Gait analysis-based obesity patient automatic identification method and application

Abstract

The invention belongs to the technical field of medical image analysis and intelligent diagnosis, and particularly relates to an obesity patient automatic identification method based on gait analysis and application. According to the method, YOLOv11 is adopted to carry out pedestrian detection, and BOT-SORT is adopted to carry out tracking; human body key points are extracted by using MMPose, and gait and facial features are calculated based on the extracted human body key points; and establishing an obesity classification model by adopting a decision tree classifier, inputting the gait and facial features in the obesity classification model, and outputting a prediction result of whether the individual is an obesity patient or not. According to the method, a computer vision technology and a deep learning method are utilized, detection, skeleton key point extraction and gait and facial feature analysis are carried out on a target individual under the non-contact and wearing equipment-free condition, and automatic classification is carried out through a machine learning model, so that efficient and accurate obesity screening is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image analysis and intelligent diagnosis, and particularly relates to an automatic recognition method and application for obesity patients based on gait analysis. Background Art

[0002] Obesity is a chronic disease that seriously affects global public health and is closely related to various health problems such as cardiovascular diseases, diabetes, and osteoarticular diseases. Traditional obesity diagnosis methods mainly rely on biometric indicators such as body mass index or perform body fat analysis through imaging. However, these methods have certain limitations, including: Limitations of the BMI index: The BMI is calculated only by the ratio of body weight to height and cannot accurately distinguish muscle mass from fat mass, thus may lead to misjudgment. For example, individuals with a high muscle mass may be misjudged as obese, while some people with a high body fat percentage but normal BMI may be overlooked.

[0003] Imaging methods are costly and difficult for large-scale screening: Although imaging methods such as MRI, CT, and DXA can provide accurate body fat distribution information, they are costly, complex to operate, and limited by the availability of equipment and professionals, and are not suitable for large-scale screening.

[0004] In recent years, gait analysis, as a non-invasive, remote, and scalable diagnostic tool, has received increasing attention in the field of healthcare. Existing research has shown that obesity patients have specific movement patterns in gait, such as reduced walking speed, smaller step length, abnormal gait angle changes, knee varus or valgus, etc. Therefore, gait feature analysis can assist in judging whether an individual has obesity.

[0005] Currently, gait-based health analysis mainly relies on wearable sensors or computer vision technology for research. The former requires users to wear additional devices, which affects natural gait and has problems with data synchronization and inconvenient sensor wearing. While computer vision-based methods have been widely used in gait analysis, existing research still has the following deficiencies: (1) Lack of in-depth mining of complex gait features: Existing methods mostly focus on basic gait parameters such as walking speed, walking frequency, and step length, while paying less attention to finer-grained gait features such as joint angles, centroid sway, and knee valgus, which affects the accuracy of obesity recognition.

[0006] (2) Difficulty in accurate detection in a multi-person environment: Traditional gait analysis methods usually target single-person videos, and in a multi-person environment, how to accurately extract the gait information of the target individual is still a challenge.

[0007] (3)Insufficient fusion of gait and facial features: Obese patients may also have significant features in facial morphology (such as the facial width-to-height ratio, changes in the jawline, etc.). However, existing gait-based obesity recognition methods often fail to effectively combine gait and facial information, resulting in limited diagnostic accuracy. Summary of the Invention

[0008] In view of the above technical problems, the present invention provides an automatic recognition method and application for obese patients based on gait analysis. The method uses computer vision technology and deep learning methods to detect, extract skeletal key points, analyze gait and facial features of a target individual without contact and without wearing devices, and perform automatic classification through a machine learning model to achieve efficient and accurate obesity screening.

[0009] The present invention is achieved through the following technical solutions: An automatic recognition method for obese patients based on gait analysis, the method comprising: (1) Data collection and processing: Collect videos containing the gait data of subjects, and preprocess the collected videos; (2) Target detection and tracking: For the videos collected and processed in step (1), use YOLOv11 for pedestrian detection to obtain all pedestrians in the video frames, and output the coordinates of the pedestrian detection boxes; use BOT-SORT for multi-target tracking to extract the complete gait sequences of the target individuals; (3) Human key point detection and gait and facial feature extraction: Use MMPose to extract human key points, the key points including body joint key points, hand key points, foot key points and facial key points, and calculate gait and facial features based on the extracted human key points; The gait and facial features include walking speed, gait angle, knee joint angle, hip joint angle, ankle joint angle, centroid swing amplitude, knee valgus angle, facial width-to-height ratio, jawline angle; (4) Establish an obesity classification model: Use a decision tree classifier to establish an obesity classification model, input the gait and facial features into the obesity classification model, and output the prediction result of whether the individual is an obese patient.

