A visual detection control method and system based on pose and motion rules

Abstract

The application relates to the technical field of visual detection, in particular to a visual detection control method and system based on a pose and a motion law. Motion data of a user is detected through an inertial measurement unit; a time sequence motion law prediction model is established based on the motion data according to the motion habit of the user; a motion cycle trend is predicted through the time sequence motion law prediction model, and images are collected according to the motion cycle; the collected images are preliminarily processed according to image pose coordinates; the images are uploaded to a backend data center; the backend data center establishes a mapping relationship between image data and motion data; the images are subjected to targeted motion filtering compensation according to the mapping relationship; image matching is carried out with real scene images pre-stored in the backend data center, and visual detection is carried out according to the image matching information. The application reduces motion blurring generated in the image collection process, ensures the image definition, and improves the efficiency of image recognition and visual detection.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, and more specifically, to a visual inspection control method and system based on pose and motion laws. Background Technology

[0002] Visual inspection uses machines to replace human eyes for measurement and judgment. It involves using machine vision products to convert the captured target into image signals, which are then transmitted to a dedicated image processing system. Based on pixel distribution and information such as brightness and color, these signals are converted into digital signals. The image system performs various calculations on these signals to extract the target's features, and then controls the on-site equipment based on the judgment results. It is a valuable mechanism for production, assembly, or packaging, and has immeasurable value in detecting defects and preventing defective products from being delivered to consumers.

[0003] When performing visual inspection using a moving camera, motion blur can easily occur due to user movement. Specifically, motion blur manifests as an overlap of the object's position within the exposure time when objects in the scene or the camera move. Due to the limitations of typical camera frame rates, objects moving with the camera will appear blurred; this is called motion blur. Motion blur, caused by the movement effect in the image, can confuse the precise location of the target and the detection results during subsequent target detection. Furthermore, because human movement and the inherent frequency characteristics of bodily motion vary from person to person, existing gimbal stabilization algorithms are not well-suited for this. Summary of the Invention

[0004] The present invention aims to overcome at least one of the defects (deficiencies) of the prior art and provide a visual detection control method and system based on pose and motion laws. The method acquires images according to the laws of human motion, processes the images according to their pose coordinates, and performs image matching with the original real-world images. This reduces motion blur generated during image acquisition, ensuring image clarity while improving the efficiency of image recognition and visual detection.

[0005] The technical solution adopted in this invention is:

[0006] Firstly, a visual detection and control method based on pose and motion patterns is provided, comprising: detecting user motion data through an inertial measurement unit; establishing a time-series motion pattern prediction model based on the motion data according to the user's motion habits; predicting the motion cycle trend through the time-series motion pattern prediction model and acquiring images according to the motion cycle; performing preliminary processing on the acquired images based on the image pose coordinates; uploading the pre-processed images to a back-end data center; establishing a mapping relationship between the image data and motion data in the back-end data center; performing targeted motion filtering compensation on the images according to the mapping relationship in the back-end data center; performing image matching between the motion-filtered compensated images and pre-stored real-scene images in the back-end data center, and performing visual detection based on the image matching information.

[0007] This invention utilizes an inertial measurement unit (IMU) mounted on a helmet-mounted device to measure an object's three-axis attitude angles (including yaw, pitch, and roll) and acceleration. Before image acquisition, a time-series motion pattern prediction model is established based on IMU data, taking into account the user's movement habits. An image sampling strategy is established using motion vibration patterns, and preliminary processing is performed based on the image pose coordinates. This reduces the workload of image data processing in the backend data center, improving efficiency and quality, and facilitating process quality analysis. It is compatible with the movement characteristics of different users, reduces motion blur during image acquisition, and ensures image clarity while improving the efficiency of image recognition, detection, and verification.

[0008] Furthermore, the aforementioned prediction model for time-series motion patterns based on motion data, tailored to the user's motion habits, specifically includes: outputting acceleration signals and yaw, pitch, and roll angle values ​​from the inertial measurement unit component; integrating the acceleration components of the yaw, pitch, and roll angles to obtain motion speed and distance; using a long short-term memory time-series model to predict the acceleration signals, extracting the vibration frequency patterns generated by the motion, and establishing a prediction model for time-series motion patterns.

