A high-precision positioning system based on UWB technology

Abstract

The application discloses a high-precision positioning system based on UWB technology, and relates to the field of industrial positioning.The system comprises a three-dimensional positioning module, a posture sensing module and a safety detection module, and high-precision personnel positioning and safety monitoring are realized through the cooperation of the modules.The three-dimensional positioning module utilizes more than four base stations, combines three-side measurement, the least square method and a filtering algorithm to obtain smooth three-dimensional coordinates; the posture sensing module adopts a double-channel analysis strategy, adaptively switches according to the complexity of the environment, and outputs synchronous space-time posture data; the safety detection module realizes high-reliability fall detection and severity judgment through three-layer analysis of threshold rules, multi-mode matching and machine learning, effectively reduces false positives, is suitable for precise positioning and emergency response in complex indoor scenes, deeply meets the complex environment requirements of industrial factories, and realizes all-round guarantee from precise positioning to safety monitoring through the cooperation of the modules and intelligent decision-making.

Description

Technical Field

[0001] This invention relates to the field of industrial positioning, specifically a high-precision positioning system based on UWB technology. Background Technology

[0002] In complex indoor environments such as industrial plants, warehousing and logistics facilities, and nuclear power plant inspections, achieving precise positioning, posture perception, and safety status monitoring of personnel is a core requirement for improving production safety management efficiency and ensuring personnel safety. However, existing indoor positioning and safety monitoring technologies still have many limitations when dealing with such complex scenarios, making it difficult to meet the comprehensive management requirements of high reliability and high precision.

[0003] Currently, mainstream indoor positioning technologies such as Wi-Fi and Bluetooth beacon (BLE) generally suffer from low positioning accuracy (typically 3-5 meters or even higher), susceptibility to signal interference, and severe multipath effects. These problems are particularly pronounced in factory environments filled with large metal equipment and complex building structures, leading to drastic jumps in positioning trajectories and failing to provide a reliable data foundation for personnel scheduling and safety early warning. Although ultra-wideband (UWB) technology improves positioning accuracy to the centimeter level with its high temporal resolution, the raw positioning data still contains "glitch" noise in complex non-line-of-sight and multipath environments. Relying solely on trilateration algorithms makes it difficult to output stable and smooth trajectories, affecting their reliability in practical industrial applications.

[0004] In terms of personnel posture perception and safety status monitoring, existing solutions mostly rely on wearing a single inertial measurement unit (IMU). Common attitude estimation algorithms, such as complementary filtering, perform well in static or simple motion scenarios. However, in industrial environments with persistent vibrations and violent movements, their mechanism of relying on accelerometers to determine the direction of gravity fails, leading to severe distortion in attitude estimation and an inability to accurately identify dangerous situations such as falls.

[0005] Therefore, the industrial sector urgently needs a system that can deeply integrate high-precision positioning, intelligent adaptive attitude perception, and high-reliability safety detection. Summary of the Invention

[0006] The purpose of this invention is to provide a high-precision positioning system based on UWB technology, which can intelligently adjust its algorithm strategy in complex industrial electromagnetic and physical environments. While ensuring the accuracy of positioning and attitude data, it also has the ability to accurately identify and provide graded early warnings for emergency events such as personnel falls, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a high-precision positioning system based on UWB technology, comprising a three-dimensional positioning module, an attitude perception module, and a safety detection module;

[0008] The three-dimensional positioning module is used to acquire the three-dimensional coordinate data of the mobile terminal, including acquiring distance information between the mobile terminal and multiple base stations through UWB ranging, and executing a positioning calculation strategy based on the distance information to acquire the three-dimensional coordinate data of the mobile terminal.

[0009] The attitude perception module is used to acquire and process the attitude data of the mobile terminal, including real-time acquisition of the acceleration and angular velocity data of the mobile terminal, and executing an attitude parsing strategy based on the acquired acceleration and angular velocity data to parse the attitude information data of the mobile terminal. At the same time, it ensures that the attitude information data is synchronized with the timestamp of the three-dimensional coordinate data to construct a three-body data set of time-attitude-coordinate.

[0010] The security detection module is used to analyze the status of the mobile terminal in real time, including acquiring and storing the three-body data group in real time, extracting multiple change features from the three-body data group based on the time series, the multiple change features including altitude change features, attitude angle change features and acceleration features, and simultaneously executing a multi-level analysis strategy to analyze the extracted multiple change features to monitor and identify the status of the mobile terminal in real time.

