Flying robot control method for monitoring and early warning falling risk of old people

Abstract

The invention relates to a flying robot control method for monitoring and early warning the falling risk of old people, and the method comprises the steps: positioning the position of the old people in each room through an ultra-wideband positioning technology and the track of a flying robot, and carrying out the real-time data collection through combining a camera and a sensor; according to the collected multi-modal data, tumble detection is carried out through a tumble detection algorithm, and an alarm signal is sent out in time; a receiver is installed in each room, indoor navigation and path planning among the rooms are carried out, and the receivers in the rooms are parked for continuous monitoring and automatic charging; gait data is collected and analyzed for a long time based on the collected data, the falling risk is predicted, and personalized health suggestions are provided; and a health report is generated by combining cloud analysis and an artificial intelligence technology. Compared with the prior art, multiple innovative technologies such as the flight control technology, intelligent gait monitoring and falling alarm, automatic following and charging are integrated, and comprehensive monitoring, early warning and emergency response to the health of old people are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring for the elderly, and in particular to a flight robot control method for monitoring and warning the risk of falls for the elderly. Background Art

[0002] In the field of health monitoring for the elderly, the current related technologies have the following defects:

[0003] 1. Limited monitoring range

[0004] Most health monitoring devices rely on wearable devices (such as smart bracelets, health monitoring watches) or fixed devices (such as bedside sensors, floor sensors). These devices can only cover a specific range and are difficult to adapt to the activity needs of the elderly in multiple rooms. That is, they have the following defects:

[0005] Wearable devices require the elderly to wear them actively. Due to the limitations of wearing comfort and compliance, many elderly people are reluctant to wear them or forget to wear them, resulting in incomplete monitoring data.

[0006] Static sensors can only monitor activities in a fixed area, lack flexibility, and cannot follow the movement of the elderly, resulting in insufficient health monitoring coverage. When the elderly leave the monitoring range, the monitoring ability immediately fails.

[0007] 2. Lag in fall detection

[0008] Traditional fall detection systems rely on the elderly pressing an emergency button or triggering device alarms to notify family members or medical institutions. The effectiveness of these systems depends on whether the elderly can actively send signals. Therefore, they have the following defects:

[0009] The elderly may be unable to move or lose consciousness after a fall, resulting in their inability to actively send a distress signal.

[0010] Alarm triggering often depends on data upload and analysis after an event occurs, with network latency and processing latency, resulting in a slow response speed and missing the best rescue opportunity.

[0011] 3. Insufficient intelligence and long-term data analysis

[0012] Most health monitoring devices only provide basic real-time data collection (such as heart rate, steps, etc.), lacking in-depth analysis and intelligent management of long-term health data. Therefore, they have the following defects:

[0013] It is impossible to conduct trend analysis on the health data of the elderly, and it is difficult to discover potential health risks in a timely manner.

[0014] Lacking dynamic learning and personalized model support, it cannot provide personalized health advice based on the personal health history of the elderly.

[0015] Gait monitoring is usually only used for simple data collection and lacks the function of predicting the risk of falls.

[0016] 4. Issues of battery life and convenience

[0017] Existing monitoring devices have limited battery life and require frequent charging for long-term operation. Usually, the charging process requires manual intervention, so there are the following defects:

[0018] Insufficient battery life will cause monitoring interruption, affecting the continuity of health monitoring. The lack of an automated charging and docking mechanism makes it difficult for the device to operate stably for a long time.

[0019] Manual charging increases the complexity of using the device, especially for the elderly or caregivers. Summary of the Invention

[0020] The purpose of the present invention is to overcome the defects of the existing technologies such as limited monitoring range, lagging fall detection, insufficient intelligent analysis ability, and low battery life, and to provide a flight robot control method for monitoring and warning the fall risk of the elderly.

[0021] The purpose of the present invention can be achieved through the following technical solutions:

[0022] A flight robot control method for monitoring and warning the fall risk of the elderly, comprising the following steps:

[0023] Flight robot monitoring and docking step: Based on the ultra-wideband positioning technology and the flight path of the flight robot, locate the position of the elderly in each room, and combine the camera and sensors to collect real-time data for health monitoring;

[0024] Fall detection and emergency response step: Collect multi-modal data for fall detection, perform fall detection through a fall detection algorithm, drive the flight robot to reach the scene after confirming a fall, and send an alarm signal;

[0025] Room following step: Install receivers in each room, and use the ultra-wideband positioning technology combined with the robot path planning algorithm to follow the movement of the elderly for indoor navigation and path planning between rooms;

[0026] Automatic docking and wireless charging step: When the flight robot enters the corresponding room, dock at the receiver in the room to continue monitoring, and automatically charge on the receiver in the room where the flight robot is located when it completes the monitoring task or runs out of power;

[0027] Gait analysis and health risk prediction steps: Long-term collection and analysis of gait data based on data collected by cameras and sensors, using artificial intelligence algorithms to predict fall risks and provide personalized health advice;

[0028] Long-term health data analysis and personalized advice steps: Generate health reports based on the long-term collected data of the flying robot, combined with cloud analysis and artificial intelligence technology, and provide health advice for the elderly.

[0029] Furthermore, the ultra-wideband positioning technology in the flying robot monitoring and docking steps is specifically:

[0030] Deploy multiple ultra-wideband positioning base stations in the room, each base station covering a certain area until all rooms are covered; when the elderly move between different rooms, the time difference of arrival technology of ultra-wideband positioning signals is used to locate the position of the elderly in each room in combination with the flight track of the flying robot.

