An indoor human target recognition and monocular distance measurement method and system
By combining a lightweight convolutional neural network model with a human target detection and tracking model, the error problem of monocular camera ranging is solved, realizing high-precision indoor human target recognition and ranging in home environments, which is suitable for embedded devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2023-01-29
- Publication Date
- 2026-05-29
AI Technical Summary
In existing technologies, the ranging method of monocular camera is greatly affected by camera calibration parameters, environmental factors and object position, and there are few methods that can perform object recognition and ranging at the same time. It has errors and limitations, especially in home environment where security is insufficient.
A lightweight convolutional neural network model is used for indoor human target recognition and monocular ranging. Combined with a human target detection and tracking model, the IOU parameter is used to determine missed detections. The coordinates of the occluded human body are calculated using the height of the human body bounding box when it is not occluded, and the ranging is adapted for embedded deployment.
It improves the accuracy of target recognition and the precision of ranging, reduces the computational resource requirements, achieves a ranging error of about 25cm, and is suitable for deployment in embedded devices.
Smart Images
Figure CN116311351B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of target detection and image processing technology, specifically to an indoor human target recognition and monocular ranging method and system. Background Technology
[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.
[0003] Today, with the rapid development of society and technology, and the improvement of people's living standards and spiritual needs, security issues are receiving increasing attention, and the demand for video surveillance systems is experiencing explosive growth. Real-time crowd detection and counting can be applied to security in public places such as hospitals, train stations, and tourist attractions. Meanwhile, the detection and analysis of abnormal human targets is also an important component of intelligent video surveillance systems. In the field of security monitoring, obtaining both the location and distance information of human targets is crucial for alarms against abnormal intrusions. Especially in home environments, where only the elderly or children are home alone, safety is significantly reduced. Therefore, systems such as human collision avoidance systems and hazard warning systems are needed to protect personal safety. To achieve these functions, it is necessary to identify and estimate the distance of human targets in the environment. Furthermore, the intelligent control of some home appliances also relies on the identification and counting of human targets and their distance from the devices. Adjusting device parameters based on the number of people and their distance from the devices allows for real-time intelligent adjustment, making the living environment more comfortable.
[0004] Currently, ranging methods include ultrasonic ranging, lidar ranging, binocular stereo vision ranging, and monocular camera ranging. Ultrasonic ranging, lidar ranging, and binocular stereo vision ranging are susceptible to environmental influences and are expensive. Camera-based ranging methods, on the other hand, process image information, offering better stability. Monocular ranging is also simpler to use and more cost-effective than binocular ranging. Regarding monocular camera ranging, since monocular cameras cannot capture depth information, they are difficult to use for distance measurement. Common methods utilize pinhole imaging and the principle of similar triangles to achieve simple object distance measurement. However, this method is affected by camera calibration parameters, environmental factors, and object position, resulting in some error compared to the actual distance. Furthermore, existing technologies rarely offer methods for simultaneous real-time object recognition and ranging, and those that exist have certain limitations. Summary of the Invention
[0005] To address the aforementioned issues, this disclosure proposes an indoor human target recognition and monocular ranging method and system. It employs a deep learning-based convolutional neural network model, trains the model using a lightweight convolutional neural network, and deploys the lightweight model on an embedded device, applying the neural network in an AI chip. Simultaneously, the detection and tracking systems complement each other to overcome the problem of missed detections, while providing important correction information for ranging of occluded targets, thereby improving performance and efficiency.
[0006] According to some embodiments, the present disclosure adopts the following technical solutions:
[0007] An indoor human target recognition and monocular ranging method includes:
[0008] Acquire image data of indoor human targets and perform preprocessing;
[0009] A lightweight neural network human target detection and tracking model is constructed. First, image data is input into the target detection and tracking model to perform target detection, and the target bounding box of the current frame image is retained.
[0010] The target human body detected in the current frame is used as the tracking target in the next frame. The image feature information of the tracking target is extracted, and a tracking box is generated in the next frame by matching the image feature information. Then, the IOU parameter is calculated with the human body target box detected in the current frame to determine whether there is a missed detection.
[0011] When a human target is occluded, find the human bounding box of the previous frame when it was not occluded. Calculate the coordinates of the bottom edge of the occluded human bounding box based on its height. Then, use monocular ranging to measure the new coordinates.
[0012] According to some embodiments, the present disclosure adopts the following technical solutions:
[0013] An indoor human target recognition and monocular ranging system includes:
[0014] The image acquisition module is used to acquire image data of indoor human targets and perform preprocessing.