[0011] Preferably, the positioning calculation strategy includes using a trilateration algorithm or the least squares method to process the distance information to calculate the three-dimensional coordinate data of the mobile terminal, and simultaneously applying a filtering algorithm to smooth the three-dimensional coordinate data.

[0012] Preferably, the attitude analysis strategy includes preprocessing the acquired acceleration and angular velocity data. The preprocessing method includes applying a sliding window mean filter to the acceleration data to eliminate high-frequency noise interference and applying a low-pass filter to the angular velocity data to suppress measurement drift.

[0013] Preferably, the attitude resolution strategy further includes establishing a dual-channel resolution path that includes a first resolution channel for low-precision application scenarios and a second resolution channel for high-precision application scenarios, so as to realize application scenarios with different precision requirements.

[0014] Preferably, the analysis method of the first analysis channel includes calculating the low-frequency components of pitch and roll angles through acceleration data, obtaining the high-frequency components of attitude angles through gyroscope integration, constructing complementary coefficients to fuse the two, and outputting a smooth and responsive attitude estimate.

[0015] The second analytical channel's analytical method includes establishing a state vector containing attitude quaternions and gyroscope bias, and achieving optimal attitude estimation through a prediction update loop. Specifically, the prediction update loop is as follows:

[0016] In the prediction phase, attitude changes are inferred using angular velocity data;

[0017] During the update phase, the gravity direction of the acceleration data is used as the observation value to correct the prediction results.

[0018] Preferably, the multi-level analysis system includes a security threshold rule layer, a pattern matching layer, and a machine learning layer;

[0019] In the security threshold rule layer, a corresponding security physical threshold is set for each single change feature among the multiple change features. When all security physical thresholds are exceeded simultaneously, the process enters the pattern matching layer.

[0020] The pattern matching layer includes constructing a fall scenario pattern library and defining feature patterns for multiple fall scenarios. Each fall scenario pattern is identified through a scene change feature template constructed from the multiple change features. Based on the similarity between the multiple change feature data corresponding to the posture estimation and the scene change feature template, if the similarity exceeds a set threshold, the fall scenario pattern is successfully matched. After successful matching, it enters the machine learning layer for reliability assessment.

[0021] Preferably, the machine learning layer trains a gradient boosting decision tree model based on historical fall data. The training feature set includes height change statistics, attitude angle change trajectory, and acceleration time-domain and frequency-domain features. The model outputs a fall probability score with a value range of 0 to 1. At the same time, the model calculates a confidence index to evaluate the reliability of the current judgment. When the model score exceeds 0.8 and the confidence index is higher than 0.7, it is judged as a reliable fall detection.

[0022] Preferably, the drop scenario modes include a first drop mode, a second drop mode, and a third drop mode:

[0023] The first drop mode is set with a first judgment condition based on altitude change characteristics, attitude angle change characteristics and acceleration characteristics. It is triggered when the altitude drop exceeds a preset altitude threshold H, the pitch angle or roll angle change exceeds a preset angle threshold A, and the resultant acceleration exhibits a characteristic sequence during the descent that is first lower than a preset low acceleration threshold LowG and then higher than a preset high acceleration threshold HighG.

[0024] The second drop mode is set with a second judgment condition based on height change characteristics and acceleration characteristics. It is triggered when the height change speed exceeds a preset speed threshold V and the resultant acceleration exhibits free fall characteristics during the height descent and is subsequently subjected to impact acceleration characteristics.

[0025] The third drop mode is set with a third judgment condition based on the characteristics of angular velocity and acceleration sequence. It is triggered when the angular velocity magnitude exceeds the preset angular velocity threshold and there is a characteristic sequence of acceleration magnitude from free fall state to impact state.

[0026] Preferably, the fall scenario mode also includes a fall escalation mode. The fall escalation mode is set with a fourth judgment condition based on the characteristics of the mobile terminal's position change and posture duration. It is triggered when it is detected that after the height decreases, the person in the low position area continues to exceed a preset time threshold and maintains an abnormal posture state. It is judged as a serious fall that cannot be recovered on its own. The first fall mode, the second fall mode and the third fall mode will automatically enter the fall escalation mode when the fourth judgment condition is met, thus classifying it as a more serious situation.