[0031] Furthermore, the processing process of the fall detection algorithm in the fall detection and emergency response steps is specifically:

[0032] Collect multi-modal data for fall detection, which includes gait interruption signals and image anomaly detection signals. The gait interruption signals are used to detect abnormal behaviors of the elderly when walking, and the abnormal behaviors include sudden stops and abnormal accelerations of the steps; the image anomaly detection signals are used to perform anomaly detection on the collected walking images of the elderly through image recognition algorithms, and the anomaly detection includes the detection of the elderly falling or showing signs of falling;

[0033] Jointly analyze the collected multi-modal data through multi-modal data fusion technology to obtain the final fall detection result;

[0034] When a fall is detected, control the flying robot to fly to the side of the elderly and activate the alarm.

[0035] Furthermore, the processing process of the robot path planning algorithm adopted in the room following step includes: determining the self-position of the flying robot based on ultra-wideband positioning technology, using the A* algorithm or Dijkstra algorithm for path planning, and dynamically adjusting the path for obstacle avoidance when new obstacles are encountered or the environment changes during the navigation of the flying robot.

[0036] Furthermore, the gait analysis and health risk prediction steps specifically include the following steps:

[0037] Capture the movement process of the elderly through a camera, and extract the positions and movement trajectories of human joint points from the video data captured by the camera based on deep learning-based pose estimation technology to obtain movement information;

[0038] Extract gait features, perform signal filtering, and extract time series features from the extracted motion information;

[0039] Pre - train a machine learning model based on the collected gait feature data and corresponding health status data for health risk prediction;

[0040] Use the trained machine learning model to predict the health risk of the measured gait features and provide personalized health advice according to the prediction results.

[0041] Furthermore, the extracted gait features include walking speed, stride length, gait cycle, gait symmetry, gait stability, and gait balance;

[0042] The walking speed is the number of steps per unit time;

[0043] The stride length is the horizontal displacement of each step;

[0044] The gait cycle is the complete cycle from the starting point of one footstep to the starting point of the next footstep;

[0045] The gait symmetry is the symmetry of the left - and right - foot gaits;

[0046] The gait stability is reflected by calculating the amplitude of the steps and the deviation between steps, reflecting the stability during walking;

[0047] The gait balance is to analyze whether the step distribution of the left and right feet is balanced.

[0048] Furthermore, the signal filtering process includes removing noise from the collected data and normalizing and standardizing the data;

[0049] The process of extracting time series features includes:

[0050] Using the sliding window technique, divide the continuous time series into time segments of a fixed length and extract the walking speed and stride length features for each time segment;

[0051] Convert the signal in the time domain to the frequency domain and extract frequency features for analyzing the frequency characteristics of the steps to identify abnormal gaits.

[0052] Furthermore, the machine learning model is a support vector machine model, a decision tree and random forest model, a K - nearest neighbor model, a neural network model, or a long short - term memory network model;

[0053] In the process of constructing the training data of the machine learning model, label data is constructed according to the health status data. The label data is a binary classification label of whether there is a health risk or a multi - classification label divided into mild health risk, moderate health risk, and severe health risk;

[0054] During the training process of the machine learning model, grid search or random search is used for parameter tuning.

[0055] Furthermore, the processing process of the long-term health data analysis and personalized advice step is specifically as follows:

[0056] Trend analysis step: The collected time series data is segmented using a sliding window. Each window contains a certain number of time step data. By calculating the data statistical features within each window, the trend features of each window are extracted;

[0057] The historical data collected is weighted and averaged, giving greater weight to the most recent data, thereby highlighting recent changes in health trends;

[0058] For the data with seasonal fluctuations collected, seasonal decomposition methods are used to extract the corresponding long-term trends and seasonal fluctuations;

[0059] Health prediction step: Based on the trend analysis results obtained from the trend analysis step and the historical data collected, the health risks and future health status of the elderly are predicted. The prediction process includes:

[0060] Using a regression model to predict the health trends of continuous variables;

[0061] Using a classification algorithm to predict future health risks based on the historical health data of the elderly;

[0062] Using time series-based algorithms to predict future health status;

[0063] Using a deep neural network to model the historical data, automatically capturing long-term dependencies, and predicting health trends;

[0064] Personalized health report generation step: Based on the trend analysis results obtained from the trend analysis step and the health prediction results obtained from the health prediction step, a personalized health report is generated. This health report includes a summary of health trends, a health risk assessment, and personalized health advice.

[0065] Furthermore, the summary of health trends includes:

[0066] Overall health trend: Outlining the health status of the elderly over the past few months or years;

[0067] Key health changes: Highlighting changes in the current health status;

[0068] Predicted trend: Based on the health prediction results, showing the health change trend over a future period of time;

[0069] The health risk assessment includes:

[0070] Individual health risk: Assess the individual's health risk based on historical data and prediction results

[0071] Risk level: According to the prediction results of health risks, give a classification of health risks and provide corresponding health advice for each risk level;

[0072] Abnormal warning: For acute health risks, automatically issue a warning and prompt emergency response measures;

[0073] The personalized health advice includes:

[0074] Exercise advice: Recommend a suitable exercise plan according to the individual's health status and exercise trends;

[0075] Diet advice: Provide diet advice according to the health data of the elderly;

[0076] Sleep management: Provide suggestions for improving sleep according to sleep quality data;

[0077] Drug and treatment advice: If the elderly have a history of chronic diseases, provide personalized drug adjustment and treatment advice based on the prediction results of health risks.

