Garbage collector working track self-adaptive collection method and system based on posture recognition

Abstract

The application discloses a kind of sanitation worker operation track self-adaptive collection method and system based on gesture recognition, it is related to wisdom city management technical field.The method includes: synchronous acquisition IMU inertial data and positioning module position data;Current operation posture category is identified based on inertial data, including bending to clean, arm picking up etc.;According to posture category, dynamically adjust track collection strategy, including positioning data reporting frequency and positioning module work power consumption state;According to strategy, target position data is obtained and is associated with posture label upload;The operation track with state label is generated.The application is linked by posture and power consumption strategy, significantly reduces terminal power consumption, enriches track information dimension, and guarantees blind area track continuity by gesture perception dead reckoning.

Description

Technical Field

[0001] This invention relates to the fields of smart city management and Internet of Things (IoT) technology, and in particular to an adaptive acquisition method and system for the work trajectory of sanitation workers based on posture recognition. Background Technology

[0002] With the increasingly refined development of smart city management, higher demands are being placed on the operation management and performance evaluation of sanitation workers. Current technologies primarily rely on smart terminal devices (such as smart name tags and shoulder lights) integrating satellite positioning modules (e.g., BeiDou / GPS) for real-time positioning and trajectory playback. However, in practical applications, existing technological solutions have the following significant drawbacks: (1) Conflict between power consumption and battery life: In order to obtain continuous operation trajectory, existing terminals usually adopt a fixed high-frequency positioning reporting strategy (such as reporting once every 10-30 seconds). However, satellite positioning modules are high-power components. Long-term high-frequency operation leads to excessive power consumption of the equipment. Sanitation workers need to charge or replace the equipment frequently, which increases management costs and inconvenience.

[0003] (2) The trajectory information is too limited in dimension to reflect the actual workload: The existing trajectory data only contains latitude and longitude coordinates and timestamps, which cannot distinguish whether the worker is in an "effective working state" (such as bending over to sweep or swinging to pick up), "moving state" or "resting state". Managers find it difficult to judge the actual work efficiency of workers through the trajectory, resulting in insufficient assessment basis.

[0004] (3) Intermittent positioning in complex environments: Sanitation work environments are complex, often involving areas with weak or no satellite signals, such as under trees, under overpasses, or indoor rest areas. Existing solutions often stop reporting directly or experience severe positioning drift when the signal is lost, resulting in track breaks and making it impossible to reconstruct the complete work path.

[0005] To address these issues, several improvement solutions have emerged in the industry. Some existing technologies focus on multi-sensor fusion algorithms to improve positioning accuracy, but do not address adjustments to the acquisition strategy based on user behavior. Other existing technologies, while proposing adaptive adjustment of fusion parameters based on user status, have coarse-grained status recognition, distinguishing only macroscopic motion patterns (such as pedestrians / vehicles) without refining to specific operational postures. Furthermore, their adjustments focus on the internal observation model of the algorithm, without addressing the coordinated control of the positioning module's hardware power consumption status (such as sleep / wake-up) and communication reporting frequency.

[0006] Therefore, there is an urgent need for a method and system that can identify the unique working postures of sanitation workers and adaptively adjust hardware power consumption, reporting frequency and positioning fusion strategy accordingly, so as to achieve low power consumption, high information dimension and continuous and complete work trajectory acquisition. Summary of the Invention

[0007] The purpose of this invention is to provide an adaptive acquisition method and system for the work trajectory of sanitation workers based on posture recognition, which aims to solve the technical problems of high terminal power consumption, single trajectory state dimension and trajectory interruption in signal blind spots in the prior art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an adaptive acquisition method for the work trajectory of sanitation workers based on posture recognition, comprising the following steps: The inertial data from the inertial measurement unit (IMU) and the position data from the positioning module are collected synchronously through the terminal device. Based on the inertial data, real-time work posture recognition is performed to obtain the current work posture category, which includes at least one of bending over to sweep, swinging arm to pick up, pushing a cart, walking normally, and standing still / resting. Based on the current working posture category, the trajectory acquisition strategy is dynamically adjusted. The trajectory acquisition strategy includes at least the positioning data reporting frequency and the working power consumption status of the positioning module. The target location data is acquired according to the adjusted trajectory acquisition strategy, and then the target location data is associated with the current operation posture category and uploaded to the cloud platform. A job trajectory with job status tags is generated based on the uploaded data.

