor 801 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0113] Memory 802 may include mass storage for data or instructions. For example, and not limitingly, memory 802 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 802 may include removable or non-removable (or fixed) media, or memory 802 may be non-volatile solid-state memory. Memory 802 may be internal or external to the integrated gateway disaster recovery device.

[0114] In one example, memory 802 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described in reference to the method determined according to the distance of this disclosure.

[0115] The processor 801 reads and executes computer program instructions stored in the memory 802 to achieve... Figure 1 The distance determination method in the illustrated embodiment.

[0116] In one example, the electronic device may also include a communication interface 803 and a bus 805. For example, Figure 8 As shown, the processor 801, memory 802, and communication interface 803 are connected through bus 805 and complete communication with each other.

[0117] The communication interface 803 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0118] Bus 805 includes hardware, software, or both, that couples components of an end device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 805 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0119] Furthermore, in conjunction with the distance determination methods in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the distance determination methods in the above embodiments.

[0120] This application also provides a computer program product, including a computer program, which, when executed, implements any of the distance determination methods described in the above embodiments.

[0121] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0122] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or text segments used to perform the required tasks. Programs or text segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (ROM), flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Text segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0123] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0124] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0125] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A distance determination method, characterized in that, include: Acquire conference video data, which includes multiple image frames and the first center x-coordinate of the image frames; Feature extraction is performed on the target image frame to obtain information on the head detection box corresponding to at least one user in the target image frame. The information on the head detection box includes the detection box height, detection box width, and the second center horizontal coordinate of the head detection box. The target image frame is any image frame included in the conference video data. For the target head detection box of the target image frame, the height of the detection box is input into the first ranging model, and the first distance information is determined according to the height of the detection box and the first calibration constant included in the first ranging model. The width of the detection box is input into the second ranging model, and the second distance information is determined according to the width of the detection box and the second calibration constant included in the second ranging model. Wherein, the target head detection box is any head detection box in the target image frame. The distance between the user and the camera corresponding to the target head detection box of the target image frame is determined based on the first center horizontal coordinate, the second center horizontal coordinate, the detection box width, the first distance information, and the second distance information.

2. The distance determination method according to claim 1, characterized in that, The step of extracting features from the target image frame to obtain information about the head detection box corresponding to at least one user in the target image frame includes: Feature extraction is performed on the target image frame to obtain information of the first head detection box of the target image frame; Obtain the information of the second head detection box of the image frame preceding the target image frame, and determine the information of the third head detection box of the user corresponding to the first head detection box in the information of the second head detection box; The target head detection box information corresponding to the user is determined based on the information of the first head detection box and the information of the third head detection box.

3. The distance determination method according to claim 2, characterized in that, The step of determining the information of the third head detection box of the user corresponding to the first head detection box in the information of the second head detection box includes: The intersection-exchange ratio and motion information of the first head detection frame and the second head detection frame are determined based on the information of the first head detection frame and the second head detection frame. If the cross-union ratio exceeds a set threshold and the motion information is the same, the information of the second head detection box is determined to be the information of the third head detection box of the user corresponding to the first head detection box.

4. The distance determination method according to claim 2, characterized in that, Determining the target head detection box information corresponding to the user based on the information of the first head detection box and the information of the third head detection box includes: The information of the third head detection box is input into the head detection box prediction model, and the information of the predicted head detection box of the user in the image frame is determined according to the information of the third head detection box and the model parameters of the head detection box prediction model. The target head detection box information corresponding to the user is determined based on the information of the first head detection box, the information of the predicted head detection box, and the preset weight parameters.

5. The distance determination method according to any one of claims 1-4, characterized in that, The second calibration constant includes a fitting coefficient, an exponential coefficient, and a bias coefficient. Determining the first distance information based on the detection frame height and the first calibration constant included in the first ranging model includes: The first distance information is determined based on the detection frame height and the first calibration constant included in the first ranging model, wherein the first distance information is the ratio of the first calibration constant to the detection frame height; The step of determining the second distance information based on the detection frame width and the second calibration constant included in the second ranging model includes: Calculate the exponent of the detection box width to obtain the result of the exponentiation; The second distance information is obtained by adding the ratio of the fitting coefficient to the result of the exponentiation operation and the offset coefficient.

6. The distance determination method according to any one of claims 1-4, characterized in that, The step of determining the distance between the user and the camera corresponding to the target head detection box of the target image frame based on the first center horizontal coordinate, the second center horizontal coordinate, the detection box width, the first distance information, and the second distance information includes: The longitudinal distance between the user and the camera corresponding to the target head detection frame is determined based on the detection frame width, the first distance information, and the second distance information. The lateral distance between the user and the camera corresponding to the target head detection frame is determined based on the vertical distance, the first center horizontal coordinate, and the second center horizontal coordinate.

7. The distance determination method according to claim 6, characterized in that, Determining the longitudinal distance between the user and the camera corresponding to the target head detection frame based on the detection frame width, the first distance information, and the second distance information includes: When the width of the detection frame is not less than the first set threshold, the longitudinal distance between the user and the camera corresponding to the target head detection frame is determined based on the first center horizontal coordinate, the second center horizontal coordinate, and the second distance information. If the width of the detection frame does not exceed the second preset threshold, the longitudinal distance between the user and the camera corresponding to the target head detection frame is determined based on the first center horizontal coordinate, the second center horizontal coordinate, and the first distance information. When the width of the detection frame is greater than a first preset threshold and less than a second preset threshold, the longitudinal distance between the user and the camera corresponding to the target head detection frame is determined based on the first center horizontal coordinate, the second center horizontal coordinate, the first distance information, the second distance information, and the preset weight parameters.

8. The distance determination method according to claim 1, characterized in that, The method further includes: Obtain the area parameters of the preset distance fence, the area parameters including a first longitudinal distance, a second longitudinal distance, a first lateral distance, and a second lateral distance; wherein, the first longitudinal distance is less than the second longitudinal distance, and the first lateral distance is less than the second lateral distance; If the vertical distance between the user and the camera corresponding to the head detection frame is between the first vertical distance and the second vertical distance, and the horizontal distance is between the first horizontal distance and the second horizontal distance, it is determined that the user is within the preset distance fence, and the operation corresponding to the preset distance fence is executed.

9. A distance determining device, characterized in that, The device includes: The acquisition module is used to acquire conference video data, which includes multiple image frames and the first center horizontal coordinate of the image frames; The extraction module is used to extract features from the target image frame to obtain information of the head detection box corresponding to at least one user in the target image frame. The information of the head detection box includes the detection box height, detection box width, and the second center horizontal coordinate of the head detection box. The target image frame is any image frame included in the conference video data. The determination module is used to, for the target head detection box of the target image frame, input the height of the detection box into a first ranging model, determine first distance information based on the detection box height and a first calibration constant included in the first ranging model, input the width of the detection box into a second ranging model, and determine second distance information based on the detection box width and a second calibration constant included in the second ranging model; wherein, the target head detection box is any head detection box in the target image frame; The determining module is further configured to determine the distance between the user and the camera corresponding to the target head detection box of the target image frame based on the first center horizontal coordinate, the second center horizontal coordinate, the detection box width, the first distance information, and the second distance information.

10. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the distance determination method as described in any one of claims 1-8.

Images