Intelligent human body motion information acquisition and real-time posture detection method

Abstract

The invention belongs to the technical field of human body motion information acquisition and real-time posture detection, and particularly relates to an intelligent human body motion information acquisition and real-time posture detection method which comprises the following specific steps: firstly, wearing an IMU (Inertial Measurement Unit) sensor on a key part of a human body, and then reasonably arranging the position of an RGB-D camera; carrying out preliminary verification on the collected original data, and checking whether abnormal values and data missing exist in sensor data, whether camera images are clear and complete, and whether timestamps of equipment data are synchronous; and performing data enhancement operation on the acquired image data and sensor data. According to the method, multi-sensor fusion and an advanced deep learning algorithm are combined, information of different data sources is fully utilized, the accuracy of posture detection is improved, and compared with a traditional single sensor or algorithm, human body motion details and posture changes can be more accurately captured.

Description

Technical Field

[0001] The present invention relates to the technical field of human motion information collection and real-time posture detection, and in particular to an intelligent human motion information collection and real-time posture detection method. Background Art

[0002] In today's era of rapid technological advancement, the collection of human motion information and real-time posture detection have crucial applications in multiple fields. In medical rehabilitation, it can assist doctors in accurately assessing patients' recovery progress and developing personalized rehabilitation plans. In sports training, it can help coaches monitor athletes' movements and postures in real time, promptly correcting errors and improving training effectiveness. In the smart home and security sectors, it can enable intelligent identification and early warning of human behavior.

[0003] However, traditional human motion information collection and posture detection technologies have numerous limitations. For example, optical motion capture systems require the deployment of multiple infrared cameras within the capture environment, and the user must wear a motion capture suit equipped with optical markers. This is not only costly but also affects the naturalness of human movement. Inertial motion capture systems, while relatively lightweight, have limited accuracy. Therefore, an intelligent method for human motion information collection and real-time posture detection was developed. Summary of the Invention

[0004] In view of the above problems and / or the problems existing in an existing intelligent human motion information collection and real-time posture detection method, the present invention is proposed.

[0005] Therefore, the purpose of the present invention is to provide an intelligent human motion information collection and real-time posture detection method, which can solve the above-mentioned problems that the existing human motion information collection and posture detection technologies have many limitations.

[0006] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions: A method for intelligent human motion information collection and real-time posture detection includes the following specific steps: S1, data collection phase: first wear the IMU sensor on the key parts of the human body, and then reasonably arrange the position of the RGB-D camera; S2, preliminary data verification: perform preliminary verification on the collected raw data to check whether there are any abnormal values ​​or missing data in the sensor data, whether the camera image is clear and complete, and whether the data timestamps of each device are synchronized; S3, Data Augmentation: Perform data augmentation operations on the collected image data and sensor data. For image data, the dataset is expanded by rotating, scaling, cropping, and adding noise. For sensor data, the diversity of the data is increased by translating and stretching the time series data. S4, data processing and feature extraction: first process the sensor data, then process the computer vision data; S5, model prediction: The processed sensor data and computer vision features are input into the trained deep learning model so that the model outputs the current posture information of the human body. This enables the model to track human movement in real time and dynamically monitor posture changes. S6, multi-model cross-validation: Introduce multiple auxiliary models with different structures or training focuses, input the same input data into the main model and auxiliary models respectively, and compare the test results of each model. When the results of each model are highly consistent, the reliability of the test results is confirmed. If there are significant differences, the exception handling mechanism is activated and the data and model are further checked; S7, Posture Assessment and Warning: This system evaluates detected postures based on preset normal posture ranges and movement patterns. When abnormal postures are detected, it issues a warning signal. It also assesses exercise intensity and fatigue levels based on physiological data, providing recommendations for adjusting exercise strategies. S8, Dynamic Model Update: Regularly collect new human motion data, update the trained deep learning model online, and use incremental learning or transfer learning methods to enable the model to continuously learn new motion postures and patterns, adapting to the motion changes of different users and different scenarios; S9, result display and feedback: First, visualize the posture detection results and then collect user feedback.