[0010] Further, in step (1), a TL-IPC48GW binocular zoom surveillance camera is used for video collection, and the subjects in the collected videos walk more than 3 meters along a fixed route to ensure the integrity of the gait data.

[0011] Further, in step (1), the preprocessing of the collected videos includes: removing blurred frames, removing redundant frames, color normalization, and background interference filtering.

[0012] Further, in step (3), MMPose is used to extract 133 human key points, including 17 body joint key points, 42 hand key points, 6 foot key points, and 68 face key points.

[0013] Further, in step (4), the obesity classification model is a binary decision tree, and the decision tree is used for binary classification of obesity, and the decision tree is a five-layer decision tree; The accuracy rate of the five-layer decision tree reaches 91%, the precision rate is 0.88, the recall rate reaches 1.0, the F1-score reaches 0.94, and the screening accuracy rate of 5-fold cross-validation is 87%.

[0014] Further, in step (4), taking the gait and facial features calculated in step (3) as the input, through layer-by-layer determination, the classification and recognition of the individual's health status are realized.

[0015] Further, the method of layer-by-layer determination specifically includes: (4.1) Root node determination condition: Determine whether the average value of the mandibular angle is less than or equal to 133.445°; If the average value of the mandibular angle ≤ 133.445°, then enter step (4.2) for left subtree determination; If the average value of the mandibular angle > 133.445°, enter step (4.3) for right subtree determination; (4.2) Determination condition: Determine whether the maximum value of the mandibular angle is less than or equal to 132.265°; If the maximum value of the mandibular angle ≤ 132.265°, it is determined as healthy; If the maximum value of the mandibular angle > 132.265°, then enter step (4.21) for the next judgment; (4.21) Determination condition: Determine whether the average value of knee valgus is less than or equal to 1.665°; If the average value of knee valgus ≤ 1.665°, it is determined as healthy; If the average value of knee valgus > 1.665°, then enter step (4.22) for the next judgment; (4.22) Determination condition: Determine whether the average value of the gait angle is less than or equal to 11.27°; If the average value of the gait angle ≤ 11.27°, then enter step (4.23) for the next judgment; If the average value of the gait angle > 11.27°, it is determined as obese; (4.23) Determination condition: Determine whether the mandibular angle is less than or equal to 17.475°; If the mandibular angle range ≤ 17.475°, then enter step (4.24) for the next judgment; If the range of the mandibular angle > 17.475°, it is determined as obese; (4.24) Judgment condition: Determine whether the maximum value of the centroid swing amplitude is less than or equal to 15.265°; If the maximum value of the centroid swing amplitude ≤ 15.265°, it is determined as healthy; If the maximum value of the centroid swing amplitude > 15.265°, it is determined as obese.

[0016] (4.3) Judgment condition: Determine whether the minimum value of the gait angle is less than or equal to 6.535°; If the minimum value of the gait angle > 6.535°, proceed to step (4.31) for the next judgment; if the minimum value of the gait angle ≤ 6.535°, proceed to step (4.32) for the next judgment; (4.31) Judgment condition: Determine whether the knee joint angle is less than or equal to 6.47°; If the knee joint angle ≤ 6.47°, it is determined as healthy, If the knee joint angle > 6.47°, it is determined as obese.

[0017] (4.32) Judgment condition: Determine whether the maximum value of the centroid swing amplitude is less than or equal to 16.655°; If the maximum value of the centroid swing amplitude ≤ 16.655°, proceed to step (4.33) for the next judgment; If the maximum value of the centroid swing amplitude > 16.655°, proceed to step (4.34) for the next judgment; (4.33) Judgment condition: Determine whether the average value of the gait angle is less than or equal to 17.26°; If the average value of the gait angle ≤ 17.26°, it is determined as healthy, If the average value of the gait angle > 17.26°, it is determined as obese; (4.34) Judgment condition: Determine whether the average value of the ankle joint angle is less than or equal to 161.17°; If the average value of the ankle joint angle ≤ 161.17°, proceed to step (4.35) for the next judgment; If the average value of the ankle joint angle > 161.17°, it is determined as obese; (4.35) Determine whether the average value of the hip joint angle is less than or equal to 175.215°; If the average value of the hip joint angle ≤ 175.215°, it is determined as healthy; If the average value of the hip joint angle > 175.215°, it is determined as obese.

[0018] Further, the method further includes step (5) result output: automatically generating an identification report, where the identification report includes individual gait and facial feature analysis and obesity screening classification results.

[0019] An application of an automatic identification method for obesity patients based on gait analysis, which is used to identify obesity in hospitals, health examination centers, telemedicine, and fitness and rehabilitation scenarios.