[0027] Preferably, the region is marked based on the acquired environmental information data. Regions with multiple interference features are marked as complex regions, and regions with a single interference feature are marked as simple regions. The mobile terminal selects the second parsing channel in the complex region and the first parsing channel in the simple region, thereby realizing automatic switching of the parsing channel during movement.

[0028] In summary, the beneficial effects of this invention are:

[0029] This invention leverages the strong penetration and multipath interference resistance of UWB technology, combined with an adaptive attitude analysis algorithm, to automatically switch to a high-precision Kalman filter channel in workshops with numerous metal equipment and complex electromagnetic environments, ensuring stable and reliable attitude data. Its core safety detection module employs a three-level progressive verification mechanism of threshold, pattern, and machine learning. This not only accurately identifies sudden events such as slips and falls, but also intelligently judges the severity of accidents through a "fall escalation" logic, significantly reducing false alarm rates in environments with vibration and high personnel density. Ultimately, it provides highly reliable decision-making support for personnel scheduling, vital sign monitoring, and emergency rescue, effectively improving the efficiency and intelligence of safety management in industrial environments. It deeply aligns with the complex environmental needs of industrial plants, achieving comprehensive protection from precise positioning to safety monitoring through multi-module collaboration and intelligent decision-making. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the process framework structure of a high-precision positioning system based on UWB technology according to the present invention;

[0032] Figure 2This invention provides a flowchart of the attitude perception module technology in a high-precision positioning system based on UWB technology.

[0033] Figure 3 This invention provides a flowchart of the technical content of a safety detection module in a high-precision positioning system based on UWB technology. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0035] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate several embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0036] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0037] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0038] Please see Figure 1 - Figure 3 The present invention provides an embodiment of a high-precision positioning system based on UWB technology, comprising a system terminal and a mobile terminal. The system terminal is equipped with a safety detection module, and the mobile terminal is worn by personnel. The mobile terminal is equipped with a three-dimensional positioning module, an attitude perception module, and a biosensing module. The biosensing module can detect some vital characteristics of the human body.

[0039] The three-dimensional positioning module is used to acquire the three-dimensional coordinate data of the mobile terminal, including obtaining distance information between the mobile terminal and multiple base stations through UWB ranging. At least four non-coplanar base stations are required because there are three unknowns (x, y, z) in three-dimensional space. Each base station provides an equation, and at least four equations are needed to eliminate uncertainties such as clock errors. Based on the distance information, a positioning calculation strategy is executed to obtain the three-dimensional coordinate data of the mobile terminal. The positioning calculation strategy includes using a trilateration algorithm or the least squares method to process the distance information to calculate the three-dimensional coordinate data of the mobile terminal. At the same time, a filtering algorithm is applied to smooth the three-dimensional coordinate data. Trilateration is the theoretical basis, while the least squares method is an enhanced implementation to deal with noise and errors in practical engineering. Typically, the system will first use the trilateration method to find an initial solution, and then use the least squares method for iterative optimization to obtain a more accurate result.

[0040] Directly calculated 3D coordinate data is often full of "gaps," with jumps and noise. This is because radio waves are affected by multipath effects, non-line-of-sight propagation, environmental interference, and other factors. The purpose of filtering algorithms is to suppress noise, smooth trajectories, and predict the true motion state of objects as accurately as possible.

[0041] The attitude perception module is used to acquire and process the attitude data of the mobile terminal, including real-time acquisition of the acceleration and angular velocity data of the mobile terminal, and executing an attitude parsing strategy based on the acquired acceleration and angular velocity data to parse the attitude information data of the mobile terminal. At the same time, it ensures that the attitude information data is synchronized with the timestamp of the three-dimensional coordinate data to construct a three-body data set of time-attitude-coordinate.

[0042] In this embodiment, the attitude analysis strategy includes preprocessing the acquired acceleration and angular velocity data. The preprocessing method includes applying a sliding window mean filter to the acceleration data to eliminate high-frequency noise interference and applying a low-pass filter to the angular velocity data to suppress measurement drift.

[0043] Specifically, the LSM6DS3TR inertial measurement unit can be used, which provides:

[0044] Triaxial accelerometer: measures the resultant acceleration, including gravitational acceleration and motion acceleration;

[0045] A three-axis gyroscope measures the angular velocity of rotation around three coordinate axes.