[0078] Compared with the prior art, the present invention has the following advantages:

[0079] (1) Comprehensive health monitoring: The present invention enables the flying robot to fly freely between multiple rooms to monitor the health status of the elderly at any time and anywhere; through the receiver at the corner of the roof, the robot covers the entire activity area during monitoring in the room, eliminating the blind spots of traditional static monitoring devices; through the built-in ultra-wideband positioning and flight control technology, the flying robot can accurately locate the position of the elderly and move freely indoors, providing all-round monitoring.

[0080] (2) Real-time fall detection and rapid response: The present invention uses sensor and image analysis technologies to capture fall events using acceleration sensors and cameras, and instantly analyzes the data and judges abnormalities through an embedded processing unit; after fall confirmation, the robot connects with family members or medical institutions through a wireless communication module to ensure the timely transmission of rescue information; it can quickly identify the fall events of the elderly and fly to the side of the elderly for alarm and calling for help in the first time; it realizes emergency response under unattended conditions and can significantly reduce the serious consequences of fall accidents.

[0081] (3) Intelligent gait analysis and health risk prediction: The flying robot of the present invention monitors and analyzes long-term gait data, uses artificial intelligence algorithms to analyze features such as walking speed, stride length, and gait cycle, predicts health risks and gives suggestions. Long-term data accumulation and model optimization make the risk prediction more accurate; it can identify potential health risks, predict through artificial intelligence algorithms, and give early warnings of falls or other health problems; improve the scientificity and accuracy of health management, and help the elderly prevent potential risks.

[0082] (4) Automatic following and intelligent switching between rooms: The present invention uses UWB positioning technology to track the activity trajectory of the elderly in real time, automatically adjusts the path through flight control algorithms, and realizes intelligent navigation; the flying robot can automatically follow when the elderly move to different rooms, and dock through the receiver at the corner of the roof to continue monitoring without manual intervention, which can ensure the continuity of monitoring. Automated movement and docking reduce the operation burden on family members or caregivers.

[0083] (5) Automatic charging and continuous operation: The docking receiver is equipped with a wireless charging module, and the robot automatically docks with the charging interface when docking; the embedded power management system monitors the power in real time and dynamically adjusts the flight tasks and charging priorities of the robot; enables the flying robot to automatically charge when docking in each room, ensuring long-term operation. Even during the charging process, the monitoring function can still work normally, avoiding monitoring interruption, and ensuring the continuity and reliability of the system.

[0084] (6) Emergency alarm and remote rescue connection: When the present invention predicts a fall or health abnormality, it can quickly connect with family members, relatives, communities or medical institutions through the wireless communication module to achieve remote rescue; the information of the emergency notification includes the current location and health status of the elderly, which is convenient for rescue personnel to quickly locate and make decisions.

[0085] (7) Personalized health advice and long-term health management: The present invention uses machine learning algorithms to analyze long-term data, discovers potential health trends and gives suggestions, which can help the elderly manage their health more scientifically and improve their quality of life; provides a precise health intervention plan based on data to reduce long-term health risks. Description of the Drawings

[0086] Figure 1 It is a schematic flow chart of a control method for a flying robot for monitoring and warning the risk of elderly people falling provided in an embodiment of the present invention. Detailed Embodiment

[0087] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention described and illustrated herein can generally be arranged and designed in a variety of different configurations.

[0088] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0089] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not require further definition and explanation in subsequent drawings.

[0090] Embodiment 1

[0091] As Figure 1 shown, this embodiment provides a control method for a flying robot for monitoring and warning of the fall risk of the elderly, including the following steps:

[0092] Flying robot monitoring and docking step S1: Based on the ultra-wideband positioning technology and the flight path of the flying robot, locate the position of the elderly in each room, and combine cameras and sensors to collect real-time data for health monitoring;

[0093] Fall detection and emergency response step S2: Collect multi-modal data for fall detection, perform fall detection through a fall detection algorithm, drive the flying robot to the scene after confirming a fall, and issue an alarm signal;

[0094] Room following step S3: Install receivers in each room, and use ultra-wideband positioning technology combined with a robot path planning algorithm to follow the movement of the elderly for indoor navigation and path planning between rooms;

[0095] Automatic docking and wireless charging step S4: When the flying robot enters the corresponding room, dock at the receiver in the room to continue monitoring, and automatically charge at the receiver in the room where the flying robot is located when it completes the monitoring task or runs out of power;

[0096] Gait analysis and health risk prediction step S5: Based on the data collected by cameras and sensors, perform long-term collection and analysis of gait data, and use artificial intelligence algorithms to predict the fall risk and provide personalized health advice;

[0097] Long-term Health Data Analysis and Personalized Advice Step S6: Generate a health report based on the long-term collected data of the flying robot, combined with cloud analysis and artificial intelligence technology, and provide health advice for the elderly.