[0009] Preferably, the terminal device is a smart name tag or shoulder light terminal device worn on the torso or shoulder of a sanitation worker, integrating the inertial measurement unit (IMU), the positioning module, and the communication module. The positioning module includes a satellite positioning module and an auxiliary positioning module, and the auxiliary positioning module is used to collect auxiliary positioning data from cellular networks and / or Wi-Fi.

[0010] Preferably, the current working posture category is divided into the following three states: The bending over to sweep and the swinging arm to pick up are identified as valid working postures; The normal walking and cart movement are identified as movement postures; The static standing / resting posture is identified as a low-activity posture.

[0011] Preferably, the real-time operation posture recognition based on the inertial data includes a step of extracting differentiated features for sanitation operation-specific movements, wherein the differentiated features include at least one of the following: Differential features used to identify bending over cleaning posture include: acceleration Z-axis dominant frequency, acceleration X-axis and Y-axis variance, and alternating peak values ​​of gyroscope X-axis angular velocity. Differential features used to identify the arm-swinging pickup posture include: short-time pulse characteristics of the gyroscope Z-axis angular velocity and sudden drop characteristics of acceleration magnitude; Differential features used to identify the trolley's movement posture include: Z-axis dominant frequency of acceleration, root mean square of acceleration magnitude, and root mean square of gyroscope Y-axis angular velocity; Differential features used to identify normal walking posture include: cadence, root mean square of acceleration magnitude, and variance of acceleration along the X and Y axes. Differential features used to identify static standing / resting postures include: acceleration magnitude variance and gyroscope three-axis angular velocity magnitude.

[0012] More preferably, the real-time operation posture recognition further includes multi-feature fusion discrimination of the differentiated features through preset temporal logic rules: The rules for identifying the bending-over cleaning posture include: determining whether the dominant frequency of the acceleration along the Z-axis is within the range of 0.8~1.5Hz, and whether the mean variance of the acceleration along the X-axis and Y-axis is less than 0.05m / s². 2 Furthermore, if the number of alternations between positive and negative peak values ​​of the gyroscope's X-axis angular velocity is ≥3 times per window, and if the above conditions are met consecutively for ≥3 sliding windows, it is determined to be a bending-over cleaning posture; The rules for identifying the arm-swinging pickup posture include: when a gyroscope Z-axis angular velocity pulse is detected that meets condition A: amplitude > 50° / s, main frequency > 3Hz, pulse width 0.3~0.8s, and also meets condition B or condition C, it is determined to be an arm-swinging pickup posture; where condition B is: the acceleration magnitude decreases by more than 40% in 0.5 seconds before and after the pulse, and condition C is: there is no periodic Z-axis fluctuation within 1 second after the pulse; The rules for identifying the trolley's movement include: determining whether the Z-axis dominant frequency of acceleration is within the range of 0.5~1.0Hz and whether the root mean square of the acceleration magnitude is within the range of 0.6~1.0m / s². 2 Within the range, and whether the root mean square of the gyroscope's Y-axis angular velocity is within the range of 2~8° / s, when the above conditions are met simultaneously, it is determined to be the trolley's moving posture; The rules for identifying normal walking posture include: determining whether the cadence is within the range of 1.5~2.5Hz and whether the root mean square magnitude of the acceleration is greater than 1.2m / s². 2 Furthermore, are the variances of the acceleration along both the X and Y axes greater than 0.1 m / s²? 2 When the above conditions are met simultaneously, and the conditions for judging the posture of pushing a cart, bending over to sweep, or waving an arm to pick up are not met, it is judged as a normal walking posture. The rules for identifying static standing / resting postures include: determining whether the variance of the acceleration magnitude is less than 0.02 m / s². 2 Furthermore, if the gyroscope's three-axis angular velocity magnitude is less than 5° / s, and the duration of the above conditions being met continuously reaches the first preset time threshold, it is determined to be a static standing / resting posture.