[0015] The human detection and tracking module is used to build a lightweight neural network human target detection model and tracking model. First, the image data is input into the target detection model to perform target detection and retain the target bounding box of the current frame image.
[0016] The target human body detected in the current frame is used as the tracking target in the next frame. The target image feature information is extracted by the tracking model and a tracking box is generated by matching the image feature information in the next frame. Then, the IOU parameter is calculated with the human body target box detected in the current frame to determine whether there is a missed detection.
[0017] The monocular ranging module is used to find the human bounding box of the occluded target in the previous frame when the target was not occluded, calculate the coordinates of the bottom edge of the occluded human bounding box based on the height of the unoccluded human bounding box, and perform monocular ranging on the new coordinates.
[0018] According to some embodiments, the present disclosure adopts the following technical solutions:
[0019] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned indoor human target recognition and monocular ranging method.
[0020] According to some embodiments, the present disclosure adopts the following technical solutions:
[0021] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the aforementioned indoor human target recognition and monocular ranging method.
[0022] Compared with the prior art, the beneficial effects of this disclosure are as follows:
[0023] This invention can utilize human target detection models and human tracking models to complement each other, effectively improving the problems of target missed detection and target occlusion ranging.
[0024] This disclosure proposes a new monocular ranging method, which achieves excellent ranging performance with a ranging error of about 25cm for targets within a certain range.
[0025] This disclosure presents a more lightweight convolutional neural network prediction model, which greatly reduces the use of computing resources while maintaining the model accuracy at a essentially unchanged level. It is also more friendly to embedded deployment and can be directly deployed on artificial intelligence chips. Attached Figure Description
[0026] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.
[0027] Figure 1 A general flowchart of the human target detection system, tracking system, and monocular ranging system provided in the embodiments of this disclosure;
[0028] Figure 2 This is a schematic diagram of the lightweight detection neural network model structure provided in the embodiments of this disclosure;
[0029] Figure 3 This is a schematic diagram of the improved SE-C3 module of the Backbone for the lightweight detection neural network model provided in the embodiments of this disclosure.
[0030] Figure 4 Training records for the lightweight detection neural network model provided in the embodiments of this disclosure;
[0031] Figure 5 A schematic diagram of the ranging area layout of a monocular ranging system provided in an embodiment of this disclosure;
[0032] Figure 6 This is a schematic diagram of camera placement for a monocular ranging system provided in an embodiment of this disclosure.
[0033] Figure 7 This is a schematic diagram illustrating the ranging principle of a monocular ranging system provided in an embodiment of this disclosure. Detailed implementation method:
[0034] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.
[0035] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.
[0036] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0037] Example 1
[0038] One embodiment of this disclosure provides an indoor human target recognition and monocular ranging method, including:
[0039] Step 1: Acquire image data of indoor human targets and perform preprocessing;
[0040] Step 2: Construct a lightweight neural network human target detection and tracking model. First, input the image data into the target detection and tracking model to perform target detection and retain the target bounding box of the current frame image.
[0041] Step 3: Take the target human body detected in the current frame as the tracking target in the next frame, extract the image feature information of the tracking target, and generate a tracking box by matching the image feature information in the next frame. Then calculate the IOU parameter with the human body target box detected in the current frame to determine whether there is a missed detection.
[0042] Step 4: When a human target is occluded, find the human bounding box of the previous frame when it was not occluded. Based on the height of the human bounding box when it was not occluded, calculate the coordinates of the bottom edge of the human bounding box when it is occluded, and use monocular ranging to measure the new coordinates.
[0043] As one embodiment, in step 1, the process of acquiring image data of indoor human targets and performing preprocessing includes:
[0044] The system uses cameras to capture images of different scenes and people in different locations and under different lighting conditions. The acquired image data is then augmented, and the original image is flipped and mirrored at a certain angle.
[0045] Specifically, (1) the same model and specifications of camera as the one mounted on the RV1126 development board are used to collect images, including fixing the height of the camera at 2.5m and tilting at 35°, and taking indoor photos of different scenes and different people in different positions and under different lighting conditions.
[0046] (2) The dataset of human images obtained by shooting is expanded by flipping and mirroring the original images at a certain angle to obtain a batch of new images, which greatly expands the training image dataset.
[0047] (3) In order to better simulate the real indoor occlusion scene, the Cutout method is used to randomly select a square area of a fixed size and then fill it with 0, which indirectly realizes the erasure of any size and retains more important areas.
[0048] (4) In order to better simulate different brightness environments, brightness enhancement and brightness reduction are used to adjust the brightness of the image to enrich the training set under backlight environment.