[0009] Since human movement and the inherent frequency characteristics of bodily motion vary from person to person, existing gimbal stabilization algorithms cannot effectively take into account these differences. This invention provides a Long Short-Term Memory (LSTM) time series model to predict acceleration signals, extract the vibration frequency patterns generated by human movement, and calculate the pose information of the image acquisition device using the angle values ​​of the three-axis attitude angles. Based on the user's movement habits, this invention more effectively reduces motion blur generated during image acquisition.

[0010] Furthermore, before image acquisition, the process includes: obtaining the yaw angle, pitch angle, and roll angle values ​​input by the inertial measurement unit component to obtain the yaw angle in three coordinate axes; detecting the yaw angle using an attitude detection algorithm, automatically correcting the yaw based on the detection results, and calibrating the zero-point reference. Specifically, this includes: calibrating the zero-point reference in three dimensions based on the user's posture, with the front being the X direction, the direction parallel to the shoulder being the Y direction, and the direction perpendicular to the shoulder being the Z direction.

[0011] In practical use, the deflection angle detection can identify whether the helmet is worn crookedly. After self-checking, the zero-point reference is recalibrated, and automatic deflection correction is performed for the wearing state. Correction is automatically completed within a 5-degree deflection angle threshold range; beyond the 5-degree threshold range, manual calibration is required based on prompts. Zero-point reference calibration based on the user's posture is compatible with different user habits, resulting in more significant image processing effects.

[0012] Furthermore, the prediction of motion cycle trends using a time-series motion pattern prediction model and the acquisition of images based on the motion cycle specifically includes: making predictions based on the time-series motion pattern prediction model, determining a relatively static state when the motion cycle is stable or the acceleration value is constant, and acquiring images when the state is relatively static.

[0013] Images acquired in a relatively static state have higher image clarity, which can effectively reduce the workload of subsequent image data processing and improve work efficiency and quality.

[0014] Furthermore, the preliminary processing of the acquired image based on the image pose coordinates specifically includes: judging the acquired image based on the deflection angle information of the three coordinate axes obtained from the inertial measurement unit component, and performing preliminary processing on the acquired image based on the judgment result.

[0015] More specifically, the process involves judging the acquired image based on the three-dimensional deflection angle information obtained from the inertial measurement unit component, and performing preliminary processing on the acquired image based on the judgment result. Specifically, this includes: when the deflection angle of the image is greater than a preset threshold angle, the image is discarded; when the deflection angle of the image is less than the preset threshold angle, the image is rotated and corrected based on the calibrated zero-point reference.

[0016] The threshold angle is adjustable and is determined based on the actual situation. The purpose of setting the threshold angle is to remove images with large deflection angles that do not need to be detected. Based on the deflection angle information of the three coordinate axes provided by the IMU, the image is assigned coordinate values. Before being uploaded to the backend data center, the image is rotated according to the calibrated zero-point reference so that the angle of the acquired image is close to that of the real-world image pre-stored in the backend data center. This allows the image to be embedded into the original image, facilitating matching and detection and reducing the workload of image processing in the backend data center.

[0017] Furthermore, the backend data center performs targeted motion filtering compensation on the image based on the mapping relationship. Specifically, this includes: the backend data center performs a sharpness self-detection on the uploaded image; if the sharpness does not meet the preset requirements, an image motion filtering compensation algorithm is performed based on the motion data to obtain the motion-filtered compensated image.

[0018] More specifically, if the clarity does not meet the preset requirements, an image motion filtering compensation algorithm is performed based on the motion data to obtain a motion-filtered compensated image. This specifically includes: capturing video of the motion process, performing motion filtering compensation in various directions based on the video, adding acceleration collected by the inertial measurement unit, fitting the blurring range, and obtaining a motion-filtered compensated image.