[0020] The beneficial technical effects of the present invention: (1) Improving the comprehensiveness of gait feature extraction: Existing gait analysis methods pay insufficient attention to gait features related to obesity. The method provided by the present invention extracts multi-dimensional gait feature parameters (including walking speed, gait angle, knee joint angle, hip joint angle, ankle joint angle, center of mass swing amplitude, knee valgus angle) through human key point detection, depicting the gait features of obesity patients from multiple angles and improving the accuracy of identification.

[0021] (2) Optimizing object detection and tracking in a multi-person environment: Traditional gait analysis is difficult to accurately obtain data of the target individual in a multi-person scenario. The present invention combines object detection and tracking technologies (YOLO + multi-object tracking), uses Ultralytics YOLO for detection and BOT-SORT for tracking to achieve accurate object identification in a multi-person environment, stably track the target individual in a complex environment, ensure the accuracy and integrity of gait data, and improve robustness.

[0022] (3) Fusing gait and facial features to improve the identification effect: In addition to gait features, the present invention introduces facial key point information to obtain facial features (including facial width-to-height ratio, mandibular line angle), and combines gait features and facial features for obesity classification, improving the accuracy of obesity identification. Compared with single-feature detection methods, it improves the identification accuracy and the discrimination ability of the algorithm.

[0023] (4) Constructing an efficient classification model to achieve automatic screening: The present invention uses machine learning methods such as decision trees to classify the extracted gait and facial features, realizing the automatic identification of obesity patients, improving the screening efficiency, and reducing the dependence on manual diagnosis.

[0024] (5) Non-contact screening: The method provided by the present invention does not require wearing equipment and can complete detection only through a camera, improving the screening efficiency, being suitable for large-scale health screenings, and reducing the consumption of medical resources.

[0025] (6) Strong scalability: The method provided by the present invention is applicable to scenarios such as hospitals, physical examination centers, telemedicine, and gyms, can be combined with mHealth for health management, and supports remote monitoring and personalized analysis. Brief Description of the Drawings

[0026] Figure 1 Schematic diagram of facial key points in an embodiment of the present invention; Figure 2 Schematic diagram of body joint key points in an embodiment of the present invention; Figure 3 Schematic diagram of hand key points in an embodiment of the present invention; Figure 4 Schematic diagram of foot key points in an embodiment of the present invention; Figure 5 Schematic diagram of gait angle calculation in an embodiment of the present invention; Figure 6 Schematic diagram of knee joint angle calculation in an embodiment of the present invention; Figure 7 Schematic diagram of hip joint angle calculation in an embodiment of the present invention; Figure 8 Schematic diagram of ankle joint angle calculation in an embodiment of the present invention; Figure 9 Schematic diagram of center of mass swing amplitude calculation in an embodiment of the present invention; Figure 10 Schematic diagram of knee valgus angle calculation in an embodiment of the present invention; Figure 11 Schematic diagram of mandibular line included angle calculation in an embodiment of the present invention; Figure 12 Schematic diagram of facial width-to-height ratio calculation in an embodiment of the present invention. Detailed Description of the Invention

[0027] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0028] On the contrary, the present invention covers any alternatives, modifications, equivalent methods and solutions made within the spirit and scope of the present invention as defined by the claims. Further, in order to enable the public to better understand the present invention, some specific details are described in detail in the following detailed description of the present invention. Those skilled in the art can fully understand the present invention without the description of these details.

[0029] In view of the problem that the health analysis of gait in the existing technology mainly relies on wearable sensors or computer vision technology for research, the present invention combines deep learning object detection, human pose estimation, and gait and facial feature extraction to realize the automatic analysis and classification of gait data, and uses a decision tree algorithm to automatically classify obese patients to achieve non-contact obesity screening; it can be widely applied to fields such as medical health monitoring, disease screening, and intelligent rehabilitation assessment.

[0030] The present invention provides an embodiment of an automatic recognition method for obese patients based on gait analysis. The method includes: (1) Data collection and processing: Collect videos containing the gait data of the subjects, and preprocess the collected videos; specifically, in this step, a TL-IPC48GW binocular zoom surveillance camera is used for video collection, and the subjects in the collected videos walk more than 3 meters along a fixed route to ensure the integrity of the gait data.

[0031] Among them, the camera needs to be installed on the regular gait path of the subjects, such as hospital corridors, walkways in health examination centers, or exercise areas in rehabilitation centers.

[0032] Camera requirements: Maximum image size: 3840×2160, that is, 4K resolution, frame rate 15FPS; Camera angle: Ensure that individuals can be completely photographed within the walking area; Video collection requirements: Multiple people can walk simultaneously within the collection range; the subjects need to walk at least 3 meters along a fixed route to ensure the integrity of the gait data.