[0007] As a preferred solution of the intelligent human motion information collection and real-time posture detection method of the present invention, the specific steps of S1 are as follows: S11, sensor deployment: First, wear IMU sensors on key parts of the human body, including the head, torso, and limb joints to ensure accurate collection of motion information from each part. Then, wear corresponding physiological monitoring sensors as needed. S12, Camera setup: Arrange the RGB-D camera position reasonably to ensure that the human body's motion range can be fully captured to obtain high-quality images and depth information.

[0008] As a preferred solution of the intelligent human motion information collection and real-time posture detection method of the present invention, the specific steps of S4 are as follows: S41, sensor data processing: First, the raw data collected by the IMU is filtered to remove noise interference. Then, the accelerometer and gyroscope data are fused using the Kalman filter algorithm to improve the accuracy of attitude solution. Finally, key features are extracted from the physiological monitoring sensor data. S42, Computer Vision Data Processing: First, use image segmentation and edge detection algorithms to preprocess the images obtained by the RGB-D camera to extract the human body contours and key points. Then, use deep learning algorithms to classify and match the key points to determine the exact location of various parts of the human body.

[0009] As a preferred solution of the intelligent human motion information collection and real-time posture detection method of the present invention, the specific steps of S5 are as follows: S51, data preprocessing and integration: first regularize the sensor data, then perform data fusion; S52, model loading and deployment: first load the pre-trained model, then deploy the inference engine; S53, real-time reasoning and prediction: data input first, then model calculation and result analysis.

[0010] As a preferred solution of the method for intelligent human motion information collection and real-time posture detection described in the present invention, the specific steps of S51 are as follows: S511, sensor data regularization: time synchronization and scale calibration of sensor data; S512, data fusion: fusing the processed sensor data and visual data to form a unified input feature vector; The specific steps of S52 are as follows: S521, loading pre-trained model: first load the pre-trained deep learning model into memory, and then load it through the model management interface to ensure the integrity and correctness of the model file to avoid prediction failure due to model damage; S522, deploying the inference engine: First, select a suitable inference engine based on the type and structure of the model to optimize the model and improve the inference speed and efficiency. Then, deploy the loaded model to the inference engine and configure the corresponding operating parameters.

[0011] As a preferred solution of the method for intelligent human motion information collection and real-time posture detection described in the present invention, the specific steps of S53 are as follows: S531, Data Input: The integrated input feature vector is first formatted according to the requirements of the inference engine and then input into the model. During the real-time detection process, new motion data is acquired at regular intervals and input into the model for inference in a timely manner. S532, Model Operation: The inference engine performs forward propagation calculations on the input data based on the model structure and parameters. During this process, its convolutional neural network extracts and abstracts image features, and its recurrent neural network or long short-term memory network models and analyzes time series data, thereby obtaining the model output results. S533, result analysis: first analyze the output result of the model and convert it into human posture information, then classify and identify the posture according to the preset posture classification standard to determine what posture the current human body movement belongs to.

[0012] As a preferred solution of the intelligent human motion information collection and real-time posture detection method of the present invention, the specific steps of S6 are as follows: S61, multi-model selection and preparation: first perform model screening, then perform model adaptation and model loading; S62, synchronous data input: data integration is performed first, followed by parallel input; S63, multi-model reasoning operation: first perform independent reasoning, then record the results; S64, result comparison and evaluation: first conduct consistency analysis, then conduct confidence assessment; S65, decision-making and feedback: make the result decision first, then provide model feedback.