[0049] As one embodiment, a lightweight neural network human target detection model and a tracking model are constructed. The improved network model is trained using the sample dataset described in step 1 for 500 training rounds with a batch size of 16, resulting in the lightweight neural network human target detection model and the tracking model.
[0050] Using the sample data obtained in step 1 above, a training set is established, and a lightweight convolutional neural network model is designed and constructed. This network model has two innovations compared to the CSP-Darknet53 network, characterized by:
[0051] By introducing the MobileOne module and integrating MobileOneBlock into the backbone network, the inference time for training and testing is reduced, achieving the goal of lightweight, real-time, and fast detection.
[0052] An attention mechanism is introduced as a plug-and-play attention module that can be added anywhere in the YOLOv5 network. The SE channel attention mechanism is added to Backbone's C3 module.
[0053] It is lightweight and highly adaptable to embedded devices, with a small amount of computation and parameters in the network model. It also supports direct conversion of Onnx or Pt format to rknn model files, which can be deployed to Rockchip chips and integrated into the human detection and tracking model and monocular ranging model built by Embedded3.
[0054] Specifically, the process of porting the model to Rockchip RV1126 is as follows:
[0055] (1) Train the model on a local PC using an improved lightweight convolutional neural network and save the model file with the best performance.
[0056] (2) Convert the format of the saved model, convert the pt format to torchscript.pt, perform dynamic node conversion, and then export it to rknn format.
[0057] (3) Connect the USBOTG of the Rockchip RV1126 core board to the PC and correctly identify the device. Use SSH to enable communication between the RV1126 and the PC for file transfer and code debugging.
[0058] (4) Copy the RKNN model file to the RV1126 and run the RKNN model on the Rockchip RV1126 NPU to identify and predict the location and distance of human targets.
[0059] As one embodiment, the specific implementation process of an indoor human target recognition and monocular ranging method is as follows:
[0060] First, target detection is performed on the input image, and the bounding boxes of the current frame are preserved. The people detected in the current frame are used as the tracking targets for the next frame. Image feature information of the tracking targets is extracted, and a tracking box is generated in the next frame by matching the image feature information. Then, the Intersection over Union (IoU) parameter is calculated between the IoU parameter and the human bounding box detected in the current frame. If the IoU is higher than a set threshold, the target from the previous frame is considered to have been detected in the current frame, indicating no missed detections. In multiple tests, a threshold of 0.75 showed the best results; that is, if the overlap between the bounding boxes of the previous and next frames is greater than 0.75, they are considered the same target. If the target disappears from the field of view in the next frame, no tracking box is generated in the current frame. Simultaneously, the tracking model generates an ID label for each target, updating the ID of the current bounding box during each detection and tracking operation. The ID label for the same target is the same in consecutive frames. Based on this, when a human target is occluded, the bounding box of the occluded target in the previous frame when it was not occluded is found by using the ID. Then, based on the height of the unoccluded bounding box, the coordinates of the bottom edge of the occluded bounding box are calculated, and the new coordinates are used to calculate the distance.
[0061] Furthermore, this disclosure proposes a novel ranging method, the specific implementation of which is as follows:
[0062] To measure the distance from the bottom of a target within a certain range to the camera, assume a rectangular area with a ranging range of (m). Divide this rectangular area into several smaller square areas with a side length of s (m). Each square area corresponds to a ranging formula. First, determine which ranging area the target falls into, and then calculate the distance using the ranging formula for the corresponding area.
[0063] Specifically, (1) Select the ranging area of the camera. Set the ranging area to 11.2×7.2 (m) and divide the area into a 14×9 square grid, with each grid having a side length of 80cm. Place the red card at the vertices of each grid.
[0064] (2) Fix the position of the camera. Set the camera height to 2.5m, the angle to be 35° downward, and place the camera 80cm in front of the center point of the foremost edge of the ranging area.
[0065] (3) Take calibration pictures. Enter the shooting command in the command line of the RV1126 development board to take several pictures with a resolution of 1920×1080.
[0066] (4) For the calibration images obtained in step (3) above, calibrate the pixel coordinates of the grid points corresponding to the red cards, and calculate the actual distance of each point from the camera.
[0067] Furthermore, the method for calculating distance is as follows:
[0068] Since each grid region corresponds to a distance measurement formula, the cross product of vectors is first used to determine which grid region the target falls into. The distance measurement formula is then calculated for the region where the target falls, and the target coordinates are substituted into the distance measurement formula to obtain the distance.