[0098] The following is a specific description of each step:

[0099] 1. Flying Robot Monitoring and Docking Step

[0100] The ultra-wideband positioning technology in the flying robot monitoring and docking step is specifically as follows:

[0101] Deploy multiple ultra-wideband positioning base stations in the room, each base station covering a certain area until all rooms are covered; when the elderly move between different rooms, the position of the elderly in each room is located by combining the time difference of arrival technology of ultra-wideband positioning signals with the flight track of the flying robot.

[0102] Specifically, the flying robot monitoring and docking step is used for health monitoring by docking through the receiver at the corner of the roof according to the characteristic that the flying robot has the ability to fly and can automatically move in the indoor environment.

[0103] Technical implementation: Adopt a flight control system and ultra-wideband (UWB) positioning technology, combined with real-time data collection of cameras and sensors, to achieve seamless monitoring of the elderly in multiple rooms.

[0104] Ultra-wideband (UWB) is a high-precision positioning technology that can perform centimeter-level positioning in the indoor environment. Combined with the flight control system, UWB can achieve precise control of real-time positioning of the elderly. The specific details are as follows:

[0105] Multi-point positioning: Deploy multiple UWB positioning base stations in the room, each base station covering a certain area, and real-time track the position changes of the elderly. When the elderly move between different rooms, the UWB system can provide accurate real-time position information, so as to ensure that the monitoring system will not miss any movement of the elderly.

[0106] High-precision indoor positioning: Through the time difference of arrival (TDOA) technology of UWB positioning signals, combined with the flight track of the flight control system, accurately locate the position of the elderly in each room, and cooperate with camera and sensor data to provide accurate health monitoring services for the elderly.

[0107] 2. Fall Detection and Emergency Response Step

[0108] The processing process of the fall detection algorithm in the fall detection and emergency response step is specifically as follows:

[0109] Collect multimodal data for fall detection. The multimodal data includes gait interruption signals and image anomaly detection signals. The gait interruption signals are used to detect abnormal behaviors of the elderly when walking, and the abnormal behaviors include sudden stops and abnormal accelerations of the steps. The image anomaly detection signals are used to perform anomaly detection on the collected walking images of the elderly through image recognition algorithms, and the anomaly detection includes the detection of the elderly falling or showing signs of falling.

[0110] Jointly analyze the collected multimodal data through multimodal data fusion technology to obtain the final fall detection result.

[0111] When a fall is detected, control the flying robot to fly to the side of the elderly and activate the alarm.

[0112] Specifically, the fall detection and emergency response steps are used to be able to detect fall events in real time through the acceleration sensor and the camera, and quickly fly from the receiver to the side of the elderly for emergency response.

[0113] Technical implementation: The fall detection algorithm combines multimodal sensor data, activates flight control after fall confirmation, and accurately locates and arrives at the scene.

[0114] Multimodal sensor data collection

[0115] The multimodal data for fall detection comes from different sensors to ensure the efficient identification and confirmation of fall events in different environments.

[0116] Gait interruption signal: The gait interruption signal comes from the built-in sensor or gait analysis through the camera. When the gait of the elderly suddenly interrupts, loses balance or stops, it may imply the occurrence of a fall event. The system can detect abnormal behaviors when walking, such as sudden stops or abnormal accelerations of the steps, through the gait analysis algorithm.

[0117] Image anomaly detection: Combine the surveillance camera to collect the image data of the elderly in real time, and use computer vision technology to analyze the abnormal behaviors in the images. When the camera captures the elderly falling or showing signs of falling, the image recognition algorithm will perform anomaly detection on the images and issue a warning in a timely manner.

[0118] Sensor data fusion: The data from different sensors will be jointly analyzed through multimodal data fusion technology. The data fusion algorithm eliminates noise signals through means such as weighting, filtering, and delay compensation to improve the accuracy of fall detection. Common data fusion methods include Kalman filtering, weighted average method, etc.

[0119] Based on the multimodal fall detection technology of sensors and cameras, a fast response mechanism for controlling the flying robot to fly down from the receiver to the fall position is realized.

[0120] 3. Room following steps

[0121] The processing procedure of the robot path planning algorithm adopted in the room following step includes: determining the position of the flying robot based on ultra-wideband positioning technology, using the A* algorithm or Dijkstra algorithm for path planning, and dynamically adjusting the path to avoid obstacles when new obstacles are encountered or the environment changes during the navigation of the flying robot.

[0122] Specifically, the room following step is used to drive the flying robot to be able to follow the movement of the elderly in real time. When the elderly enter a new room, the robot automatically flies to the target room and parks at the receiver in the new room to continue monitoring.

[0123] Technical implementation: Adopting UWB positioning technology combined with the robot path planning algorithm to achieve indoor navigation and intelligent switching between rooms; coordinating and managing the parking positions through wireless communication between receivers.

[0124] 3.1. Path planning is the core part of realizing indoor navigation. Especially in a complex indoor environment, the robot needs to efficiently find the shortest and safest path. Combining with UWB positioning technology, the robot can accurately know its own position, and thus perform intelligent path planning according to the environmental situation.

[0125] A* algorithm: A* is a heuristic search algorithm suitable for path planning in a known map. It selects the optimal path by calculating the cost of each position (the distance from the starting point to the target and the heuristic estimate from the current position to the target).

[0126] Dijkstra algorithm: Suitable for unknown or changing maps, it ensures the optimality of the path by calculating the shortest path from the starting point to each node.