[0033] Among them, preprocessing the collected videos includes: removing blurred frames (using motion blur detection methods to eliminate low-quality frames), removing redundant frames (cropping irrelevant video parts and only retaining the gait data of the target individuals). The above preprocessing can improve the robustness of subsequent detections.

[0034] (2) Object detection and tracking: For the videos collected and processed in step (1), use YOLOv11 for pedestrian detection to obtain all pedestrians in the video frames, and output the coordinates of the pedestrian detection boxes; use BOT-SORT for multi-object tracking to extract the complete gait sequences of the target individuals; Specifically, since multiple people can walk simultaneously within the collection range, it is first necessary to detect the target individuals in the video frames and track them to ensure that the gait data of the target individuals can be accurately extracted in the subsequent steps. In order to accurately detect the target individuals in a multi-person scenario, the present invention uses Ultralytics YOLO (YOLOv11) for pedestrian detection to obtain the positions of the target individuals in the video frames; this model has the characteristics of high precision and low latency and can accurately detect human contours in a multi-person environment (adapt to a multi-person environment).

[0035] All pedestrians in the video frames are obtained through Ultralytics YOLO, and the coordinates of the detection boxes are output. Input of Ultralytics YOLO: video frames; Output: all pedestrian detection boxes (x, y, w, h) and confidence scores in each frame; The specific process is as follows: Read each frame image of the gait video; Use YOLO for human detection and output the bounding boxes of each individual; Filter out targets with too small detection boxes or low confidence to reduce false detections.

[0036] The present invention uses BOT-SORT for multi-object tracking to ensure stable tracking of target individuals in a multi-person environment and extract complete gait sequences.

[0037] In the present invention, BOT-SORT mainly includes trajectory prediction, data association, and identity assignment; Trajectory prediction: Use Kalman filter to predict the position of the target in the next frame; Data association: Calculate the IoU between the current frame target and the previous frame trajectory, and combine the Bernoulli Bayesian update mechanism to improve the recovery ability under short-term occlusion; Identity assignment: Assign a unique ID to each detected target to ensure target consistency.

[0038] (3) Human key point detection and gait feature extraction: Use MMPose to extract human key points, which include body joint key points, hand key points, foot key points, and face key points, and calculate gait and face features based on the extracted human key points; The gait and face features include walking speed, gait angle, knee joint angle, hip joint angle, ankle joint angle, centroid swing amplitude, knee valgus angle, face aspect ratio, mandibular line angle; (4) Establish an obesity classification model: Use a decision tree classifier to establish an obesity classification model, input the gait and face features into the obesity classification model, and output the prediction result of whether an individual is an obesity patient.

[0039] In step (3) of this embodiment, MMPose is used to extract 133 human key points, such as Figures 1 - 4 shown, including 17 body joint key points, 42 hand key points, 6 foot key points, and 68 face key points.

[0040] Specifically, the 17 body joint key points include 5 face contour key points, 2 shoulder key points, 2 elbow key points, 2 wrist key points, 2 hip joint key points, 2 knee joint key points, and 2 ankle joint key points.

[0041] 42 hand key points are symmetrically distributed on two hands. Each hand includes 2 wrist key points and 19 knuckle key points (3 for the thumb and 4 for each of the other fingers).

[0042] 6 foot key points are symmetrically distributed on two feet. Each foot includes 1 big toe key point, 1 heel key point and 1 fifth toe key point; 68 face key points include 17 key points distributed on the face contour, 10 key points distributed on the eyebrow bones (symmetrically distributed on both sides of the eyebrow bones), 14 key points distributed on the two eyes (symmetrically distributed on both sides of the eyes), 9 key points distributed on the nose (4 on the nose bridge and 5 below the nostrils), 1 key point distributed between the eyebrows and 17 key points distributed on the edge of the lips In this embodiment, gait and facial features are calculated based on the extracted human key points, where: the walking speed is a basic gait parameter, the gait angle, knee joint angle, hip joint angle, and ankle joint angle are joint angle features, the centroid swing amplitude and knee valgus angle are centroid stability features, and the face width-to-height ratio and mandibular line angle are facial features.

[0043] Among them, the walking speed (steps / second) refers to the number of steps taken per second; the gait angle (°) is used to calculate the average value, maximum value, minimum value, and range of the gait angle within the gait cycle; the knee joint angle (°) is used to detect the maximum and minimum angles and the range of change of the knee during the gait cycle; the hip joint angle (°) is used to calculate the average angle change of the hip joint; the ankle joint angle (°) is used to detect the range of motion of the ankle joint and reflect gait stability. The centroid swing amplitude is used to analyze the lateral and longitudinal offsets of the centroid during walking; knee valgus is used to calculate the knee valgus angle and the range of change, and obese patients usually show a larger knee valgus angle. The face width-to-height ratio is obtained by extracting the face contour through MMPose and calculating the width-to-height ratio; the mandibular line angle and range are obtained by calculating the change in the inclination angle of the mandibular line to assist in identifying obesity characteristics.