[0013] In this embodiment, the first preset time threshold is preferably 3 seconds. This threshold can be adjusted according to the actual application scenario, for example, set to any value between 2 seconds and 5 seconds, so as to achieve a balance between recognition accuracy and response speed.

[0014] Preferably, the step of dynamically adjusting the trajectory acquisition strategy according to the current working posture category includes establishing a quantitative mapping relationship between posture and localization strategy: The high-frequency mode is defined as follows: the location data reporting frequency is adjusted to the first frequency, and the power consumption state of the location module is adjusted to full power mode. The intermediate frequency mode is defined as follows: the location data reporting frequency is adjusted to a second frequency lower than the first frequency, and the power consumption state of the location module is adjusted to a low power consumption mode. The low-frequency heartbeat mode is defined as follows: the location data reporting frequency is adjusted to a third frequency lower than the second frequency, and the working power consumption state of the location module is adjusted to a deep sleep mode. When the identified working posture is bending over to sweep or swinging an arm to pick up, switch to the high-frequency mode; When the identified working posture is normal walking or pushing a cart, switch to the mid-frequency mode; When the identified working posture is static standing / resting, switch to the low-frequency heartbeat mode.

[0015] More preferably, the reporting interval corresponding to the first frequency is 15 to 30 seconds, the reporting interval corresponding to the second frequency is 1 to 2 minutes, and the reporting interval corresponding to the third frequency is 10 minutes.

[0016] More preferably, the step of dynamically adjusting the trajectory acquisition strategy according to the current working posture category further includes a jitter prevention and hysteresis wake-up mechanism, the mechanism including at least one of the following: Time integration judgment rules: The switch from the current mode to the high-frequency mode is triggered only when the cumulative proportion of bending over to sweep and / or swinging arms to pick up within a 30-second sliding window is ≥70%; the switch to the low-frequency heartbeat mode is triggered only when the continuous duration of a static standing / resting posture is ≥2 minutes. Strategy Oscillation Suppression Rule: When the number of attitude switching times is ≥4 within 10 seconds, the system is forced to enter the frequency stabilization lock state, maintaining the current acquisition strategy unchanged for 30 seconds, and does not respond to any attitude changes during the lock period; Delayed wake-up rule: When switching from the low-frequency heartbeat mode to the medium-frequency mode or high-frequency mode, the positioning module must be woken up only if ≥3 valid working postures are detected continuously and the time interval between two adjacent detections is ≤5 seconds. The positioning module maintains the high-frequency mode for the first 30 seconds after being woken up and does not perform frequency reduction operation.

[0017] Preferably, the step of acquiring target location data according to the adjusted trajectory acquisition strategy includes the sub-steps of multimodal positioning fusion and blind spot compensation: Detect satellite signal strength; When the satellite signal strength is greater than or equal to the signal threshold, the target location data is generated based on the satellite positioning data. When the satellite signal strength is below the signal threshold or is lost, the fusion weights of satellite positioning data, auxiliary positioning data, and dead reckoning data are dynamically adjusted according to the current operational attitude category, and dead reckoning based on inertial data is triggered to fill the trajectory gap in the positioning blind spot.

[0018] More preferably, the step of dynamically adjusting the fusion weights of satellite positioning data, assisted positioning data, and dead reckoning data according to the current operational attitude category includes: When the satellite signal strength is below the signal threshold, the satellite positioning data weight is set to 0, and the weights of the auxiliary positioning data and dead reckoning data are assigned according to the current operational attitude category. If the current working posture category is normal walking or cart movement, set the fusion weight ratio of auxiliary positioning data and dead reckoning data to 0.5:0.5; If the current operation posture category is bending over to sweep or swinging arm to pick up, set the fusion weight ratio of auxiliary positioning data and dead reckoning data to 0.8:0.2; If the current working posture category is stationary standing / resting, set the auxiliary positioning data weight to 1.0 and disable dead reckoning.