[0069] Three grid points are selected for each region. The coordinates of each pixel and its corresponding distance are substituted into the following equation to fit the ranging parameters a, b, and c for each grid region. The specific method for calculating the ranging formula is as follows:
[0070] R 2 =((x-960)·k x ) 2 +((y-1080)·k y ) 2 +c
[0071] Where R represents the distance from the target to the camera, x and y are the coordinates of the midpoint of the bottom edge of the human bounding box, and k x k y c represents the distance parameter.
[0072] The above formula is rearranged to obtain the following formula:
[0073] R 2 = a(x-960) 2 +b(y-1080) 2 +c
[0074] in
[0075] When a human target is occluded, assuming the coordinates of the human bounding box in the frame before the occlusion occurred are (x... 0, (y0, x1, y1), where (x0, y0) are the coordinates of the top-left corner of the frame and (x1, y1) are the coordinates of the bottom-right corner of the frame. The height h and width w of the human body frame when it is not occluded are calculated as follows:
[0076] h = y1 - y0
[0077] w = x1 - x0
[0078] Calculate the corrected coordinates (x″1, y″1) of the bottom right corner of the human frame used for distance measurement when occluded, based on the height h and width w of the human frame when unoccluded. Assume the coordinates of the human frame when occluded are (x′0, y′0, x′1, y′1):
[0079] x″1=x′0+w
[0080] y″1=y′0+h
[0081] The basic idea of fitting the distance measurement formula using the nonlinear least squares method and finding the undetermined coefficients a, b, and c is as follows:
[0082] Let (x, y, r) be a set of observations, where x = [x1, x2, x3, x4]. T y = [y1, y2, y3, y4] T , where (x i ,y i (i = 1, 2, 3, 4) are the pixel coordinates of the vertices of the square grid, r = R 2 And it satisfies the following theoretical functions:
[0083] r = f(x, y, w)
[0084] Where w = [w1, w2, w3] T The parameters are undetermined, w1 = a, w2 = b, w3 = c; to find the optimal estimate of the parameter w of the function f(x,y,w), given 4 sets of observation data:
[0085] (x i ,y i ,r i (i = 1, 2, 3, 4)
[0086] Find the parameter w that minimizes the objective function L(r,f(x,y,w)). i (i = 1, 2, 3, 4), the objective function is as follows:
[0087]
[0088] Finally, the parameters a, b, and c in the distance measurement formula are obtained.
[0089] Finally, the source file of the host computer display program was compiled on the PC to obtain an executable file, and a push-pull image transmission system was built on the RV1126 and the PC to display the detection screen on the host computer in real time.
[0090] The camera mounted on the embedded RV1126 development board detects objects in real time and sends the images to the detection and tracking network model for prediction. The detection results and distance estimates of human targets are obtained, and the images and distance parameters are transmitted to the host computer for display in real time.
[0091] Example 2
[0092] One embodiment of this disclosure provides an indoor human target recognition and monocular ranging system, including:
[0093] The image acquisition module is used to acquire image data of indoor human targets and perform preprocessing.
[0094] The human detection and tracking module is used to build a lightweight neural network human target detection model and tracking model. First, the image data is input into the target detection model to perform target detection and retain the target bounding box of the current frame image.
[0095] The target human body detected in the current frame is used as the tracking target in the next frame. The target image feature information is extracted by the tracking model and a tracking box is generated by matching the image feature information in the next frame. Then, the IOU parameter is calculated with the human body target box detected in the current frame to determine whether there is a missed detection.
[0096] The monocular ranging module is used to find the human bounding box of the occluded target in the previous frame when the target was not occluded, calculate the coordinates of the bottom edge of the occluded human bounding box based on the height of the unoccluded human bounding box, and perform monocular ranging on the new coordinates.
[0097] Example 3
[0098] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When executed by a processor, the computer instructions implement the steps of the indoor human target recognition and monocular ranging method.
[0099] Example 4
[0100] One embodiment of this disclosure provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform the steps of the indoor human target recognition and monocular ranging method described above.
[0101] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0103] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.