[0013] As a preferred solution of the intelligent human motion information collection and real-time posture detection method of the present invention, the specific steps of S61 are as follows: S611, Model Screening: Select multiple deep learning models for cross-validation based on their structure, training methods, and focus. These deep learning models include models based on a convolutional neural network architecture to capture static human posture features in images, and models based on recurrent neural networks or long short-term memory networks to analyze motion information over time series. Simultaneously, similar models with different loss functions and training data are included to enrich model diversity. S612, Model Adaptation: To ensure that the input data format and feature dimensions of the selected model are compatible with each other, the model is first fine-tuned to adapt it to the unified input specification; S613, model loading: loading multiple pre-trained models into memory simultaneously; The specific steps of S62 are as follows: S621, data integration: the unified input feature vector obtained through data preprocessing and integration is formatted according to the input requirements of each model; S622, Parallel Input: Utilize multi-threaded or distributed computing technology to synchronously input the same input data into each model to fully utilize computing resources, improve the efficiency of model inference, reduce the time overhead of multi-model cross-validation, and meet the needs of real-time detection; The specific steps of S63 are as follows: S631, Independent Inference: Each model performs independent inference operations on the input data based on its own structure and parameters. During the inference process, the inference engine of each model performs forward propagation calculations according to the model's calculation logic, extracts data features, and generates prediction results; S632, result recording: recording the output results of each model, including the posture category and joint angle information predicted by the model, and standardizing the output results for subsequent comparative analysis; The specific steps of S64 are as follows: S641, consistency analysis: compare the prediction results of different models and calculate the similarity or difference between them; S641, Confidence Assessment: First, a confidence score is assigned to the prediction results of each model. The confidence score can be determined based on the model's training accuracy and the model's performance on the validation set. Then, the reliability of each model's prediction results is evaluated by comprehensively considering the consistency and confidence of the prediction results. The specific steps of S65 are as follows: S651, result decision: Based on the comparison and evaluation results, the final posture detection result is selected. When the prediction results of multiple models are highly consistent and the confidence scores meet the preset threshold, the prediction results of the majority model are used as the final output. If the prediction results between the models are significantly different, the exception handling mechanism is activated; S652, model feedback: Feedback the final decision results to each model for subsequent model update and optimization.

[0014] As a preferred solution of the intelligent human motion information collection and real-time posture detection method of the present invention, the specific steps of S9 are as follows: S91, Visual Display: The posture detection results are displayed in the form of an intuitive 3D model, so that users can clearly see the real-time posture of the human body. At the same time, relevant motion data and physiological data charts can be displayed to facilitate users to fully understand the exercise situation; S92, user feedback collection: setting up a user feedback portal on the visual display interface to collect users’ evaluation and suggestions on the test results; S93, automatic report generation: Based on the posture detection and analysis results, as well as user feedback information, the system automatically generates a detailed exercise report, which includes exercise data statistics, posture evaluation results, and exercise suggestions.

[0015] Compared with existing technologies: 1. High Precision: Multi-sensor fusion combined with advanced deep learning algorithms fully utilizes information from different data sources to improve the accuracy of posture detection. Compared with traditional single sensors or algorithms, it can more accurately capture human motion details and posture changes. 2. Real-time performance: Through optimized data processing and efficient model calculations, the system achieves real-time collection of human motion information and real-time detection of posture, meeting the requirements of application scenarios with high real-time performance, such as sports training and security monitoring. 3. Convenience: The wearable sensor is small, lightweight, and comfortable to wear, without interfering with normal human movement. The markerless, non-invasive detection method of the RGB-D camera avoids the cumbersome donning process of traditional motion capture equipment, making it more convenient to use. 4. Multi-field applicability: This solution has broad application potential in multiple fields such as medical care, sports, smart home, and security. It can provide personalized solutions for users in different fields and promote the intelligent development of various fields. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below.