[0019] More preferably, the multimodal positioning fusion is implemented using an extended Kalman filter algorithm. The observation noise covariance matrix of the extended Kalman filter is dynamically adjusted according to the current operational attitude category and satellite signal strength to change the fusion weights of satellite positioning data, auxiliary positioning data, and dead reckoning data.

[0020] More preferably, the dead reckoning step size parameter is dynamically adjusted according to the current operational attitude category, including: The step length calculation model is dynamically adjusted based on the real-time identified work posture category. When the walking posture is identified as normal, the stride length L = L0 × (1 + a × (f - f0)), where L0 is the historical statistical base stride length, f is the current stride frequency, f0 is the reference stride frequency (e.g., 1.8 Hz), and a is the adjustment coefficient (e.g., 0.15). When the movement is identified as a pushcart, the step length L = L0 × 0.7; When the posture is identified as bending over to sweep, the step length L = L0 × 0.4; When identified as a stationary standing / resting posture, step length L=0.

[0021] In a preferred embodiment, the historical statistical base step size L0 is dynamically calibrated using positioning results when satellite signals are good.

[0022] More preferably, the dead reckoning further includes attitude-assisted smoothing of the bearing reckoning, including: When the posture is identified as static standing / resting, the static output of the gyroscope is collected as the zero bias value, which is then updated and used for temperature drift compensation in subsequent directional integration. When the posture is identified as bending over to clean, the change in heading angle is linearly extrapolated using the historical directional trend of auxiliary positioning data (base station / Wi-Fi) to shield the short-term drift of the inertial measurement unit; When the movement is identified as normal walking or pushing a cart, the rate of change of heading angle is limited to ≤30° / second.

[0023] In a second aspect, the present invention provides an adaptive acquisition system for the work trajectory of sanitation workers based on posture recognition, used to implement the method described in any one of the first aspects, the system comprising: Terminal equipment, including: An inertial measurement unit is used to collect inertial data from sanitation workers. The positioning module includes a satellite positioning module and an auxiliary positioning module. The power consumption state of the satellite positioning module switches between full power, low power and deep sleep according to control commands. The auxiliary positioning module is used to collect auxiliary positioning data from cellular networks and / or Wi-Fi. The communication module is used for data interaction with the cloud platform; The pose recognition engine, policy management engine, and localization fusion engine are deployed on the terminal device or cloud platform, wherein: An attitude recognition engine is used to identify the current work attitude category based on the inertial data. The current work attitude category includes at least one of bending over to sweep, swinging arm to pick up, pushing a cart, walking normally, and standing still / resting. The strategy management engine is used to dynamically adjust the trajectory acquisition strategy according to the current operation posture category. The strategy includes the positioning reporting frequency and the working power consumption status of the positioning module. A positioning fusion engine is used to acquire target location data according to the trajectory acquisition strategy and trigger dead reckoning based on the current operational attitude category in satellite signal blind zones. The cloud platform is used to receive uploaded target location data and corresponding job posture categories, and generate job trajectories with job status tags.

[0024] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: 1) Significantly reduced power consumption and improved device endurance: This invention utilizes dynamic frequency control based on posture recognition and switching of the positioning module's operating power state. During effective operation (bending over to sweep / sweeping with an arm), a high-frequency full-power mode is used to ensure trajectory accuracy. During movement (normal walking / pushing a cart), a medium-frequency low-power mode is used to balance power consumption and accuracy. During rest (standing still / resting), it switches to a deep sleep mode, sending only heartbeat signals, thus avoiding invalid high-frequency positioning and reporting. Actual testing has verified that the terminal's average daily power consumption is reduced by more than 64% compared to traditional fixed high-frequency reporting solutions, significantly improving the terminal's endurance and addressing the management pain point of frequent charging for sanitation workers.