[0046] The raw data contains noise and must be filtered before use. One method is sliding window mean filtering for the accelerometer: a fixed-length data queue is created. Each time new acceleration data arrives, it is added to the queue, and the oldest data is removed. The arithmetic mean of all data within the window is then calculated as the current output. This effectively smooths out brief, sharp high-frequency noise, such as interference from slight hand tremors or vibrations. It makes the acceleration data curve smoother, facilitating the extraction of stable gravity components.

[0047] Low-pass filtering for gyroscopes allows low-frequency signals to pass through while suppressing high-frequency signals. In code, a first-order infinite impulse response filter is often used. Gyroscopes exhibit zero drift (non-zero output when stationary) and high-frequency measurement noise. Low-pass filtering significantly suppresses this high-frequency noise but introduces a small delay. The parameter α (between 0 and 1) determines the filter's cutoff frequency; the smaller α is, the stronger the filtering effect, but the greater the delay.

[0048] It should be noted that, in this embodiment, the attitude resolution strategy further includes establishing a dual-channel resolution path comprising a first resolution channel for low-precision application scenarios and a second resolution channel for high-precision application scenarios, thereby enabling application to scenarios with different precision requirements. Specifically:

[0049] First analysis channel: In scenarios with limited computing resources or non-strenuous motion, quickly output sufficiently smooth and timely attitude angles;

[0050] The low-frequency components of pitch and roll angles are calculated using acceleration data, and the high-frequency components of attitude angles are obtained by integrating the gyroscope. Complementary coefficients are constructed to fuse the two components, resulting in a smooth and responsive attitude estimate. The low-frequency characteristics of the accelerometer (long-term stability, no drift) are used to correct the high-frequency characteristics of the gyroscope (short-term accuracy, but with drift). The complementary coefficients are generally set to a value close to 1, such as 0.98.

[0051] The second parsing channel provides the optimal and most accurate pose estimation in complex motion scenarios.

[0052] A state vector containing attitude quaternions and gyroscope bias is established, and optimal attitude estimation is achieved through a prediction update loop. The prediction update loop specifically consists of:

[0053] In the prediction phase, attitude changes are inferred using angular velocity data;

[0054] During the update phase, the gravity direction of the acceleration data is used as the observation value to correct the prediction result. The error between the "observed gravity direction" and the "predicted gravity direction" is compared. Combined with the uncertainty (covariance matrix) in the prediction phase, the Kalman gain is calculated. This gain is then used to correct the predicted attitude and update the bias estimate of the gyroscope. The accelerometer acts like a "guide", constantly pulling the gyroscope back from the drift path, thus obtaining a high-precision attitude that is smooth, fast-responding, and stable over a long period of time.

[0055] The security detection module is used to analyze the status of the mobile terminal in real time, including acquiring and storing the three-body data group in real time, extracting multiple change features from the three-body data group based on the time series, the multiple change features including altitude change features, attitude angle change features and acceleration features, and simultaneously executing a multi-level analysis strategy to analyze the extracted multiple change features to monitor and identify the status of the mobile terminal in real time.

[0056] Specifically, the time-attitude-coordinate three-body data set provides a unified spatiotemporal context:

[0057] Time: the foundation of sequence analysis.

[0058] Posture (pitch angle, roll angle, etc.): describes changes in body orientation and angle.

[0059] Coordinates (x, y, z): Especially the height (z), is the key to determining whether a fall has occurred.

[0060] The system calculates the following three core features in real time from the three-body data stream:

[0061] Height change characteristics: Extract the height value (z-axis) from the three-dimensional coordinates, calculate its rate of change (descent speed) and total descent. Falling is essentially a process of rapid descent, which is the most direct characteristic.

[0062] Characteristics of attitude angle changes: Analyze the trajectory, rate of change and final posture of pitch angle (forward / backward) and roll angle (lateral tilt) in time series. Falls are usually accompanied by loss of balance, resulting in a violent and rapid overturning of the trunk angle.

[0063] Acceleration characteristics: Calculate the resultant acceleration magnitude of the triaxial acceleration, analyze its time domain characteristics (such as maximum and minimum values) and frequency domain characteristics (obtain the energy distribution through Fourier transform). The falling process usually conforms to a typical acceleration sequence: weightlessness (free fall) - violent impact (landing) - stillness or weak motion.