[0044] As Figure 5 is a schematic diagram for calculating the gait angle: In gait analysis, the gait angle can reflect the symmetry of a person's gait and the distribution of steps. If the angle is too large or too small, it may indicate gait abnormalities. Generally, it is considered that the gait angle of obese patients is larger than that of normal people when standing with their legs together, and then smaller than that of normal people during walking; Figure 5 In, the dots represent the key points used to calculate the gait angle. The three key points are located at the sacrum, the left ankle joint, and the right ankle joint respectively. The black lines: represent the connection relationships between the key points and are used to calculate the gait angle. Two lines are respectively connected to the ankle joints on both sides to form two sides of the gait angle.

[0045] Figure 6Schematic diagram for knee joint angle calculation; in obese patients, the increased knee joint load may lead to a reduction in the flexion and extension angle of the knee joint. Figure 6 In the figure, the dots represent the key points for calculating the knee joint angle, which are located at the hip joint, knee joint, and ankle joint respectively. The black connecting lines: represent the connection relationships between the key points and are used to calculate the knee joint angle. Taking the calculation of the left knee joint angle as an example: with the left knee joint as the vertex, connect to the left hip joint and the left foot ankle joint respectively to form two sides of the knee joint angle.

[0046] Figure 7 Schematic diagram for hip joint angle calculation; in obese patients, the range of motion of the hip joint may be restricted and the angle change may be small. Figure 7 In the figure, the dots represent the key points for calculating the hip joint angle, which are located at the left hip joint, left knee joint, and left shoulder joint respectively. The black connecting lines: represent the connection relationships between the key points and are used to calculate the hip joint angle. Taking the calculation of the hip joint angle on the left side of the body as an example: with the left hip joint as the vertex, connect the left shoulder joint and the left knee joint respectively to form two sides of the hip joint angle.

[0047] Figure 8 Schematic diagram for ankle joint angle calculation; in obese patients, the range of motion of the ankle joint may be restricted and the angle change may be small. Figure 8 In the figure, the dots represent the key points for calculating the ankle joint angle, which are located at the left knee joint, left foot ankle joint, and left big toe respectively. The black connecting lines: represent the connection relationships between the key points and are used to calculate the ankle joint angle. Taking the calculation of the left foot ankle joint angle as an example: with the left foot ankle joint as the vertex, connect the left knee joint and the left foot toe respectively to form two sides of the ankle joint angle.

[0048] Figure 9 Schematic diagram for calculating the center of mass sway amplitude; the center of mass sway amplitude = |∠1 - ∠2|. In obese patients, the center of mass sway amplitude may be relatively large. Figure 9 In the figure, the dots represent the key points for calculating the center of mass sway amplitude. The three key points are located at the sacrum, left foot ankle joint, and right foot ankle joint respectively. The black connecting lines: represent the connection relationships between the key points and are used to quantify the center of mass sway amplitude. Calculate two angles. The first angle is the included angle formed by the connecting line between the left foot ankle joint and the sacrum and the body midline, which is called ∠1. The second angle is the included angle formed by the connecting line between the right foot ankle joint and the sacrum and the body midline, which is called ∠2. Calculate |∠1 - ∠2|. When healthy volunteers walk, their bodies do not sway too much, and |∠1 - ∠2| is small. When obese patients walk, their body sway amplitude is large, ∠1 and ∠2 change greatly, and |∠1 - ∠2| is large.

[0049] Figure 10Schematic diagram for calculating the genu valgum angle: In the figure, the dots represent the key points for calculating the genu valgum angle, which are located at the right knee joint and the right hip joint respectively. The black connecting lines: represent the connection relationships between the key points and are used to calculate the genu valgum angle. Taking the genu valgum angle on the right side of the body as an example, the genu valgum angle is the included angle formed by the connecting line between the right knee joint and the right hip joint and the midline of the body. Figure 11 Schematic diagram for calculating the mandibular line included angle: Measuring the mandibular angle helps to judge the changes in the facial morphology of obese patients; In the figure, the dots: represent the key points for calculating the mandibular angle, and the three key points are located at the turning points of the right facial mandibular bone contour, the right ear, and the chin respectively. The black connecting lines: represent the connection relationships between the key points and are used to calculate the mandibular line included angle. Taking the calculation of the right mandibular line included angle as an example: Connect the right ear and the chin respectively from the turning point of the right facial mandibular bone contour to form two sides of the mandibular line included angle.