[0019] If better image quality is required, an image motion filtering compensation algorithm can be applied based on motion data. This algorithm compensates for acceleration and velocity values ​​in various directions and re-matches pixel values ​​between consecutive frames to further eliminate motion blur. Whether motion filtering is needed in practical applications depends on the requirements of visual inspection.

[0020] Furthermore, it also includes: locating users based on image matching information, and the backend data center monitoring dangerous situations based on visual detection information and location information, issuing alarm commands when dangerous situations occur.

[0021] This invention performs indoor positioning by matching user-collected images with real-world images. This method is more accurate than simply using a panoramic camera for positioning. It also avoids situations where the exact location cannot be determined when there are obstructions. Furthermore, visual detection provides more timely hazard monitoring and alarms than manual alarms, and the monitored area is also wider.

[0022] Secondly, a vision detection and control system based on pose and motion patterns is provided, including a helmet device and a back-end data center;

[0023] The helmet device includes an attitude detection module, an image acquisition module, an image processing module, a communication module, and an alarm module;

[0024] The attitude detection module acquires acceleration signals and yaw, pitch, and roll angle values ​​based on the inertial measurement unit. It uses attitude detection algorithms to determine whether the helmet device is worn correctly, automatically corrects the yaw state after wearing, and establishes a time series motion pattern prediction model based on motion data according to the user's movement habits. The time series motion pattern prediction model is used to predict the trend of the motion cycle.

[0025] The image acquisition module determines a relatively stationary state when the motion cycle is stable or the acceleration value is constant, and performs image acquisition in the relatively stationary state.

[0026] The image processing module judges the acquired image based on the deflection angle obtained by the posture detection module. When the deflection angle of the image is greater than the preset threshold angle, the image is discarded. When the deflection angle of the image is less than the preset threshold angle, the image is rotated and corrected according to the calibrated zero-point reference.

[0027] The communication module uploads the pre-processed images to the backend data center and receives alarm commands from the backend data center.

[0028] The alarm module triggers an alarm when the communication module receives an alarm command.

[0029] The backend data center includes a communication node module and a data processing module;

[0030] The communication node module is used to connect to the communication module of the helmet device, receive images uploaded by the communication module, and issue alarm commands to the communication module when a dangerous situation occurs.

[0031] The data processing module establishes a mapping relationship between the image data and motion data received by the communication node module, performs targeted motion filtering compensation on the image based on the mapping relationship, matches the motion-filtered image with pre-stored real-scene images, performs visual detection based on the image matching information, locates the helmet device based on the image matching information, and monitors dangerous situations based on the visual detection information and the location information.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] (1) The present invention utilizes an inertial measurement unit (IMU) installed on the helmet device. Before image acquisition, a time series motion pattern prediction model is established based on the IMU data according to the user's motion habits. This model is compatible with the motion characteristics of different users, reduces motion blur generated during image acquisition, and improves the efficiency of image recognition, detection and verification while ensuring image clarity.

[0034] (2) Rotating the image based on the image pose coordinates can reduce the workload of image data processing in the back-end data center, improve efficiency and quality, and facilitate process quality analysis.

[0035] (3) Based on the motion data, an image motion filtering compensation algorithm is performed to compensate for the acceleration and velocity values ​​in various directions. Combined with the pixel values ​​between consecutive frames, the motion blur of the image can be further eliminated. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention.

[0037] Figure 2 This is a system structure diagram of Embodiment 2 of the present invention. Detailed Implementation

[0038] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To better illustrate the following embodiments, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0039] Example 1

[0040] like Figure 1 As shown, this embodiment provides a visual detection and control method based on pose and motion laws, including:

[0041] S1. Detect the user's motion data through an inertial measurement unit;

[0042] S2. Based on the user's exercise habits, establish a time-series exercise pattern prediction model based on exercise data;

[0043] S3. Predict the trend of motion cycle using a time series motion pattern prediction model, and collect images based on the motion cycle;

[0044] S4. Perform preliminary processing on the acquired image based on the image pose coordinates;

[0045] S5. Upload the pre-processed image to the backend data center;

[0046] S6, the backend data center establishes a mapping relationship between image data and motion data;

[0047] S7. The backend data center performs targeted motion filtering compensation on the image based on the mapping relationship;

[0048] S8. Perform image matching between the motion-filtered compensated image and the pre-stored real-scene images in the backend data center, and perform visual detection based on the image matching information.