[0064] It is worth mentioning that, in this embodiment, the multi-level analysis system includes a security threshold rule layer, a pattern matching layer, and a machine learning layer;

[0065] The security threshold rule layer quickly filters out the vast majority of normal activities, triggering an initial alarm. This is a highly sensitive, low-specificity filter.

[0066] For each individual change feature among the multiple change features, a corresponding security physical threshold is set, for example:

[0067] Height threshold: A drop in height exceeding 0.4 meters within 0.5 seconds.

[0068] Attitude angle threshold: Angle change exceeding 45 degrees within 0.5 seconds.

[0069] Acceleration threshold: The resultant acceleration exceeds 3g (impact) or is less than 0.7g (weightlessness).

[0070] The next pattern matching layer will only be triggered when all thresholds are exceeded simultaneously or in succession within a short period of time. This "AND" logic is crucial to avoid false triggering caused by a single disturbance (such as jumping or bending over quickly).

[0071] The pattern matching layer performs pattern recognition on the triggered alarms to determine whether they conform to known fall patterns. This is a medium-granularity filter that determines whether the current "situation" conforms to a certain known fall pattern. It has a built-in fall scene pattern library that defines feature patterns for multiple fall scenes. Each fall scene pattern is identified through a scene change feature template constructed from the multiple change features. Based on the similarity between the multiple change feature data corresponding to the posture estimation and the scene change feature template, if the similarity exceeds a set threshold, the fall scene pattern is successfully matched. After successful matching, it enters the machine learning layer for reliability assessment.

[0072] The drop scenario modes include a first drop mode, a second drop mode, and a third drop mode:

[0073] The first drop mode is set with a first judgment condition based on altitude change characteristics, attitude angle change characteristics and acceleration characteristics. It is triggered when the altitude drop exceeds a preset altitude threshold H, the pitch angle or roll angle change exceeds a preset angle threshold A, and the resultant acceleration exhibits a characteristic sequence during the descent that is first lower than a preset low acceleration threshold LowG and then higher than a preset high acceleration threshold HighG.

[0074] This is the most intuitive and classic fall pattern, which requires the simultaneous occurrence of three events: a drop in height, a body roll (change in posture angle), and a weightlessness-impact sequence. This pattern has an extremely low false alarm rate because it is difficult to reproduce these three features simultaneously in daily activities.

[0075] The second drop mode is set with a second judgment condition based on height change characteristics and acceleration characteristics. It is triggered when the height change speed exceeds a preset speed threshold V and the resultant acceleration exhibits free fall characteristics during the height descent and is subsequently subjected to impact acceleration characteristics.

[0076] It focuses more on the principles of physics, detecting the rate of change of height (close to the acceleration due to gravity) and typical characteristics of free fall acceleration. This mode is very effective for falls such as slips and straight falls where the change in posture angle is not drastic but conforms to the laws of physics.

[0077] The third drop mode is set with a third judgment condition based on the characteristics of angular velocity and acceleration sequence. It is triggered when the angular velocity modulus exceeds the preset angular velocity threshold and there is a characteristic sequence of acceleration modulus from free fall state to impact state at the same time.

[0078] For falls that may involve rapid spinning or struggling, the analysis combines angular velocity magnitude (reflecting the intensity of rotation) with acceleration sequences to make a judgment. This model can capture some atypical, highly sudden falls.

[0079] The system will calculate the similarity between the multiple variation features extracted in real time and each template in the pattern library (such as Euclidean distance, dynamic time warping DTW). If the similarity exceeds the threshold, the match is successful.

[0080] The machine learning layer performs the final, most intelligent "authenticity" arbitration on the matched patterns, greatly reducing false positives; it is a highly specific filter.

[0081] The gradient boosting decision tree model is trained based on historical fall data. The training feature set includes height change statistics, attitude angle change trajectory, and acceleration time-domain and frequency-domain features. The model outputs a fall probability score, with a value ranging from 0 to 1. At the same time, the model calculates a confidence index to evaluate the reliability of the current judgment. When the model score exceeds 0.8 and the confidence index is higher than 0.7, it is judged as a reliable fall detection.

[0082] It should be noted that in this embodiment, the fall scenario mode also includes fall escalation, which is concerned not only with "whether a fall occurs", but also with "what happens after the fall".