[0025] 2) Rich trajectory information dimensions to support accurate performance evaluation: The work trajectory output by this invention is accompanied by work status labels such as "bending over to sweep", "swinging arm to pick up", "walking normally", "pushing cart", and "standing still / resting". Managers can intuitively view the effective working time, movement time and rest time of sanitation workers on the management platform to achieve accurate assessment and intelligent scheduling of workload.

[0026] 3) Continuous and complete trajectory in blind zones, adapting to complex operating environments: This invention uses multimodal positioning fusion and attitude-aware dead reckoning to dynamically adjust the fusion weight of auxiliary positioning data and dead reckoning based on the current operating attitude in satellite signal blind zones (such as under trees, under overpasses, and indoor rest areas). It also adjusts the step size model and direction reckoning parameters based on attitude adaptive adjustment, effectively compensating for the trajectory interruption problem in satellite signal blind zones and achieving “seamless” continuous trajectory recording in complex environments.

[0027] 4) High posture recognition accuracy, designed specifically for sanitation scenarios: This invention designs a differentiated feature extraction scheme (such as the Z-axis main frequency for bending over to sweep, the gyroscope pulse for swinging arms to pick up, etc.) and multi-feature fusion discrimination logic for the unique movements of sanitation operations. Validated by actual test datasets, the accuracy of bending over to sweep reaches 93.7%, and the accuracy of swinging arms to pick up reaches 91.2%, which is significantly better than general motion classifiers (general SVM accuracy is only 78%), providing reliable posture input for subsequent strategy adjustments. Attached Figure Description

[0028] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0029] Figure 1A schematic diagram of the system architecture of an adaptive acquisition system for sanitation worker work trajectory based on posture recognition provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall process of an adaptive acquisition method for the work trajectory of sanitation workers based on posture recognition, provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the above and other features and advantages of the present invention clearer, the invention will be further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explanation to those skilled in the art and are exemplary only, not restrictive.

[0031] Example 1: like Figure 1 As shown, this embodiment provides an adaptive acquisition system for sanitation worker work trajectory based on posture recognition, including a terminal device 100, a posture recognition engine 200, a strategy management engine 300, a positioning fusion engine 400, and a cloud platform 500.

[0032] The terminal device 100 is worn on the torso or shoulder of the sanitation worker, and may specifically be a smart name tag or a shoulder light terminal. The terminal device 100 includes: Inertial Measurement Unit (IMU) 110: Used to collect inertial data of sanitation workers in real time, including triaxial acceleration and triaxial angular velocity, with a sampling frequency preferably of 50Hz; Positioning module 120: includes satellite positioning module 121 and auxiliary positioning module 122. The power consumption state of satellite positioning module 121 (such as Beidou / GPS) can be switched between full power, low power and deep sleep according to control commands; auxiliary positioning module 122 is used to collect auxiliary positioning data from cellular network (base station) and / or Wi-Fi; Communication module 130: Used to interact with the cloud platform 500, and upload target location data and operation posture tags.

[0033] Main control unit 140: integrates attitude recognition engine 200, policy management engine 300 and localization fusion engine 400, used to run various algorithm modules and coordinate data flow and control commands.

[0034] The pose recognition engine 200, policy management engine 300, and localization fusion engine 400 run in the main control unit 140, wherein: The posture recognition engine 200 is used to process IMU data, extract differentiated features for sanitation operations, and identify and output the current operation posture category in real time. The strategy management engine 300 is used to dynamically adjust the trajectory acquisition strategy according to the current operation posture category. The strategy includes the positioning data reporting frequency and the working power consumption status of the positioning module 120. The positioning fusion engine 400 is used to acquire target location data according to the trajectory acquisition strategy, and to trigger and optimize dead reckoning based on the current operational attitude category in satellite signal blind spots.

[0035] The three engines mentioned above can also be deployed on the cloud platform 500, or a distributed deployment scheme of terminal and cloud can be adopted. In this embodiment, the three engines run in the main control unit 140 to achieve local real-time decision-making and reduce dependence on cloud computing.