[0049] Taking an aluminum casting workshop as an example: Workers wearing helmets monitor the casting well in real time during their work, observing and recording different angles of the casting well through various movements such as walking, bending over, and looking down. During this process, the multi-functional camera worn on the head follows the human body in a dynamic process, and each person has their own habitual movement cycle and specific frequency, so the video footage is difficult to guarantee effectiveness. Moreover, the worker is in a free state, and the angle of the helmet facing the casting site, including pitch, yaw, and roll angles, results in a certain angle between the captured image and the casting scene, which has a certain impact on data processing and recognition. Adding gimbal stabilization facilities would increase costs and weight, and most of the footage would be unusable after data processing, occupying information resources and making it difficult to use for evaluation and recognition, severely limiting the function of the multi-functional helmet.

[0050] This embodiment utilizes an inertial measurement unit (IMU) mounted on the helmet-mounted device to measure the object's three axes (yaw, pitch, and roll) and acceleration. Before image acquisition, a time-series motion pattern prediction model is established based on the user's movement habits and IMU data. An image sampling strategy is established using motion vibration patterns, and preliminary processing is performed based on the image pose coordinates. This reduces the workload of image data processing in the backend data center, improves efficiency and quality, and facilitates process quality analysis. It is compatible with the movement characteristics of different users, reduces motion blur during image acquisition, ensures image clarity, and improves the efficiency of image recognition, detection, and verification.

[0051] Step S2 in this embodiment specifically includes:

[0052] S201, the inertial measurement unit component outputs acceleration signals as well as yaw, pitch, and roll angle values;

[0053] S202. Integrate the acceleration components of yaw angle, pitch angle, and roll angle to obtain the motion speed and distance;

[0054] S203. Use a long short-term memory time series model to predict acceleration signals, extract the vibration frequency patterns generated by motion, and establish a time series motion pattern prediction model.

[0055] Since human movement and the inherent motion frequency characteristics of the body vary from person to person, the existing gimbal stabilization processing algorithms cannot take good care of them. This embodiment provides a Long Short-Term Memory (LSTM) time series model to predict acceleration signals, extract the vibration frequency patterns generated by human movement, and calculate the pose information of the image acquisition device through the angle values ​​of the three-axis attitude angles. Based on the user's movement habits, it can more effectively reduce motion blur generated during image acquisition.

[0056] Before image acquisition, this embodiment also includes:

[0057] The yaw angle, pitch angle, and roll angle values ​​are input from the inertial measurement unit component to obtain the yaw angle in three coordinate axes. The yaw angle is detected by the attitude detection algorithm, and automatic yaw correction is performed based on the detection results. The zero-point reference is calibrated, which specifically includes: zero-point reference calibration in three dimensions based on the user's posture, with the front being the X direction, the direction parallel to the shoulder being the Y direction, and the direction perpendicular to the shoulder being the Z direction.

[0058] In practical use, the deflection angle detection can identify whether the helmet is worn crookedly. After self-checking, the zero-point reference is recalibrated, and automatic deflection correction is performed for the wearing state. Correction is automatically completed within a 5-degree deflection angle threshold range; beyond the 5-degree threshold range, manual calibration is required based on prompts. Zero-point reference calibration based on the user's posture is compatible with different user habits, resulting in more significant image processing effects.

[0059] Step S3 in this embodiment specifically includes: making a prediction based on the time series motion law prediction model, determining a relatively static state when the motion period is stable or the acceleration value is constant, and acquiring images when the state is relatively static.

[0060] Step S4 in this embodiment specifically includes: judging the acquired image based on the deflection angle information of the three coordinate axes obtained from the inertial measurement unit component, and performing preliminary processing on the acquired image based on the judgment result.