[0127] 3.2. Path optimization and dynamic adjustment:

[0128] Real-time dynamic adjustment: When the robot encounters new obstacles or the environment changes during navigation, the path planning algorithm will dynamically adjust the path based on UWB positioning information. The robot can avoid collisions while ensuring efficiency.

[0129] Optimization algorithm: For example, using the advantages of the A* algorithm combined with the Dijkstra algorithm can speed up the calculation while ensuring accuracy and adapt to a rapidly changing environment.

[0130] 3.3. Multi-objective path planning: For tasks in a multi-room and complex environment, the robot needs to achieve multi-objective path planning from one room to another. The system can perform real-time path planning according to task requirements (such as visiting multiple target points).

[0131] 4. Automatic parking and wireless charging steps

[0132] The receiver in each room is not only used for the robot to dock, but also has a wireless charging function. When the robot completes the monitoring task or runs out of power, it will automatically return to the receiver for charging.

[0133] During the docking process, the robot can still maintain the monitoring function to ensure that the monitoring is not interrupted.

[0134] The automated charging design achieved thereby ensures that the robot always has enough power to perform the monitoring task, reduces manual intervention, and improves the convenience and reliability of equipment use.

[0135] 5. Gait analysis and health risk prediction steps

[0136] The gait analysis and health risk prediction steps specifically include the following steps:

[0137] Capture the movement process of the elderly through a camera, and extract the positions and movement trajectories of human joint points from the video data captured by the camera based on deep learning-based pose estimation technology to obtain movement information;

[0138] Extract gait features, perform signal filtering processing, and extract time series features from the extracted movement information;

[0139] Pre-train a machine learning model according to the collected gait feature data and corresponding health condition data for health risk prediction;

[0140] Use the trained machine learning model to perform health risk prediction on the measured gait features, and provide personalized health advice according to the prediction results.

[0141] Specifically, the gait analysis and health risk prediction steps are used to enable the flying robot to collect and analyze the gait data of the elderly for a long time, use artificial intelligence algorithms to predict the fall risk and provide personalized health advice.

[0142] Technical implementation: Based on the data collected by the camera and sensors, combined with parameters such as step frequency, step length, and gait cycle, use machine learning algorithms to establish a health risk prediction model; provide personalized health advice through data analysis.

[0143] 5.1. Data collection and sensor selection

[0144] First of all, data collection is the basis of the health risk prediction model, and data is usually collected through a camera

[0145] Function and role: The camera is used to capture the movement process of the user, and extract the gait features of the user through image processing and computer vision technology.

[0146] Application Technology: The pose estimation (PoseNet) technology based on deep learning can extract the positions and movement trajectories of human joint points from video frames. These key points include ankles, knees, hips, shoulders, wrists, etc., helping to identify motion characteristics such as gait cycles and stride lengths.

[0147] Data Extraction: Extract information such as joint angles, pace rhythms, and gait symmetries of users during walking through real-time video streams or static images.

[0148] 5.2. Feature Extraction and Processing

[0149] After data collection, a series of feature extractions and processes are carried out to be input into the machine learning model for training and prediction.

[0150] 5.2.1 Gait Feature Extraction

[0151] According to the data from the camera, the following gait features are extracted:

[0152] Cadence: The number of steps within a unit of time, reflecting the rhythm of the gait.

[0153] Stride Length: The horizontal displacement of each step, reflecting the length of the step.

[0154] Gait Cycle: The complete cycle from the starting point of one footstep to the starting point of the next footstep, including the stance phase and the swing phase.

[0155] Gait Symmetry: The symmetry of the left and right foot gaits. If the steps on one side are significantly larger or smaller than those on the other side, it may indicate a health problem.

[0156] Gait Stability: By calculating the amplitude of the steps, the deviation between steps, etc., it reflects the stability during walking.

[0157] Pace Balance: The balance of the steps, mainly analyzing whether the step distributions of the left and right feet are balanced.

[0158] 5.2.2 Signal Filtering and Processing

[0159] Noise Removal: The data collected by sensors may contain noise. Techniques such as Kalman filtering and low-pass filtering are used to smooth the data and eliminate unnecessary fluctuations.

[0160] Standardization and Normalization: Standardize or normalize different sensor data to ensure that different features are compared on the same scale.

[0161] 5.2.3 Time Series Feature Extraction

[0162] Gait analysis involves the processing of time series data, so time-based features need to be extracted:

[0163] Sliding window: Using the sliding window technique, the continuous time series is segmented into time segments of a fixed length, and features such as step frequency and step length are extracted for each time period.

[0164] Fourier transform: It is used to transform the signal in the time domain to the frequency domain, extract frequency features, and analyze the frequency features of the gait to identify abnormal gaits.

[0165] 5.3. Establishment of machine learning model

[0166] Use the collected gait data and health status data to train a machine learning model for health risk prediction.

[0167] 5.3.1. Model selection

[0168] According to the task requirements, select an appropriate machine learning model:

[0169] Support Vector Machine (SVM): Suitable for binary classification tasks (such as health risk or no risk), and classifies data through support vectors.

[0170] Decision tree and random forest: Suitable for tasks with more non-linear features, can handle complex health data patterns, and are easy to interpret.

[0171] K-Nearest Neighbor (KNN): Predicts an individual's health risk based on gait similarity and can handle data of different gait types.

[0172] Deep Neural Network (DNN): Uses deep neural networks (such as Convolutional Neural Network CNN) to extract more complex gait features from camera images and is suitable for processing high-dimensional and non-linear data.