[0050] Figure 12 Schematic diagram for calculating the lower face width-to-height ratio; The face width-to-height ratio of obese patients may be relatively large, showing characteristics of a "round face" or a "wide face". In the figure, the dots: are used to mark the key points for calculating the face width and height, and the four key points are located at the top of the head, the chin, and the left and right ears respectively. The black connecting lines: represent the connection relationships between the key points and are used to calculate the face width-to-height ratio. The face width is determined by the pixel difference between the left and right ears, and the face height is determined by the pixel difference from the top of the head to the chin. The face width-to-height ratio = face width / face height.

[0051] In the present invention, the construction of the obesity classification model includes: Feature dataset: 39 patients (20 obese and 19 healthy) were collected, and a total of 233 samples were extracted, including 145 obese samples and 88 healthy samples, which were used for the training of the machine learning model.

[0052] Decision tree classification: A decision tree model was used for obesity classification, aiming to perform binary classification (obese / healthy) on individuals using gait features. In order to improve the performance and generalization ability of the model, in-depth exploration was carried out in feature engineering and hyperparameter optimization, including feature correlation analysis, feature importance evaluation, and hyperparameter optimization, to ensure that the classifier can accurately capture the gait features of obese patients and improve the stability and reliability of prediction.

[0053] Feature importance evaluation: After screening out the key features, the importance scores of each feature in the decision tree model were further calculated to ensure that the classifier mainly relies on the most discriminative features for decision-making and is not interfered by redundant information.

[0054] In the present invention, the feature importance of the decision tree is calculated through the Gini index, and the specific steps are as follows: Calculate the contribution degree of each feature to the target variable when splitting at each node; Normalize the contribution degrees of all features so that their sum is 1; Select the features with top-ranked feature importance for visualization analysis to understand the decision-making basis of the model.

[0055] The features with the highest importance include the mandibular angle and the gait angle; the importance of the mandibular angle is much higher than that of other features, indicating that the mandibular angle is closely related to obesity recognition. This may be due to the changes in the facial structure of obese individuals, resulting in different mandibular angles. The gait angle reflects the gait angle of leg movement during walking and is related to factors such as gait stability and movement mode. The gait angle of obese individuals may be affected by body mass, showing different gait patterns.

[0056] Other features with high importance include genu valgum, the range of center of mass sway, knee joint angle, and hip joint angle; genu valgum describes the valgus angle of the knee joint and may be related to the gait pattern and joint force of obese individuals, affecting their walking stability; the range of center of mass sway: individuals with a larger range of center of mass sway may show poor gait stability, and obese individuals may have different center of mass control patterns due to different mass distributions; knee joint angle: the knee joint load of obese patients increases, which may lead to a decrease in the flexion and extension angle of the knee joint; hip joint angle: the range of motion of the hip joint of obese patients may be limited, and the angle change may be small.

[0057] Use 5-fold cross-validation to optimize the model performance: To further evaluate the stability of the model, 5-fold cross-validation is used to evaluate the model: The dataset is randomly divided into 5 parts, 4 of which are used for training and 1 for testing, and this is repeated 5 times to ensure the balanced performance of the model on different training / test sets.

[0058] Calculate the accuracy, precision, recall, and F1-score of the model under different folds to evaluate the comprehensive performance of the model. Through the cross-validation results, ensure the robustness of the model and avoid performance fluctuations caused by single data partitioning. Compare decision trees with different depths, and the results are shown in Table 1: Table 1 Comparison of decision tree results with different depths

[0059] According to the comparison results, the present invention selects a 5-layer decision tree, which ensures high accuracy while avoiding the overfitting problem caused by an overly deep tree and can effectively use gait features for classification.

[0060] In step (4), using the gait features calculated in step (3) as the input, through layer-by-layer determination, the classification and recognition of the individual's health status are realized. The specific method of the layer-by-layer determination includes: (4.1) Root node determination condition: Determine whether the average value of the mandibular angle is less than or equal to 133.445°; If the average value of the mandibular angle ≤ 133.445°, then proceed to step (4.2) for the left subtree determination; If the average value of the mandibular angle > 133.445°, proceed to step (4.3) for the right subtree determination; (4.2) Determination condition: Determine whether the maximum value of the mandibular angle is less than or equal to 132.265°; If the maximum value of the mandibular angle ≤ 132.265°, it is determined to be healthy; If the maximum value of the mandibular angle > 132.265°, then proceed to step (4.21) for the next judgment; (4.21) Determination condition: Determine whether the average value of the knee valgus is less than or equal to 1.665; If the average value of the knee valgus ≤ 1.665, it is determined to be healthy; If the average value of the knee valgus > 1.665, then proceed to step (4.22) for the next judgment; (4.22) Determination condition: Determine whether the average value of the gait angle is less than or equal to 11.27°; If the average value of the gait angle ≤ 11.27°, then proceed to step (4.23) for the next judgment; If the average value of the gait angle > 11.27°, it is determined to be obese; (4.23) Determination condition: Determine whether the mandibular angle is less than or equal to 17.475°; If the mandibular angle range ≤ 17.475°, then proceed to step (4.24) for the next judgment; If the mandibular angle range > 17.475°, it is determined to be obese; (4.24) Determination condition: Determine whether the maximum value of the centroid swing is less than or equal to 15.265; If the maximum value of the centroid swing ≤ 15.265, it is determined to be healthy; If the maximum value of the centroid swing > 15.265, it is determined to be obese.