[0083] The fall escalation mode includes a fourth judgment condition based on the mobile terminal's position change and posture duration characteristics. It is triggered when a person remains in a low-position area for more than a preset time threshold after a descent in altitude, maintaining an abnormal posture. This is considered a severe fall from which recovery is impossible. Once "fall escalation" is triggered, it is determined to be a "severe fall, unable to recover independently," requiring immediate medical intervention. This makes the alarm information more valuable, helping rescuers prioritize the most critical situations.

[0084] When the first drop mode, the second drop mode, and the third drop mode meet the fourth judgment condition, they will all automatically enter the drop escalation mode, thus being classified as more serious situations.

[0085] Based on the acquired environmental information data, areas are labeled. Areas with multiple interference features are labeled as complex areas, such as large metal equipment areas in a workshop, metal stairwells and warehouse shelving areas in a workshop, etc. Areas with a single interference feature are labeled as simple areas, such as open offices, residential living rooms, and empty corridors. The mobile terminal selects the second parsing channel in the complex area and the first parsing channel in the simple area, thereby realizing automatic switching of parsing channels during movement.

[0086] In actual operation, within a factory, there are "simple areas" (such as open office areas) and "complex areas" (such as workshops and metal staircases). The system terminal is set as the factory's central monitoring platform, and the mobile terminal is a smartwatch with built-in UWB, LSM6DS3TRIMU, and biosensors.

[0087] Scene 1: Tripping in the workshop

[0088] As employees enter the workshop, their watches continuously monitor environmental information: reduced UWB signal-to-noise ratio, abnormal magnetic field, and increased IMU vibration.

[0089] The system immediately marks the region as a "complex region" and automatically switches the attitude analysis module to the second analysis channel (high-precision Kalman filter) to ensure the stability of attitude data under the worst conditions.

[0090] Fall occurrence and data collection:

[0091] 10:00:00:00: An employee tripped over a ground cable and fell forward abruptly.

[0092] 3D positioning module: The UWB base station uses TDOA ranging and a filtering algorithm to calculate the employee's height coordinates, which drop sharply from 1.2 meters to 0.3 meters within 0.5 seconds.

[0093] Attitude perception module: The Kalman filter channel outputs attitude data, showing that the pitch angle changes from 10° to 85° in 0.5 seconds (body rolls forward), while the triaxial accelerometer records a typical sequence of [resultant acceleration less than 0.8g (weightlessness) - instantaneous impact exceeding 5g].

[0094] Multi-level analysis of the security detection module:

[0095] Safety threshold rule layer: The system detects a drop in altitude > 0.4 meters, a pitch angle change > 45 degrees, and an acceleration sequence that matches the "low-high" characteristic. When all thresholds are exceeded simultaneously, a preliminary alarm is triggered, and the system proceeds to the next layer.

[0096] Pattern matching layer: The system compares real-time features with the pattern library. The matching degree with the first drop pattern (comprehensive features) is as high as 92%, far exceeding the set threshold (such as 80%). If the pattern matching is successful, it proceeds to the final arbitration.

[0097] Machine learning layer: The GBDT model extracts more refined features and calculates a drop probability score of 0.95 with a confidence level of 0.85. Since the score is greater than 0.8 and the confidence level is greater than 0.7, the system determines it as a reliable drop detection.

[0098] Fall escalation check: If the system detects that an employee has been in a lying position for more than 15 seconds in an area with a height of less than 0.5 meters, it will automatically trigger the "fall escalation mode" and mark it as "severe fall, unable to recover on its own".

[0099] System response:

[0100] At 10:00:15.500, the central monitoring platform received the highest level alarm: "An employee has suffered a serious fall in workshop A area, with abnormal vital signs (sudden increase in heart rate), and is suspected of being unable to move. Please provide immediate assistance!" The alarm was accompanied by precise three-dimensional coordinates, and rescue personnel rushed to the scene.

[0101] Scenario 2: Slipping on the stairs

[0102] The employee walks on a metal staircase, and the environment is marked as a "complex area." The attitude resolution uses the second resolution channel.

[0103] The fall occurred when the foot slipped, the buttocks landed directly on the ground, and the person slid down several steps. The body did not twist violently, but the descent was rapid.

[0104] Height change characteristics: The height drops by 1.5 meters in 0.8 seconds, and the descent speed far exceeds the normal threshold for going downstairs.

[0105] Acceleration characteristics: The combined acceleration sequence clearly shows free fall characteristics (approximately 1g) accompanied by multiple step-like impacts.