Claims
1. A method for indoor human target recognition and monocular ranging, characterized in that, include: Acquire image data of indoor human targets and perform preprocessing; A lightweight neural network human target detection and tracking model is constructed. First, image data is input into the target detection and tracking model to perform target detection, and the target bounding box of the current frame image is retained. The target human body detected in the current frame is used as the tracking target in the next frame. The image feature information of the tracking target is extracted, and a tracking box is generated in the next frame by matching the image feature information. Then, the IOU parameter is calculated with the human body target box detected in the current frame to determine whether there is a missed detection. When a human target is occluded, find the human bounding box of the previous frame when it was not occluded. Calculate the coordinates of the bottom edge of the occluded human bounding box based on its height. Then, use monocular ranging to measure the new coordinates. The monocular ranging process includes: measuring the distance from the bottom of a target within a certain range to the camera. Assuming the ranging range is within a rectangular area, this rectangular area is divided into several smaller square areas. Each square area corresponds to a ranging formula. Assuming the image resolution captured by the camera is 1920*1080 pixels, the target is first determined to fall within its specified ranging area. Then, the distance is calculated using the ranging formula for that area. Each square area corresponds to a ranging formula as follows: Where R represents the distance from the target to the camera, and x and y are the coordinates of the midpoint of the bottom edge of the human bounding box. c is the distance parameter; The above formula can be rearranged to obtain: , where a = b = ; When a human target is occluded, assuming the coordinates of the human bounding box in the frame before the occlusion occurred are... ,in The coordinates of the top left corner of the box. Using the coordinates of the bottom right corner of the bounding box, the height h and width w of the human bounding box when it is not obscured are calculated as follows: Calculate the coordinates of the bottom right corner midpoint of the bottom edge of the human frame used for distance measurement when occluded, based on the height h and width w of the human frame when unobstructed. Assuming the coordinates of the human bounding box when occluded are... : ， ， Then, the distance measurement formula is fitted using the nonlinear least squares method, and the undetermined coefficients a, b, and c are obtained: set up It is a set of observations. , ,in These are the pixel coordinates of the vertices of the square grid. Let represent the distance from the target to the camera, and satisfy the following theoretical function: ， in These are parameters to be determined. In order to find the function parameters The optimal estimate for the given 4 sets of observation data: Solve the objective function Parameter that takes the minimum value The objective function is as follows: ， Finally, the parameters a, b, and c in the distance measurement formula are obtained; Compared to the CSP-Darknet53 network, the lightweight convolutional neural network model integrates MobileOneBlock into the backbone network and adds an SE channel attention mechanism to the C3 module of the backbone.
2. The indoor human target recognition and monocular ranging method as described in claim 1, characterized in that, The steps of acquiring image data of indoor human targets and performing preprocessing include: The system uses cameras to capture images of different scenes and people in different locations and under different lighting conditions. The acquired image data is then augmented, and the original image is flipped and mirrored at a certain angle.
3. The indoor human target recognition and monocular ranging method as described in claim 1, characterized in that, The process of determining whether a target is missed includes: setting an IOU threshold and comparing it with the calculated IOU parameter value. If the calculated IOU parameter value is higher than the set IOU threshold, it is considered that the target in the previous frame was detected in the current frame and no target was missed. If the target disappears from the field of view in the next frame, no tracking box will be generated in the current frame.
4. The indoor human target recognition and monocular ranging method as described in claim 1, characterized in that, In the tracking model, an ID label is generated for each target. The ID of the current target box is updated each time detection and tracking occurs, and the ID label of the same target is the same in consecutive frames.
5. The indoor human target recognition and monocular ranging method as described in claim 4, characterized in that, When a human target is occluded, the bounding box of the occluded target in the previous frame when it was not occluded is determined by its ID. Then, based on the height of the bounding box when it was not occluded, the coordinates of the bottom edge of the bounding box when it is occluded are calculated, and the distance is calculated using the new coordinates.
6. The indoor human target recognition and monocular ranging method as described in claim 1, characterized in that, The distance parameters obtained after monocular ranging and the detection results after human target detection are transmitted to the host computer in real time and the detection screen is displayed.
7. An indoor human target recognition and monocular ranging system, employing the indoor human target recognition and monocular ranging method as described in any one of claims 1-6, characterized in that, include: The image acquisition module is used to acquire image data of indoor human targets and perform preprocessing. The human detection and tracking module is used to build a lightweight neural network human target detection model and tracking model. First, the image data is input into the target detection and tracking model to perform target detection and retain the target bounding box of the current frame image. The target human body detected in the current frame is used as the tracking target in the next frame. The image feature information of the tracking target is extracted, and a tracking box is generated in the next frame by matching the image feature information. Then, the IOU parameter is calculated with the human body target box detected in the current frame to determine whether there is a missed detection. The monocular ranging module is used to find the human bounding box of the occluded target in the previous frame when the target was not occluded, calculate the coordinates of the bottom edge of the occluded human bounding box based on the height of the unoccluded human bounding box, and perform monocular ranging on the new coordinates.
8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement an indoor human target recognition and monocular ranging method as described in any one of claims 1-6.
9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform an indoor human target recognition and monocular ranging method as described in any one of claims 1-6.