[0061] When the deflection angle of an image is greater than a preset threshold angle, the image is discarded; when the deflection angle of an image is less than a preset threshold angle, the image is rotated and corrected according to the calibrated zero-point reference.

[0062] The threshold angle is adjustable and is determined based on the actual situation. In this embodiment, the position directly facing the workstation is taken as the zero point, and 50 degrees is set as the threshold angle. The purpose of setting the threshold angle is to remove images with large deflection angles that do not need to be detected. Based on the deflection angle information of the three coordinate axes provided by the IMU, the image is assigned coordinate values. Before being uploaded to the backend data center, the image is rotated according to the calibrated zero-point reference so that the angle of the acquired image is close to that of the pre-stored real-scene image in the backend data center. This allows the image to be embedded into the original image, facilitating subsequent matching and detection and reducing the workload of image processing in the backend data center.

[0063] Step S7 in this embodiment specifically includes: the backend data center performs a self-detection of the image sharpness based on the uploaded image; if the sharpness does not meet the preset requirements, an image motion filtering compensation algorithm is performed based on the motion data to obtain the motion-filtered compensated image.

[0064] The motion process is captured by video, and motion filtering compensation is performed in various directions based on the video. Acceleration collected by the inertial measurement unit is added, and the blurring range is fitted to obtain the image after motion filtering compensation.

[0065] If better image quality is required, an image motion filtering compensation algorithm can be applied based on motion data. This algorithm compensates for acceleration and velocity values ​​in various directions and re-matches pixel values ​​between consecutive frames to further eliminate motion blur. Whether motion filtering is needed in practical applications depends on the requirements of visual inspection.

[0066] This embodiment also includes: locating the user based on image matching information, and the backend data center monitoring dangerous situations based on visual detection information and location information, and issuing an alarm command when a dangerous situation occurs.

[0067] Indoor positioning is achieved by matching user-collected images with real-world images, which is more accurate than simply using panoramic cameras. It avoids situations where the exact location cannot be determined when there are obstructions. Furthermore, visual detection provides more timely hazard monitoring and alarms than manual alarms, and the monitored range is also wider.

[0068] Example 2

[0069] like Figure 2 As shown, this embodiment provides a visual detection and control system based on pose and motion patterns, including a helmet device 100 and a back-end data center 200;

[0070] The helmet device 100 includes an attitude detection module 101, an image acquisition module 102, an image processing module 103, a communication module 104, and an alarm module 105;

[0071] The attitude detection module 101 acquires acceleration signals and angle values ​​of yaw, pitch, and roll angles based on the inertial measurement unit. It determines whether the helmet device is worn correctly through the attitude detection algorithm, performs automatic deflection correction on the state after wearing, and establishes a time series motion pattern prediction model based on motion data according to the user's movement habits. It predicts the trend of motion cycle through the time series motion pattern prediction model.

[0072] The image acquisition module 102 determines that the motion cycle is stable or the acceleration value is constant, and performs image acquisition in the relatively static state.

[0073] The image processing module 103 judges the acquired image based on the deflection angle obtained by the attitude detection module 102. When the deflection angle of the image is greater than the preset threshold angle, the image is discarded. When the deflection angle of the image is less than the preset threshold angle, the image is rotated and corrected according to the calibrated zero-point reference.

[0074] The communication module 104 uploads the pre-processed image to the backend data center 200 and receives alarm commands from the backend data center 200.

[0075] The alarm module 105 triggers an alarm when the communication module 104 receives an alarm command, typically through sound, light, or vibration.

[0076] This embodiment takes the casting well workspace as an example. The casting well is a cuboid structure. Therefore, when a worker faces the casting well, the image has a long straight edge. By using HOG edge extraction, the planar contour can be obtained. The tilt angle of the human body facing the casting well can be determined based on the number of pixels in the image. At the same time, the yaw angle of the IMU is combined with the estimated and measured values ​​of the two using a filtering algorithm to obtain a more accurate angle value. When the image information is insufficient, the IMU value is given greater weight. Similarly, the pitch angle and roll angle are also detected and judged using this principle.