[0017] The present invention provides a method for collecting intelligent human motion information and detecting real-time posture, which includes the following specific steps: S1, data collection phase: first wear the IMU sensor on the key parts of the human body, and then reasonably arrange the position of the RGB-D camera; The specific steps of S1 are as follows: S11, sensor deployment: First, wear IMU sensors on key parts of the human body, including the head, torso, and limb joints to ensure accurate collection of motion information from each part. Then, wear corresponding physiological monitoring sensors as needed. S12, Camera setup: Arrange the RGB-D camera position appropriately to ensure that the human body's motion range can be fully captured to obtain high-quality images and depth information; S2, preliminary data verification: perform preliminary verification on the collected raw data to check whether there are any abnormal values ​​or missing data in the sensor data, whether the camera image is clear and complete, and whether the data timestamps of each device are synchronized; By setting up preliminary data verification, potential problems in the data collection process can be discovered in advance, avoiding inaccurate analysis results due to data quality issues in the subsequent processing stage, and reducing the waste of time and resources caused by invalid data processing; S3, Data Augmentation: Perform data augmentation operations on the collected image data and sensor data. For image data, the dataset is expanded by rotating, scaling, cropping, and adding noise. For sensor data, the diversity of the data is increased by translating and stretching the time series data. By setting up data augmentation, we can increase the scale and diversity of training data, improve the generalization ability of deep learning models, and enable them to maintain high detection accuracy in the face of complex and changing real-world scenarios; S4, data processing and feature extraction: first process the sensor data, then process the computer vision data; The specific steps of S4 are as follows: S41, sensor data processing: First, the raw data collected by the IMU is filtered to remove noise interference. Then, the accelerometer and gyroscope data are fused using the Kalman filter algorithm to improve the accuracy of attitude solution. Finally, key features are extracted from the physiological monitoring sensor data. S42, Computer Vision Data Processing: First, use image segmentation and edge detection algorithms to pre-process the images acquired by the RGB-D camera to extract the human body contours and key points. Then, use deep learning algorithms to classify and match the key points to determine the exact location of various body parts. S5, model prediction: The processed sensor data and computer vision features are input into the trained deep learning model so that the model outputs the current posture information of the human body. This enables the model to track human movement in real time and dynamically monitor posture changes. The specific steps of S5 are as follows: S51, data preprocessing and integration: first regularize the sensor data, then perform data fusion; The specific steps of S51 are as follows: S511, sensor data regularization: time synchronization and scale calibration of sensor data; S512, data fusion: fusing the processed sensor data and visual data to form a unified input feature vector; S52, model loading and deployment: first load the pre-trained model, then deploy the inference engine; The specific steps of S52 are as follows: S521, loading pre-trained model: first load the pre-trained deep learning model into memory, and then load it through the model management interface to ensure the integrity and correctness of the model file to avoid prediction failure due to model damage; S522, deploying the inference engine: First, select an appropriate inference engine based on the model type and structure to optimize the model and improve inference speed and efficiency. Then, deploy the loaded model to the inference engine and configure the corresponding operating parameters. S53, real-time reasoning and prediction: data input first, followed by model calculation and result analysis; The specific steps of S53 are as follows: S531, Data Input: The integrated input feature vector is first formatted according to the requirements of the inference engine and then input into the model. During the real-time detection process, new motion data is acquired at regular intervals and input into the model for inference in a timely manner. S532, Model Operation: The inference engine performs forward propagation calculations on the input data based on the model structure and parameters. During this process, its convolutional neural network extracts and abstracts image features, and its recurrent neural network or long short-term memory network models and analyzes time series data, thereby obtaining the model output results. S533, result analysis: first analyze the output results of the model and convert them into human posture information, then classify and identify the posture according to the preset posture classification standard to determine what posture the current human body movement belongs to; S6, multi-model cross-validation: Introduce multiple auxiliary models with different structures or training focuses, input the same input data into the main model and auxiliary models respectively, and compare the test results of each model. When the results of each model are highly consistent, the reliability of the test results is confirmed. If there are significant differences, the exception handling mechanism is activated and the data and model are further checked; By setting up multi-model cross-validation, the accuracy and credibility of posture detection results can be improved, the risk of misjudgment that may occur in a single model can be reduced, and the stability and robustness of the system can be enhanced; The specific steps of S6 are as follows: S61, multi-model selection and preparation: first perform model screening, then perform model adaptation and model loading; The specific steps of S61 are as follows: S611, Model Screening: Select multiple deep learning models for cross-validation based on their structure, training methods, and focus. These deep learning models include models based on a convolutional neural network architecture to capture static human posture features in images, and models based on recurrent neural networks or long short-term memory networks to analyze motion information over time series. Simultaneously, similar models with different loss functions and training data are included to enrich model diversity. S612, Model Adaptation: To ensure that the input data format and feature dimensions of the selected model are compatible with each other, the model is first fine-tuned to adapt it to the unified input specification; S613, model loading: loading multiple pre-trained models into memory simultaneously; S62, synchronous data input: data integration is performed first, followed by parallel input; The specific steps of S62 are as follows: S621, data integration: the unified input feature vector obtained through data preprocessing and integration is formatted according to the input requirements of each model; S622, Parallel Input: Utilize multi-threaded or distributed computing technology to synchronously input the same input data into each model to fully utilize computing resources, improve the efficiency of model inference, reduce the time overhead of multi-model cross-validation, and meet the needs of real-time detection; S63, multi-model reasoning operation: first perform independent reasoning, then record the results; The specific steps of S63 are as follows: S631, Independent Inference: Each model performs independent inference operations on the input data based on its own structure and parameters. During the inference process, the inference engine of each model performs forward propagation calculations according to the model's calculation logic, extracts data features, and generates prediction results; S632, result recording: recording the output results of each model, including the posture category and joint angle information predicted by the model, and standardizing the output results for subsequent comparative analysis; S64, result comparison and evaluation: first conduct consistency analysis, then conduct confidence assessment; The specific steps of S64 are as follows: S641, consistency analysis: compare the prediction results of different models and calculate the similarity or difference between them; S641, Confidence Assessment: First, a confidence score is assigned to the prediction results of each model. The confidence score can be determined based on the model's training accuracy and the model's performance on the validation set. Then, the reliability of each model's prediction results is evaluated by comprehensively considering the consistency and confidence of the prediction results. S65, decision-making and feedback: first make the result decision, then provide model feedback; The specific steps of S65 are as follows: S651, result decision: Based on the comparison and evaluation results, the final posture detection result is selected. When the prediction results of multiple models are highly consistent and the confidence scores meet the preset threshold, the prediction results of the majority model are used as the final output. If the prediction results between the models are significantly different, the exception handling mechanism is activated; S652, model feedback: Feedback the final decision results to each model for subsequent model update and optimization; S7, Posture Assessment and Warning: This system evaluates detected postures based on preset normal posture ranges and movement patterns. When abnormal postures are detected, it issues a warning signal. It also assesses exercise intensity and fatigue levels based on physiological data, providing recommendations for adjusting exercise strategies. By setting posture assessment and early warning, the model can be continuously optimized over time and with changes in application scenarios, maintaining the accuracy and adaptability of human motion posture detection and extending the service life of the system. S8, Dynamic Model Update: Regularly collect new human motion data, update the trained deep learning model online, and use incremental learning or transfer learning methods to enable the model to continuously learn new motion postures and patterns, adapting to the motion changes of different users and different scenarios; S9, result display and feedback: first visualize the posture detection results and then collect user feedback; The specific steps of S9 are as follows: S91, Visual Display: The posture detection results are displayed in the form of an intuitive 3D model, so that users can clearly see the real-time posture of the human body. At the same time, relevant motion data and physiological data charts can be displayed to facilitate users to fully understand the exercise situation; S92, user feedback collection: setting up a user feedback portal on the visual display interface to collect users’ evaluation and suggestions on the test results; By setting up user feedback collection, the interaction between the system and users can be strengthened, and system functions and detection algorithms can be optimized in a targeted manner based on user feedback, thereby improving user experience and system practicality. S93, automatic report generation: Based on the posture detection and analysis results, as well as user feedback, the system automatically generates a detailed exercise report, which includes exercise data statistics, posture assessment results, and exercise suggestions; By setting up automatic report generation, users can be provided with a systematic and standardized summary of exercise information, saving time in collating and analyzing data, making it easier for users to fully understand their exercise situation and providing strong support for subsequent decision-making.