[0173] Long Short-Term Memory Network (LSTM): Suitable for processing time series data, can capture the long-term dependencies of gait, and is particularly effective for predicting dynamic gait changes.

[0174] 5.3.2. Feature engineering and dataset construction

[0175] Label construction: According to the user's health status (such as whether there is a history of falls, whether there is cardiovascular disease, hypertension, etc.), construct label data. The label can be binary classification (risky / no risk) or multi-classification (mild, moderate, severe health risks).

[0176] Feature selection: Use feature selection methods (such as L1 regularization, information gain, etc.) to select the features that have the greatest impact on the prediction results.

[0177] 5.3.3. Training and validation

[0178] Training set and test set division: The dataset is divided into a training set and a test set, and cross-validation is used to avoid overfitting.

[0179] Model tuning: Use Grid Search or Random Search to tune the hyperparameters of the model and optimize the prediction performance.

[0180] Evaluation metrics: Use metrics such as accuracy, recall, F1-score, and ROC curve to evaluate the prediction ability of the model.

[0181] 5.4 Health risk prediction and personalized health advice

[0182] Based on the trained machine learning model, real-time health risk prediction is carried out, and personalized health advice is provided according to the prediction results.

[0183] 5.4.1 Health risk prediction

[0184] According to the output results of the model (such as health risk categories), provide users with real-time health status assessment:

[0185] High risk: If the gait abnormality is severe and obvious deviations occur in parameters such as walking speed and stride length, it may indicate a decline in motor function or a risk of falling. It is recommended to conduct further physical examinations or professional interventions.

[0186] Medium risk: If the gait is slightly abnormal, it is recommended to regularly monitor the gait and maintain appropriate exercise and physical training.

[0187] Low risk: The gait is stable and the risk is low, but moderate exercise still needs to be maintained to prevent potential health problems.

[0188] 5.4.2 Personalized health advice

[0189] Exercise advice: Provide users with personalized exercise plans based on the gait analysis results. For example, if the walking speed is too low, it is recommended to increase the walking speed training; if the stride length is uneven, it is recommended to strengthen the gait coordination training.

[0190] Health warning: Combine heart rate, gait, and historical health data to give early warnings of possible health risks, such as heart problems and fall risks.

[0191] Lifestyle advice: Provide optimization advice on diet, sleep, etc. according to the individual's health status to help users maintain good health.

[0192] 5.5 Emergency alarm and remote rescue connection

[0193] When the robot detects that an elderly person has fallen through sensors and cameras, it immediately flies down from the receiver on the roof and quickly flies to the side of the elderly person.

[0194] The robot activates the emergency call mode, communicates with the elderly through the built-in voice system to confirm their status, and sends an alarm signal to family members, relatives, the community or the hospital through the wireless communication module.

[0195] Technical implementation: Triggered in real time by the alarm module in combination with multi-modal sensor data; The communication module sends the alarm information and health data to the preset contacts through Wi-Fi or 5G network.

[0196] 6. Steps for long-term health data analysis and personalized advice

[0197] The processing process of the steps for long-term health data analysis and personalized advice is specifically as follows:

[0198] Trend analysis step: Use a sliding window to segment the collected time series data. Each window contains a certain number of time step data. By calculating the data statistical features within each window, extract the trend features of each window;

[0199] Perform a weighted average on the collected historical data, giving greater weight to the most recent data, so as to highlight the recent changes in health trends;

[0200] For the collected data with seasonal fluctuations, use the seasonal decomposition method to extract the corresponding long-term trend and seasonal fluctuations;

[0201] Health prediction step: Based on the trend analysis results obtained from the trend analysis step and the collected historical data, predict the health risks and future health status of the elderly. The prediction process includes:

[0202] Use a regression model to predict the health trend of continuous variables;

[0203] Use a classification algorithm to predict future health risks based on the historical health data of the elderly;

[0204] Use a time series-based algorithm to predict the future health status;

[0205] Adopt a deep neural network to model the historical data, automatically capture long-term dependencies, and predict the health trend;

[0206] Personalized health report generation step: Generate a personalized health report according to the trend analysis results obtained from the trend analysis step and the health prediction results obtained from the health prediction step. This health report includes a summary of health trends, a health risk assessment, and personalized health advice.

[0207] Specifically, based on the long-term monitoring data accumulation of the flying robot, the long-term health data analysis and personalized advice step generates a health report by combining cloud analysis and artificial intelligence technology, and provides health advice exclusive to the elderly.

[0208] Technical implementation: The data is uploaded to the cloud through the wireless module; AI algorithms are used for trend analysis and health prediction, and personalized health reports are generated in combination with the historical data of the elderly.

[0209] 6.1. Trend analysis

[0210] The purpose of trend analysis is to identify potential long-term change patterns and trends in health data. The main steps are as follows:

[0211] Sliding window technique: The time series data is segmented using a sliding window, and each window contains a certain number of time step data. By calculating the data statistical features (such as mean, standard deviation, maximum value, minimum value, etc.) within each window, the trend features of each window are extracted.

[0212] Weighted moving average method (WMA): The historical data is weighted and averaged, with greater weight given to the most recent data, thus highlighting recent changes in health trends.

[0213] Seasonal decomposition: For data with seasonal fluctuations (such as blood pressure, weight, etc.), seasonal decomposition methods (such as STL decomposition) are used to extract the long-term trend, seasonal fluctuations, and residual components.