[0077] In practical use, workers wear the helmet device. A posture detection algorithm checks whether the worker is wearing the helmet correctly, determining the worker's walking direction based on acceleration values ​​to confirm proper helmet wearing. It also detects tilting angles to check for misalignment. After self-checking, it automatically corrects the tilting after wearing the helmet. Each helmet is uniquely associated with a worker's ID. Upon power-up and when a worker enters the foundry well work area wearing the helmet device, the device's communication module automatically connects to a communication node module within the production space. Once connected, the communication node module captures the worker's ID, and after the signal connection is established, the image acquisition module is activated.

[0078] When performing image processing, staff wearing helmets may exhibit various postures such as tilting their heads, looking sideways, looking up, or looking down while shooting. Therefore, image rotation processing is performed. The first step is posture calculation to determine the angle of the lens view. Then, the image is rotated to achieve a better visual perspective.

[0079] The backend data center 200 includes a communication node module 201 and a data processing module 202;

[0080] The communication node module 201 is used to connect to the communication module of the helmet device, receive images uploaded by the communication module 104, and issue alarm commands to the communication module when a dangerous situation occurs. It usually uses Wi-Fi or Bluetooth communication.

[0081] The data processing module 202 establishes a mapping relationship between the image data and motion data received by the communication node module 201, performs targeted motion filtering compensation on the image based on the mapping relationship, matches the motion-filtered image with a pre-stored real-scene image, performs visual detection based on the image matching information, locates the helmet device 100 based on the image matching information, and monitors dangerous situations based on the visual detection information and the location information.

[0082] In this embodiment, the helmet device, combined with a backend data center, can effectively collect close-up images and transmit them back for inspection, playing a crucial role in compliance inspection of the production process. Simultaneously, by matching the images collected by the helmet with pre-stored real-world images, visual inspection is performed based on the image matching information. The helmet device is then located based on this matching information. Furthermore, based on both visual inspection and location information, hazardous situations are monitored, effectively overseeing dangerous situations during the production process and promptly notifying workers when such situations occur.

[0083] In this embodiment, the helmet device establishes a predictive model based on the law of motion vibration to acquire images and performs preliminary screening and rotation processing on the images, which reduces the workload of the back-end data center in processing images and improves the overall efficiency of image recognition and visual inspection.

[0084] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A visual detection and control method based on pose and motion laws, characterized in that, include: The user's motion data is detected through an inertial measurement unit; Based on users' exercise habits, a time-series exercise pattern prediction model is established using exercise data. The motion cycle trend is predicted by using a time series motion pattern prediction model, and images are collected based on the motion cycle. The acquired image is initially processed based on its pose coordinates. The pre-processed images are uploaded to the backend data center; The backend data center establishes a mapping relationship between image data and motion data; The backend data center performs targeted motion filtering compensation on the image based on the mapping relationship; The motion-filtered image is matched with the real-world images pre-stored in the backend data center, and visual detection is performed based on the image matching information. The aforementioned prediction model for predicting time-series exercise patterns based on exercise data, targeting users' exercise habits, specifically includes: The inertial measurement unit (IMU) outputs acceleration signals as well as yaw, pitch, and roll angle values. Integrate the acceleration components of yaw, pitch, and roll to obtain the velocity and distance. By using a long short-term memory time series model to predict acceleration signals, the vibration frequency patterns generated by motion are extracted, and a time series motion pattern prediction model is established.

2. The visual detection and control method based on pose and motion laws according to claim 1, characterized in that, Before image acquisition, the following steps are also included: The yaw angle, pitch angle, and roll angle values ​​output by the inertial measurement unit component are used to obtain the yaw angles in the three coordinate axes. The deflection angle is detected by a posture detection algorithm, and automatic deflection correction is performed based on the detection results. The zero-point reference is calibrated, which includes: zero-point reference calibration in three dimensions based on the user's posture: the front is the X direction, the direction parallel to the shoulder is the Y direction, and the direction perpendicular to the shoulder is the Z direction.