[0106] Multi-level analysis:

[0107] Threshold layer: The thresholds for altitude change rate and acceleration features are exceeded (attitude angle changes are not significant).

[0108] Pattern matching layer: The system is highly matched with the second drop pattern (free fall physics pattern) (88%).

[0109] Machine learning layer: The model identifies the characteristic sequence of multiple impacts, giving a probability score of 0.87 and a confidence level of 0.80. It is confirmed as a reliable fall.

[0110] Fall escalation: The employee is in pain but conscious and attempts to sit up. The system detects that the "duration of low position" has not exceeded the threshold, so it is not escalated, but it is still marked as a fall event requiring attention.

[0111] Scene 3: Bending down to pick something up in the hallway

[0112] When employees are in an open corridor (simple area), the system uses the first analytical channel (complementary filtering) to save power.

[0113] Action occurs: He quickly bends down to pick up the tool, his height decreases briefly, and his body leans forward.

[0114] Height change: The height decreased from 1.2 meters to 0.7 meters.

[0115] Attitude angle: The pitch angle changes rapidly by approximately 50 degrees.

[0116] Acceleration: There are slight changes in acceleration, but no obvious "weightlessness-impact" sequence.

[0117] Multi-level analysis:

[0118] Safety threshold rule layer: The altitude and attitude angle thresholds were triggered, but the acceleration feature threshold was not triggered (the crucial LowG->HighG sequence is missing). Since not all thresholds were exceeded simultaneously, the alarm was not triggered, and the process terminated here.

[0119] The system intelligently determined that this was normal activity, thus avoiding false alarms.

[0120] Scenario 4: Fainting suddenly in the office

[0121] In the office (simple area), use the first resolution channel.

[0122] Incident: An employee, feeling unwell, slowly slid from their chair to the floor. It was a slow, non-impact fall.

[0123] Altitude change: The altitude drops from 0.9 meters to 0.4 meters within 2 seconds.

[0124] Attitude angle: The roll angle changes slowly, indicating a lateral slip.

[0125] Acceleration: There was no violent impact, but the biosensing module detected an abnormal heart rate.

[0126] Multi-level analysis:

[0127] Threshold layer: Changes in altitude and attitude angle are relatively slow, and may only just reach or not fully reach the threshold. However, combined with abnormal biometric features, the system still triggers an initial alarm.

[0128] Pattern matching layer: has a low degree of direct matching with any pattern.

[0129] Machine Learning Layer: The GBDT model played a crucial role. It integrated data on slow descent, anomalous final posture, and mutated heart rate to calculate a probability score of 0.82. Although the confidence level may be slightly lower (0.65), the system still classified it as a high-risk suspected fall based on the probability score and biological data.

[0130] Fall escalation: The system detected that the employee had been lying still for an extended period of time, which triggered a fall escalation and issued an alarm: "The employee appears to have fainted and fallen in the office area. His vital signs are abnormal and he needs immediate medical attention!"

[0131] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any variations or substitutions conceived without inventive effort should be included within the scope of protection of the invention. Therefore, the scope of protection of the invention should be determined by the scope defined in the claims.