[0018] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for intelligent human motion information collection and real-time posture detection, characterized in that: The specific steps are as follows: S1, data collection phase: first wear the IMU sensor on the key parts of the human body, and then reasonably arrange the position of the RGB-D camera; S2, preliminary data verification: perform preliminary verification on the collected raw data to check whether there are any abnormal values ​​or missing data in the sensor data, whether the camera image is clear and complete, and whether the data timestamps of each device are synchronized; S3, data enhancement: performing data enhancement operations on the collected image data and sensor data; For image data, the dataset is expanded by rotating, scaling, cropping, and adding noise. For sensor data, the diversity of the data is increased by translating and stretching the time series data. S4, data processing and feature extraction: first process the sensor data, then process the computer vision data; S5, model prediction: The processed sensor data and computer vision features are input into the trained deep learning model so that the model outputs the current posture information of the human body. This enables the model to track human movement in real time and dynamically monitor posture changes. S6, multi-model cross-validation: Introduce multiple auxiliary models with different structures or training focuses, input the same input data into the main model and auxiliary models respectively, and compare the test results of each model. When the results of each model are highly consistent, the reliability of the test results is confirmed. If there are significant differences, the exception handling mechanism is activated and the data and model are further checked; S7, Posture Assessment and Warning: This system evaluates detected postures based on preset normal posture ranges and movement patterns. When abnormal postures are detected, it issues a warning signal. It also assesses exercise intensity and fatigue levels based on physiological data, providing recommendations for adjusting exercise strategies. S8, Dynamic Model Update: Regularly collect new human motion data, update the trained deep learning model online, and use incremental learning or transfer learning methods to enable the model to continuously learn new motion postures and patterns, adapting to the motion changes of different users and different scenarios; S9, result display and feedback: First, visualize the posture detection results and then collect user feedback.