[0036] The cloud platform 500 is used to receive target location data and corresponding work posture categories uploaded by terminal device 100, and generate continuous work trajectories with work status tags for managers to view and evaluate performance.

[0037] After the system powers on, all modules initialize and enter the default low-power mode. At this time, the satellite positioning module 121 performs positioning every minute. Sanitation workers wearing terminal devices 100 begin their work, and the system executes the following closed-loop process in real time: Multi-source data synchronous acquisition; Real-time job posture recognition based on IMU; Attitude-based adaptive acquisition strategy decision-making; Multimodal localization fusion and blind spot compensation; Track data with status labels is reported.

[0038] Example 2: like Figure 2 As shown, this embodiment provides an adaptive acquisition method for the work trajectory of sanitation workers based on posture recognition. Based on the system described in Embodiment 1, it specifically includes the following steps: Step S1: Synchronously collect inertial data from the inertial measurement unit (IMU) and position data from the positioning module through the terminal device.

[0039] After the system is powered on, each module initializes. The IMU continuously acquires three-axis acceleration and three-axis angular velocity data at a frequency of 50Hz; the satellite positioning module operates in low-power mode by default, and satellite positioning module 121 performs positioning once every minute; the cellular / Wi-Fi module is in standby mode. The main control unit 140 synchronously reads the IMU data buffer and the latest calculated coordinates from the positioning module via the bus.

[0040] Step S2: Based on the inertial data, perform real-time operation posture recognition to obtain the current operation posture category.

[0041] The current work posture category includes at least one of bending over to sweep, swinging arms to pick up items, pushing a cart, walking normally, and standing still / resting. The specific identification process includes: Step S21: Data preprocessing and feature extraction.

[0042] The raw IMU data is subjected to DC removal and low-pass filtering. A sliding window is used to segment the data, with a window length of 2 seconds and a step size of 0.5 seconds. Fast Fourier Transform (FFT) is performed on the data within the window to extract frequency domain features.

[0043] Within each sliding window, extract the following differential features for different job posture categories: (1) Regarding the posture of bending over to sweep: Extracting the Z-axis dominant frequency of acceleration: The spectrum is calculated using Fast Fourier Transform (FFT), and the maximum peak frequency in the range of 0.8~1.5Hz is extracted; X-axis and Y-axis acceleration variances: Calculate the variances of the X-axis and Y-axis accelerations within the window respectively, and take the mean; Gyroscope X-axis angular velocity peak alternation characteristic: the number of times positive and negative peaks alternate within the detection window.

[0044] (2) Regarding the arm-swinging picking posture: Gyroscope Z-axis angular velocity short-time pulse: Employing a sliding peak detection algorithm, pulse events with amplitude > 50° / s, pulse width 0.3~0.8s, and main frequency > 3Hz are identified; Acceleration magnitude drop characteristics: Calculate the acceleration magnitude 0.5 seconds before and after the pulse ( ) average, calculate the percentage decrease.

[0045] (3) Regarding the stroller's movement posture: Z-axis acceleration frequency: Extract the maximum peak frequency within the range of 0.5~1.0Hz; Root mean square (RMS) angular velocity of the Y-axis of the gyroscope: characterizes low-frequency swaying of the vehicle body, with a typical value of 2~8° / s; Root mean square of acceleration modulus: The RMS of acceleration modulus within the calculation window, distinguished from normal walking (cart movement <1.0 m / s²). 2 ).

[0046] (4) For normal walking posture: Step frequency: Detect the periodicity of acceleration and extract the step frequency within the range of 1.5~2.5Hz; Root mean square magnitude of acceleration: >1.2 m / s² 2 ; Variance of acceleration along the X and Y axes: >0.1 m / s² 2 .

[0047] (5) For static standing / resting postures: Acceleration magnitude variance: <0.02m / s 2 Lasting more than 3 seconds; The gyroscope's three-axis angular velocity magnitude is <5° / s.

[0048] Step S22: Perform multi-feature fusion discrimination on the differentiated features using preset temporal logic rules.