3. The visual detection and control method based on pose and motion laws according to claim 2, characterized in that, The prediction of motion cycle trends using a time-series motion pattern prediction model, which involves acquiring images based on the motion cycle, specifically includes: Prediction is made based on the time series motion law prediction model. When the motion period is stable or the acceleration value is constant, it is judged as a relatively static state, and image acquisition is performed when it is in a relatively static state.

4. The visual detection and control method based on pose and motion laws according to claim 3, characterized in that, The preliminary processing of the acquired image based on the image pose coordinates specifically includes: The inertial measurement unit (IMU) obtains the deflection angle information of the three coordinate axes to judge the acquired image, and performs preliminary processing on the acquired image based on the judgment result.

5. The visual detection and control method based on pose and motion laws according to claim 4, characterized in that, The process involves judging the acquired image based on the three-dimensional deflection angle information obtained from the inertial measurement unit component, and performing preliminary processing on the acquired image based on the judgment result, specifically including: If the deflection angle of an image is greater than a preset threshold angle, the image is discarded. When the deflection angle of the image is less than the preset threshold angle, the image is rotated and corrected according to the calibrated zero-point reference.

6. The visual detection and control method based on pose and motion laws according to claim 3, characterized in that, The backend data center performs targeted motion filtering compensation on the image based on the mapping relationship, specifically including: The backend data center performs self-resolution detection based on the uploaded images; If the clarity does not meet the preset requirements, an image motion filtering compensation algorithm is performed based on the motion data to obtain the motion-filtered compensated image.

7. The visual detection and control method based on pose and motion laws according to claim 6, characterized in that, If the clarity does not meet the preset requirements, an image motion filtering compensation algorithm is performed based on the motion data to obtain a motion-filtered compensated image. Specifically, this includes: capturing video of the motion process, performing motion filtering compensation in each direction based on the video, adding acceleration collected by the inertial measurement unit, fitting the blurring range, and obtaining a motion-filtered compensated image.

8. A visual detection and control method based on pose and motion laws according to any one of claims 1-7, characterized in that, Also includes: The user is located based on image matching information. The backend data center monitors dangerous situations based on visual detection information and location information, and issues an alarm command when a dangerous situation occurs.

9. A visual detection and control system based on pose and motion laws, applied to the visual detection and control method based on pose and motion laws as described in claim 1, characterized in that, Including helmet-mounted devices and backend data centers; The helmet device includes an attitude detection module, an image acquisition module, an image processing module, a communication module, and an alarm module; The attitude detection module acquires acceleration signals and yaw, pitch, and roll angle values ​​based on the inertial measurement unit. It uses attitude detection algorithms to determine whether the helmet device is worn correctly, automatically corrects the yaw state after wearing, and establishes a time series motion pattern prediction model based on motion data according to the user's movement habits. The time series motion pattern prediction model is used to predict the trend of the motion cycle. The image acquisition module determines a relatively stationary state when the motion cycle is stable or the acceleration value is constant, and performs image acquisition in the relatively stationary state. The image processing module judges the acquired image based on the deflection angle obtained by the posture detection module. When the deflection angle of the image is greater than the preset threshold angle, the image is discarded. When the deflection angle of the image is less than the preset threshold angle, the image is rotated and corrected according to the calibrated zero-point reference. The communication module uploads the pre-processed images to the backend data center and receives alarm commands from the backend data center. The alarm module triggers an alarm when the communication module receives an alarm command. The backend data center includes a communication node module and a data processing module; The communication node module is used to connect to the communication module of the helmet device, receive images uploaded by the communication module, and issue alarm commands to the communication module when a dangerous situation occurs. The data processing module establishes a mapping relationship between the image data and motion data received by the communication node module, performs targeted motion filtering compensation on the image based on the mapping relationship, matches the motion-filtered image with pre-stored real-scene images, performs visual detection based on the image matching information, locates the helmet device based on the image matching information, and monitors dangerous situations based on the visual detection information and the location information.

Images