[0061] (4.3) Determination condition: Determine whether the minimum value of the gait angle is less than or equal to 6.535°; If the minimum value of the gait angle > 6.535°, then proceed to step (4.31) for the next judgment; if the minimum value of the gait angle ≤ 6.535°, then proceed to step (4.32) for the next judgment; (4.31) Judgment condition: Determine whether the knee joint angle is less than or equal to 6.47°; If the knee joint angle ≤ 6.47°, it is determined to be healthy, If the knee joint angle > 6.47°, it is determined to be obese.

[0062] (4.32) Judgment condition: Determine whether the maximum value of the center of mass swing amplitude is less than or equal to 16.655; If the maximum value of the center of mass swing amplitude ≤ 16.655, proceed to step (4.33) for the next judgment; If the maximum value of the center of mass swing amplitude > 16.655, proceed to step (4.34) for the next judgment; (4.33) Judgment condition: Determine whether the average gait angle is less than or equal to 17.26°; If the average gait angle ≤ 17.26°, it is determined to be healthy, If the average gait angle > 17.26°, it is determined to be obese; (4.34) Judgment condition: Determine whether the average ankle joint angle is less than or equal to 161.17°; If the average ankle joint angle ≤ 161.17°, proceed to step (4.35) for the next judgment; If the average ankle joint angle > 161.17°, it is determined to be obese; (4.35) Determine whether the average hip joint angle is less than or equal to 175.215°; If the average hip joint angle ≤ 175.215°, it is determined to be healthy; If the average hip joint angle > 175.215°, it is determined to be obese.

[0063] The method further includes step (5) result output: Automatically generate an identification report, and the identification report includes an analysis of individual gait characteristics and the results of obesity screening and classification. Specifically, the results of obesity screening and classification include: a prediction label of "obese" or "healthy" for the subject; the analysis of individual gait characteristics outputs the gait characteristics of the subject and a visualization chart.

[0064] Through feature correlation analysis, feature importance evaluation, hyperparameter optimization, and cross-validation, the present invention improves the performance of the decision tree model in obesity identification. Finally, a screening accuracy rate of 87% is achieved, which is significantly better than traditional discrimination methods. Compared with traditional screening methods based on body mass index, this method can more accurately identify gait abnormalities, improve the reliability and applicability of screening, and provide an efficient and low-cost solution for intelligent healthcare.

[0065] Embodiment 2: Application of an automatic recognition method for obesity patients based on gait analysis. The automatic recognition method for obesity patients based on gait analysis in Embodiment 1 is used for the recognition of obesity in hospitals, health examination centers, telemedicine, and fitness and rehabilitation scenarios.

[0066] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An automatic recognition method for obese patients based on gait analysis, characterized in that, The method includes: (1) Data collection and processing: Collect videos containing the gait data of the subjects, and preprocess the collected videos; (2) Object detection and tracking: For the videos collected and processed in step (1), use YOLOv11 for pedestrian detection to obtain all pedestrians in the video frames, and output the coordinates of the pedestrian detection boxes; use BOT-SORT for multi-object tracking to extract the complete gait sequences of the target individuals; (3) Human key point detection and gait and facial feature extraction: Use MMPose to extract human key points, including body joint key points, hand key points, foot key points and facial key points, and calculate gait and facial features based on the extracted human key points; The gait and facial features include walking speed, gait angle, knee joint angle, hip joint angle, ankle joint angle, centroid swing amplitude, knee valgus angle, face width-to-height ratio, and mandibular line angle; (4) Establish an obesity classification model: Use a decision tree classifier to establish an obesity classification model, input the gait and facial features into the obesity classification model, and output the prediction result of whether the individual is an obesity patient.

2. The automatic recognition method for obese patients based on gait analysis according to claim 1, wherein In step (1), a TL-IPC48GW binocular zoom surveillance camera is used for video collection, and the subjects in the collected videos walk more than 3 meters along a fixed route to ensure the integrity of the gait data.

3. The automatic recognition method for obese patients based on gait analysis according to claim 1, wherein In step (1), the preprocessing of the collected videos includes: removing blurred frames and redundant frames.