[0049] To avoid misjudgment based on a single feature, the pose recognition engine 200 employs temporal logic rules for multi-feature fusion and judgment. The specific judgment rules are as follows: (1) Rules for determining bending over to clean: Condition A: The dominant frequency of acceleration along the Z-axis is ∈ [0.8, 1.5 Hz]; Condition B: The mean variance of the acceleration along the X and Y axes is <0.05 m / s². 2 ; Condition C: The number of times the positive and negative peak values ​​of the gyroscope's X-axis angular velocity alternate ≥ 3 times per window; Judgment logic: If A∩B∩C is consecutively true for ≥3 sliding windows, then it is judged as "bending over to clean".

[0050] (2) Arm-swing picking judgment rules: Condition A: A gyroscope Z-axis angular velocity pulse is detected (amplitude > 50° / s, main frequency > 3Hz, pulse width 0.3~0.8s); Condition B: The magnitude of acceleration decreases by more than 40% within 0.5 seconds before and after the pulse; Condition C: No periodic Z-axis fluctuations within 1 second after the pulse; Decision logic: If (A∩B)∪(A∩C) is satisfied, then it is determined as "arm-waving pickup".

[0051] (3) Rules for determining the movement of the cart: Condition A: The dominant frequency of acceleration along the Z-axis is ∈ [0.5, 1.0 Hz]; Condition B: Acceleration magnitude RMS ∈ [0.6, 1.0 m / s²] 2 ]; Condition C: The gyroscope's Y-axis angular velocity RMS ∈ [2, 8° / s]; Decision logic: If A∩B∩C is satisfied, then it is determined as "pushing the cart forward".

[0052] (4) Rules for determining normal walking: Condition A: Step frequency ∈ [1.5, 2.5Hz]; Condition B: Acceleration modulus RMS > 1.2 m / s² 2 ; Condition C: The variance of acceleration along the X and Y axes > 0.1 m / s²2 ; Judgment logic: If A∩B∩C is satisfied and the cart / cleaning characteristics are not satisfied, then it is judged as "normal walking".

[0053] (5) Rules for determining standing / resting: Condition A: Acceleration magnitude variance < 0.02 m / s² 2 ; Condition B: Gyroscope angular velocity magnitude < 5° / s; Condition C: Duration ≥ 3 seconds; Judgment logic: If A∩B∩C is satisfied, then it is judged as "standing still / resting".

[0054] The above feature extraction and discrimination logic is specifically designed for sanitation operation scenarios. Validated by actual test datasets, the accuracy rate for recognizing bending over to sweep reaches 93.7%, and the accuracy rate for recognizing waving arms to pick up items reaches 91.2%, significantly better than general motion classifiers (general SVM accuracy is only 78%).

[0055] Step S3: Dynamically adjust the trajectory acquisition strategy according to the current working posture category.

[0056] This embodiment discloses the attitude-localization strategy quantization mapping rules and the anti-jitter / hysteresis wake-up mechanism to ensure power consumption optimization and strategy stability.

[0057] 3.1 Attitude-Localization Strategy Quantization Mapping The strategy management engine 300 dynamically adjusts the trajectory acquisition strategy based on the current job posture category output by the posture recognition engine 200 (updated every 2 seconds). This embodiment establishes the following mapping relationship (see Table 1): Table 1. Attitude-Localization Strategy Quantization Mapping Table The specific values ​​of the first frequency, the second frequency, and the third frequency can be configured according to the actual application scenario. In this embodiment, the preferred values ​​are: the reporting interval corresponding to the first frequency is 15 to 30 seconds, the reporting interval corresponding to the second frequency is 1 to 2 minutes, and the reporting interval corresponding to the third frequency is 10 minutes.