Claims

1. A high-precision positioning system based on UWB technology, comprising a three-dimensional positioning module, an attitude sensing module, and a safety detection module, characterized in that: The three-dimensional positioning module is used to acquire the three-dimensional coordinate data of the mobile terminal, including acquiring distance information between the mobile terminal and multiple base stations through UWB ranging, and executing a positioning calculation strategy based on the distance information to acquire the three-dimensional coordinate data of the mobile terminal. The attitude perception module is used to acquire and process the attitude data of the mobile terminal, including real-time acquisition of the acceleration and angular velocity data of the mobile terminal, and executing an attitude parsing strategy based on the acquired acceleration and angular velocity data to parse the attitude information data of the mobile terminal. At the same time, it ensures that the attitude information data is synchronized with the timestamp of the three-dimensional coordinate data, and constructs a three-body data group of time-attitude-coordinate. The attitude parsing strategy includes establishing a dual-channel parsing path that includes a first parsing channel for low-precision application scenarios and a second parsing channel for high-precision application scenarios, so as to realize the application of different precision requirements. The analysis method of the first analysis channel includes calculating the low-frequency components of pitch and roll angles through acceleration data, obtaining the high-frequency components of attitude angles through gyroscope integration, constructing complementary coefficients to fuse the two, and outputting a smooth and responsive attitude estimate. The analytical method of the second analytical channel includes establishing a state vector containing attitude quaternions and gyroscope bias, and achieving optimal attitude estimation through a prediction update loop. The prediction update loop specifically consists of: in the prediction stage, using angular velocity data to deduce attitude changes; and in the update stage, using the gravity direction of acceleration data as an observation value to correct the prediction result. The security detection module is used to analyze the mobile terminal status in real time, including acquiring and storing the three-body data set in real time, extracting multiple change features from the three-body data set based on time series, the multiple change features including altitude change features, attitude angle change features, and acceleration features, and simultaneously executing a multi-level analysis strategy to analyze the extracted multiple change features to monitor and identify the mobile terminal status in real time. The multi-level analysis system includes a security threshold rule layer, a pattern matching layer, and a machine learning layer; in the security threshold rule layer, a corresponding security physical threshold is set for each individual change feature among the multiple change features, and the process enters the pattern matching layer when all security physical thresholds are exceeded simultaneously. The pattern matching layer includes constructing a fall scenario pattern library and defining feature patterns for multiple fall scenarios. Each fall scenario pattern is identified through a scene change feature template constructed from the multiple change features. Based on the similarity between the multiple change feature data corresponding to the posture estimation and the scene change feature template, if the similarity exceeds a set threshold, the fall scenario pattern is successfully matched. After successful matching, it enters the machine learning layer for reliability assessment.

2. The high-precision positioning system based on UWB technology according to claim 1, characterized in that: The positioning calculation strategy includes using a trilateration algorithm or the least squares method to process the distance information to calculate the three-dimensional coordinate data of the mobile terminal, and applying a filtering algorithm to smooth the three-dimensional coordinate data.

3. A high-precision positioning system based on UWB technology according to claim 2, characterized in that: The machine learning layer trains a gradient boosting decision tree model based on historical fall data. The training feature set includes height change statistics, attitude angle change trajectory, and acceleration time-domain and frequency-domain features. The model outputs a fall probability score and calculates a confidence index to evaluate the reliability of the current fall scenario judgment.

4. A high-precision positioning system based on UWB technology according to claim 3, characterized in that: The drop scenario modes include a first drop mode, a second drop mode, and a third drop mode: The first drop mode is set with a first judgment condition based on altitude change characteristics, attitude angle change characteristics and acceleration characteristics. It is triggered when the altitude drop exceeds a preset altitude threshold, the pitch angle or roll angle change exceeds a preset angle threshold, and the resultant acceleration exhibits a characteristic sequence during the descent that is first lower than a preset low acceleration threshold and then higher than a preset high acceleration threshold. The second drop mode is set with a second judgment condition based on height change characteristics and acceleration characteristics. It is triggered when the height change speed exceeds a preset speed threshold and the combined acceleration exhibits free fall characteristics during the height descent and then impact acceleration characteristics occur. The third drop mode is set with a third judgment condition based on the characteristics of angular velocity and acceleration sequence. It is triggered when the angular velocity magnitude exceeds the preset angular velocity threshold and there is a characteristic sequence of acceleration magnitude from free fall state to impact state.

5. A high-precision positioning system based on UWB technology according to claim 4, characterized in that: The fall scenario mode also includes a fall escalation mode, which is set with a fourth judgment condition based on the characteristics of mobile terminal position change and posture duration. When it is detected that a person has been in a low position area for more than a preset time threshold and has maintained an abnormal posture state after a drop in height, it is triggered and judged as a serious fall that cannot be recovered on its own.

6. A high-precision positioning system based on UWB technology according to claim 5, characterized in that: Based on the acquired environmental information data, regions are marked. Regions with multiple interference features are marked as complex regions, and regions with a single interference feature are marked as simple regions. The mobile terminal selects the second parsing channel in the complex region and the first parsing channel in the simple region, thereby realizing automatic switching of parsing channels during movement.

7. A high-precision positioning system based on UWB technology according to claim 6, characterized in that: The attitude analysis strategy also includes preprocessing the acquired acceleration and angular velocity data. The preprocessing method includes applying a sliding window mean filter to the acceleration data to eliminate high-frequency noise interference and applying a low-pass filter to the angular velocity data to suppress measurement drift.

Images