2. The method for collecting intelligent human motion information and detecting real-time posture according to claim 1, characterized in that: The specific steps of S1 are as follows: S11, sensor deployment: First, wear IMU sensors on key parts of the human body, including the head, torso, and limb joints to ensure accurate collection of motion information from each part. Then, wear corresponding physiological monitoring sensors as needed. S12, Camera setup: Arrange the RGB-D camera position reasonably to ensure that the human body's motion range can be fully captured to obtain high-quality images and depth information.

3. The method for collecting intelligent human motion information and detecting real-time posture according to claim 1, characterized in that: The specific steps of S4 are as follows: S41, sensor data processing: First, the raw data collected by the IMU is filtered to remove noise interference. Then, the accelerometer and gyroscope data are fused using the Kalman filter algorithm to improve the accuracy of attitude solution. Finally, key features are extracted from the physiological monitoring sensor data. S42, Computer Vision Data Processing: First, use image segmentation and edge detection algorithms to preprocess the images obtained by the RGB-D camera to extract the human body contours and key points. Then, use deep learning algorithms to classify and match the key points to determine the exact location of various parts of the human body.

4. The method for collecting intelligent human motion information and detecting real-time posture according to claim 1, characterized in that: The specific steps of S5 are as follows: S51, data preprocessing and integration: first regularize the sensor data, then perform data fusion; S52, model loading and deployment: first load the pre-trained model, then deploy the inference engine; S53, real-time reasoning and prediction: data input first, then model calculation and result analysis.

5. The method for collecting intelligent human motion information and detecting real-time posture according to claim 4, characterized in that: The specific steps of S51 are as follows: S511, sensor data regularization: time synchronization and scale calibration of sensor data; S512, data fusion: fusing the processed sensor data and visual data to form a unified input feature vector; The specific steps of S52 are as follows: S521, loading pre-trained model: first load the pre-trained deep learning model into memory, and then load it through the model management interface to ensure the integrity and correctness of the model file to avoid prediction failure due to model damage; S522, deploying the inference engine: First, select a suitable inference engine based on the type and structure of the model to optimize the model and improve the inference speed and efficiency. Then, deploy the loaded model to the inference engine and configure the corresponding operating parameters.