[0214] Anomaly detection: AI algorithms (such as Isolation Forest, DBSCAN, etc.) are used to detect abnormal fluctuations or abnormal points in health data, and these abnormal points may indicate potential health problems (such as acute disease onset, sharp deterioration of health status, etc.).

[0215] 6.2. Health prediction

[0216] Health prediction is based on the results of trend analysis and historical data to predict the health risks and future health status of the elderly:

[0217] Regression analysis: Regression models (such as linear regression, ridge regression, etc.) are used to predict the health trends of continuous variables. For example, predicting blood glucose levels or weight changes in the next few months.

[0218] Classification prediction: Classification algorithms (such as support vector machines, random forests, XGBoost, etc.) are used to predict future health risks based on the historical health data of the elderly. For example, predicting the probability of cardiovascular disease, diabetes, or falls.

[0219] Time series prediction: Use time series-based algorithms (such as ARIMA, LSTM, etc.) to predict future health status. For example, predict blood pressure trends or gait stability in the next few months.

[0220] Deep learning models: Employ deep neural networks (such as RNN, LSTM) to model historical data, automatically capture long-term dependencies, and predict health trends.

[0221] 6.3. Generation of personalized health reports

[0222] Based on trend analysis and health prediction results, generate personalized health reports for the elderly, which should include the following aspects:

[0223] 6.3.1. Summary of health trends

[0224] Overall health trend: Outline the health status of the elderly in the past few months or years, such as the change trends of major health indicators like weight, blood sugar, and blood pressure.

[0225] Key health changes: Highlight the changes in the current health status, such as whether there are significant fluctuations in certain health indicators (such as heart rate, sleep quality, exercise volume).

[0226] Predicted trend: Based on the prediction results of the AI model, show the health change trends in the future period, such as possible weight changes or blood pressure fluctuations in the next three months.

[0227] 6.3.2. Health risk assessment

[0228] Individual health risk: Evaluate the individual's health risks based on historical data and prediction results. For example, evaluate the occurrence risks of diseases such as heart disease, diabetes, and falls.

[0229] Risk level: According to the output of the AI model, give a classification of health risks (such as low, medium, high risks), and provide corresponding health suggestions for each risk level.

[0230] Abnormal warning: For possible acute health risks (such as sudden heart problems, acute disease attacks), the system automatically issues a warning and prompts emergency response measures.

[0231] 6.3.3. Personalized health suggestions

[0232] Based on trend analysis and health prediction results, provide personalized health management suggestions for the elderly, including:

[0233] Exercise suggestions: According to the individual's health status and exercise trends, recommend suitable exercise plans. For example, if gait analysis shows that the elderly have unsteady steps, recommend gait recovery training; if the weight increases, recommend controlling diet and performing moderate aerobic exercise.

[0234] Dietary advice: Provide scientific dietary advice based on the health data of the elderly, such as blood sugar, blood lipids, weight, etc. For example, provide low-sugar dietary advice for diabetic patients and recommend low-salt diets for hypertensive patients.

[0235] Sleep management: Suggest improving the sleep environment, adjusting the schedule, etc. based on sleep quality data.

[0236] Drug and treatment advice: If the elderly have a history of chronic diseases, the system can provide personalized drug adjustment and treatment advice based on the output of the prediction model.

[0237] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. A control method for a flying robot used to monitor and give early warnings of the risk of the elderly falling, characterized in that, It includes the following steps: Flying robot monitoring and docking step: Based on the ultra-wideband positioning technology and the flight track of the flying robot, locate the position of the elderly in each room, and combine cameras and sensors to collect real-time data for health monitoring; Fall detection and emergency response step: Collect multi-modal data for fall detection, perform fall detection through a fall detection algorithm, drive the flying robot to the scene after confirming a fall, and send an alarm signal; Room following step: Install receivers in each room, and use ultra-wideband positioning technology combined with a robot path planning algorithm to follow the movement of the elderly for indoor navigation and path planning between rooms; Automatic docking and wireless charging step: When the flying robot enters the corresponding room, dock at the receiver in the room to continue monitoring, and automatically charge on the receiver in the room where the flying robot is located when it completes the monitoring task or runs out of power; Gait analysis and health risk prediction step: Based on the data collected by cameras and sensors, perform long-term collection and analysis of gait data, and use artificial intelligence algorithms to predict fall risks and provide personalized health advice; Long-term health data analysis and personalized advice step: Generate a health report based on the long-term collected data of the flying robot, combined with cloud analysis and artificial intelligence technology, and provide health advice for the elderly.

2. The flight robot control method for monitoring and warning the fall risk of the elderly according to claim 1, wherein The ultra-wideband positioning technology in the flying robot monitoring and docking step is specifically: Arrange multiple ultra-wideband positioning base stations in the room, each base station covers a certain area until all rooms are covered; when the elderly move between different rooms, use the time difference of arrival technology of ultra-wideband positioning signals, combined with the flight track of the flying robot, to locate the position of the elderly in each room.