4. The automatic recognition method for obesity patients based on gait analysis according to claim 1, characterized in that, In step (3), MMPose is used to extract 133 human key points, including 17 body joint key points, 42 hand key points, 6 foot key points and 68 facial key points.

5. The automatic recognition method for obese patients based on gait analysis according to claim 4, characterized in that In step (4), the obesity classification model is a binary decision tree, and a decision tree is used for binary classification of obesity, and the decision tree is a five-layer decision tree; The accuracy rate of the five-layer decision tree reaches 91%, the precision rate is 0.88, the recall rate reaches 1.0, the F1-score reaches 0.94, and the screening accuracy rate of 5-fold cross-validation is 87%.

6. The automatic recognition method for obese patients based on gait analysis according to claim 5, characterized in that, In step (4), taking the gait and facial features calculated in step (3) as the input, through layer-by-layer determination, the classification and recognition of the individual's health status are realized.

7. The automatic recognition method for obese patients based on gait analysis according to claim 6, characterized in that, The specific method of the layer-by-layer determination includes: (4.1) Root node determination condition: Judge whether the average value of the mandibular angle is less than or equal to 133.445°; If the average value of the mandibular angle ≤ 133.445°, then enter step (4.2) for left subtree determination; If the average value of the mandibular angle > 133.445°, enter step (4.3) for right subtree determination; (4.2) Determination condition: Judge whether the maximum value of the mandibular angle is less than or equal to 132.265°; If the maximum value of the mandibular angle ≤ 132.265°, it is determined to be healthy; If the maximum value of the mandibular angle > 132.265°, then enter step (4.21) for the next judgment; (4.21) Determination condition: Judge whether the average value of the knee valgus is less than or equal to 1.665°; If the average value of the knee valgus ≤ 1.665°, it is determined to be healthy; If the average value of genu valgum > 1.665°, then proceed to step (4.22) for the next judgment; (4.22) Judgment condition: Determine whether the average value of the gait angle is less than or equal to 11.27°; If the average value of the gait angle ≤ 11.27°, then proceed to step (4.23) for the next judgment; If the average value of the gait angle > 11.27°, it is determined as obese; (4.23) Judgment condition: Determine whether the mandibular angle is less than or equal to 17.475°; If the range of the mandibular angle ≤ 17.475°, then proceed to step (4.24) for the next judgment; If the range of the mandibular angle > 17.475°, it is determined as obese; (4.24) Judgment condition: Determine whether the maximum value of the center-of-mass swing amplitude is less than or equal to 15.265°; If the maximum value of the center-of-mass swing amplitude ≤ 15.265°, it is determined as healthy; If the maximum value of the center-of-mass swing amplitude > 15.265°, it is determined as obese; (4.3) Judgment condition: Determine whether the minimum value of the gait angle is less than or equal to 6.535°; If the minimum value of the gait angle > 6.535°, then proceed to step (4.31) for the next judgment; if the minimum value of the gait angle ≤ 6.535°, then proceed to step (4.32) for the next judgment; (4.31) Judgment condition: Determine whether the knee joint angle is less than or equal to 6.47°; If the knee joint angle ≤ 6.47°, it is determined as healthy, If the knee joint angle > 6.47°, it is determined as obese; (4.32) Judgment condition: Determine whether the maximum value of the center-of-mass swing amplitude is less than or equal to 16.655°; If the maximum value of the center-of-mass swing amplitude ≤ 16.655°, then proceed to step (4.33) for the next judgment; If the maximum value of the center-of-mass swing amplitude > 16.655°, then proceed to step (4.34) for the next judgment; (4.33) Judgment condition: Determine whether the average value of the gait angle is less than or equal to 17.26°; If the average value of the gait angle ≤ 17.26°, it is determined as healthy, If the average value of the gait angle > 17.26°, it is determined as obese; (4.34) Judgment condition: Determine whether the average value of the ankle joint angle is less than or equal to 161.17°; If the average value of the ankle joint angle ≤ 161.17°, then proceed to step (4.35) for the next judgment; If the average value of the ankle joint angle > 161.17°, it is determined as obese; (4.35) Determine whether the average value of the hip joint angle is less than or equal to 175.215°; If the average value of the hip joint angle ≤ 175.215°, it is determined as healthy; If the average value of the hip joint angle > 175.215°, it is determined as obese.

8. The automatic recognition method for obesity patients based on gait analysis according to claim 1, characterized in that, The method further includes step (5) Result output: Automatically generate an identification report, and the identification report includes the analysis of individual gait and facial features and the classification results of obesity screening.

9. Application of an automatic recognition method for obesity patients based on gait analysis, characterized in that, Use the method for automatically identifying obese patients based on gait analysis according to any one of claims 1-8 for the identification of obesity in hospitals, health examination centers, telemedicine, and fitness and rehabilitation scenarios.

Images