6. The method for collecting intelligent human motion information and detecting real-time posture according to claim 5, characterized in that: The specific steps of S53 are as follows: S531, Data Input: The integrated input feature vector is first formatted according to the requirements of the inference engine and then input into the model. During the real-time detection process, new motion data is acquired at regular intervals and input into the model for inference in a timely manner. S532, Model Operation: The inference engine performs forward propagation calculations on the input data based on the model structure and parameters. During this process, its convolutional neural network extracts and abstracts image features, and its recurrent neural network or long short-term memory network models and analyzes time series data, thereby obtaining the model output results. S533, result analysis: first analyze the output result of the model and convert it into human posture information, then classify and identify the posture according to the preset posture classification standard to determine what posture the current human body movement belongs to.

7. The method for collecting intelligent human motion information and detecting real-time posture according to claim 1, characterized in that: The specific steps of S6 are as follows: S61, multi-model selection and preparation: first perform model screening, then perform model adaptation and model loading; S62, synchronous data input: data integration is performed first, followed by parallel input; S63, multi-model reasoning operation: first perform independent reasoning, then record the results; S64, result comparison and evaluation: first conduct consistency analysis, then conduct confidence assessment; S65, decision-making and feedback: make the result decision first, then provide model feedback.

8. The method for collecting intelligent human motion information and detecting real-time posture according to claim 7, characterized in that: The specific steps of S61 are as follows: S611, Model Screening: Select multiple deep learning models for cross-validation based on their structure, training methods, and focus. These deep learning models include models based on a convolutional neural network architecture to capture static human posture features in images, and models based on recurrent neural networks or long short-term memory networks to analyze motion information over time series. Simultaneously, similar models with different loss functions and training data are included to enrich model diversity. S612, Model Adaptation: To ensure that the input data format and feature dimensions of the selected model are compatible with each other, the model is first fine-tuned to adapt it to the unified input specification; S613, model loading: loading multiple pre-trained models into memory simultaneously; The specific steps of S62 are as follows: S621, data integration: the unified input feature vector obtained through data preprocessing and integration is formatted according to the input requirements of each model; S622, Parallel Input: Utilize multi-threaded or distributed computing technology to synchronously input the same input data into each model to fully utilize computing resources, improve the efficiency of model inference, reduce the time overhead of multi-model cross-validation, and meet the needs of real-time detection; The specific steps of S63 are as follows: S631, Independent Inference: Each model performs independent inference operations on the input data based on its own structure and parameters. During the inference process, the inference engine of each model performs forward propagation calculations according to the model's calculation logic, extracts data features, and generates prediction results; S632, result recording: recording the output results of each model, including the posture category and joint angle information predicted by the model, and standardizing the output results for subsequent comparative analysis; The specific steps of S64 are as follows: S641, consistency analysis: compare the prediction results of different models and calculate the similarity or difference between them; S641, Confidence Assessment: First, a confidence score is assigned to the prediction results of each model. The confidence score can be determined based on the model's training accuracy and the model's performance on the validation set. Then, the reliability of each model's prediction results is evaluated by comprehensively considering the consistency and confidence of the prediction results. The specific steps of S65 are as follows: S651, result decision: Based on the comparison and evaluation results, the final posture detection result is selected. When the prediction results of multiple models are highly consistent and the confidence scores meet the preset threshold, the prediction results of the majority model are used as the final output. If the prediction results between the models are significantly different, the exception handling mechanism is activated; S652, model feedback: Feedback the final decision results to each model for subsequent model update and optimization.

9. The method for collecting intelligent human motion information and detecting real-time posture according to claim 1, characterized in that: The specific steps of S9 are as follows: S91, Visual Display: The posture detection results are displayed in the form of an intuitive 3D model, so that users can clearly see the real-time posture of the human body. At the same time, relevant motion data and physiological data charts can be displayed to facilitate users to fully understand the exercise situation; S92, user feedback collection: setting up a user feedback portal on the visual display interface to collect users’ evaluation and suggestions on the test results; S93, automatic report generation: Based on the posture detection and analysis results, as well as user feedback information, the system automatically generates a detailed exercise report, which includes exercise data statistics, posture evaluation results, and exercise suggestions.