3. A control method for a flying robot for monitoring and warning of the risk of elderly people falling according to claim 1, characterized in that, The processing process of the fall detection algorithm in the fall detection and emergency response step is specifically: Collect multi-modal data for fall detection, which includes a gait interruption signal and an image anomaly detection signal. The gait interruption signal is used to detect abnormal behaviors when the elderly walk, and the abnormal behaviors include sudden stops and abnormal accelerations of the steps; the image anomaly detection signal is used to perform anomaly detection on the collected walking images of the elderly through an image recognition algorithm, and the anomaly detection includes the detection of the elderly falling or showing signs of falling; Jointly analyze the collected multi-modal data through multi-modal data fusion technology to obtain the final fall detection result; When a fall is detected, control the flying robot to fly to the side of the elderly and activate the alarm.

4. A control method for a flying robot for monitoring and warning of the risk of elderly people falling according to claim 1, characterized in that, The processing process of the robot path planning algorithm used in the room following step includes: Determine the own position of the flying robot based on ultra-wideband positioning technology, use the A* algorithm or Dijkstra algorithm for path planning, and dynamically adjust the path to avoid obstacles when new obstacles are encountered or the environment changes during the navigation of the flying robot.

5. A control method for a flying robot for monitoring and warning of the risk of elderly people falling according to claim 1, characterized in that, The gait analysis and health risk prediction step specifically includes the following steps: Capture the movement process of the elderly through a camera, and use deep learning-based pose estimation technology to extract the positions and movement trajectories of human joint points from the video data captured by the camera to obtain movement information; Extract gait features, perform signal filtering processing, and extract time series features from the extracted motion information; Pre-train a machine learning model based on the collected gait feature data and corresponding health status data for health risk prediction; Use the trained machine learning model to predict the health risk of the measured gait features and provide personalized health advice according to the prediction results.

6. A control method for a flying robot for monitoring and warning of the risk of elderly people falling according to claim 5, characterized in that, The extracted gait features include walking speed, stride length, gait cycle, gait symmetry, gait stability, and gait balance; The walking speed is the number of steps per unit time; The stride length is the horizontal displacement of each step; The gait cycle is the complete cycle from the starting point of one footstep to the starting point of the next footstep; The gait symmetry is the symmetry of the left and right foot gaits; The gait stability reflects the stability during walking by calculating the amplitude of the steps and the deviation between steps; The gait balance analyzes whether the step distribution of the left and right feet is balanced.

7. A control method for a flying robot for monitoring and warning of the risk of elderly people falling according to claim 5, characterized in that, The signal filtering processing includes noise removal from the collected data and standardization and normalization of the data; The process of extracting time series features includes: Using the sliding window technique, dividing the continuous time series into time segments of a fixed length, and extracting the walking speed and stride length features for each time period; Converting the signal in the time domain to the frequency domain and extracting frequency features for analyzing the frequency characteristics of the steps to identify abnormal gaits.

8. A flight robot control method for monitoring and warning of the risk of elderly people falling according to claim 5, characterized in that, The machine learning model is a support vector machine model, a decision tree and random forest model, a K-nearest neighbor model, a neural network model, or a long short-term memory network model; In the process of constructing the training data of the machine learning model, label data is constructed according to the health status data, and the label data is a binary classification label of whether there is a health risk, or a multi-classification label divided into mild health risk, moderate health risk, and severe health risk; Grid search or random search is used during the training process of the machine learning model for parameter tuning.

9. A flight robot control method for monitoring and warning the fall risk of the elderly according to claim 1, characterized in that The processing process of the long-term health data analysis and personalized advice step is specifically as follows: Trend analysis step: Use a sliding window to segment the collected time series data. Each window contains a certain number of time step data. By calculating the data statistical features within each window, extract the trend features of each window; Perform a weighted average on the collected historical data, giving greater weight to the most recent data, so as to highlight the recent changes in health trends; For the collected data with seasonal fluctuations, use the seasonal decomposition method to extract the corresponding long-term trend and seasonal fluctuations; Health prediction step: Based on the trend analysis results obtained in the trend analysis step and the collected historical data, predict the health risks and future health status of the elderly. The prediction process includes: Use a regression model to predict the health trend of continuous variables; Use a classification algorithm to predict future health risks based on the historical health data of the elderly; Use a time series-based algorithm to predict the future health status; Use a deep neural network to model the historical data, automatically capture long-term dependencies, and predict the health trend; Steps for generating a personalized health report: Based on the trend analysis results obtained from the trend analysis step and the health prediction results obtained from the health prediction step, generate a personalized health report, which includes a summary of health trends, a health risk assessment, and personalized health recommendations.

10. A control method for a flying robot for monitoring and warning of the risk of elderly people falling according to claim 9, characterized in that, The summary of health trends includes: Overall health trend: Outlines the health status of the elderly over the past few months or years; Key health changes: Highlights the changes in the current health status; Predicted trend: Based on the health prediction results, shows the health change trend over a period of time in the future; The health risk assessment includes: Individual health risk: Evaluates the individual's health risk based on historical data and prediction results Risk level: According to the prediction results of health risks, gives the grading of health risks and provides corresponding health recommendations for each risk level; Abnormal warning: For acute health risks, automatically issues a warning and prompts emergency response measures; The personalized health recommendations include: Exercise recommendations: Based on the individual's health status and exercise trends, recommends a suitable exercise plan; Dietary recommendations: Provides dietary recommendations based on the health data of the elderly; Sleep management: Based on the sleep quality data, provides suggestions for improving sleep; Drug and treatment recommendations: If the elderly have a history of chronic diseases, provides personalized drug adjustment and treatment recommendations based on the prediction results of health risks.

Images