[0058] 3.2 Anti-shake and delayed wake-up mechanism To avoid frequent policy jumps caused by instantaneous attitude misjudgment, the policy management engine 300 introduces the following mechanism: (1) Time integration determination rules: High-frequency mode trigger: The current mode will only be switched to high-frequency mode when the "bending over to sweep / swinging to pick up" posture accounts for ≥70% of the total within a 30-second sliding window. Low-frequency sleep trigger: The low-power mode will switch to the low-frequency heartbeat mode only when the "standing / resting" posture lasts for ≥2 minutes.

[0059] (2) Strategy Oscillation Suppression Rules: If the attitude changes ≥ 4 times within 10 seconds, the system will be forced into a frequency stabilization lock state: the current acquisition strategy will remain unchanged for 30 seconds, and no attitude changes will be responded to during the lock period to avoid strategy oscillations caused by noise or brief actions.

[0060] (3) Delayed wake-up rule: When recovering from low-frequency heartbeat mode to medium / high-frequency mode, the satellite positioning module must be activated only after ≥3 consecutive detections of valid operating attitude, with the time interval between two adjacent detections being ≤5 seconds. For the first 30 seconds after activation, the satellite positioning module maintains high-frequency mode and does not perform frequency reduction operations to ensure continuous and reliable trajectory in the initial segment.

[0061] The above mechanism has been verified on the low-power ARM Cortex-M4 platform. Compared with the solution without anti-jitter strategy, the number of strategy switching is reduced by 72% and the average power consumption is reduced by 18%.

[0062] Step S4: Obtain target location data according to the adjusted trajectory acquisition strategy, associate the target location data with the current operation posture category, and upload it to the cloud platform.

[0063] This embodiment discloses an adaptive adjustment mechanism for attitude-aware dead reckoning parameters to solve the problems of poor accuracy and easy divergence in general INS reckoning in sanitation scenarios.

[0064] 4.1 Adaptive Step Size Estimation Based on Attitude Awareness The Positioning Fusion Engine 400 abandons fixed step size or general step frequency-step size empirical formulas, and dynamically adjusts the step size model based on real-time job posture categories: When the walking posture is identified as normal: stride length L = L0 × (1 + a × (f - f0)); where L0 is the historical statistical base stride length (default 0.65m), f is the current step frequency, f0 is the reference step frequency (e.g. 1.8Hz), and a is the adjustment coefficient (e.g. 0.15). When the movement is identified as pushing a cart: stride length L = L0 × 0.7 (stride length is limited when pushing a cart); When the posture is identified as bending over to sweep: step length L = L0 × 0.4 (mainly leaning forward and moving laterally); When identified as a stationary standing / resting posture: step length L=0, displacement calculation is prohibited.

[0065] In a preferred embodiment, the historical statistical base step size L0 is dynamically calibrated using the positioning results when the satellite signal is good, further improving the estimation accuracy.

[0066] It should be noted that the above default and preferred values ​​are empirical values ​​obtained from a large amount of measured data. In practical applications, they can be individually calibrated according to the height and gait characteristics of different sanitation workers.

[0067] 4.2 Attitude-Auxiliary Smoothing Based on Direction Calculation The positioning fusion engine 400 also employs the following attitude-assisted orientation error suppression strategy: Gyroscope zero bias real-time suppression: When the "stationary standing / resting" posture is detected, the current static output of the three axes of the gyroscope (average value for 2 seconds) is automatically collected as the zero bias value and subtracted in the subsequent direction integration, which significantly suppresses temperature drift; Walking direction change rate constraint: Under the "normal walking / cart walking" posture, the heading angle change rate is limited to ≤30° / second to eliminate sudden changes in direction caused by instantaneous torso swaying; Attitude-direction binding strategy: When the posture is identified as "bending over to clean", due to the forward lean of the body and the lateral movement, the gyroscope integration is prone to cumulative error. At this time, the change of heading angle is directly extrapolated linearly using the historical direction trend of auxiliary positioning data (base station / Wi-Fi) to completely shield the short-term drift of IMU.

[0068] 4.3 Dynamic Adjustment of Multimodal Fusion Weights The Positioning Fusion Engine 400 uses an Extended Kalman Filter (EKF) to achieve multi-source data fusion, and its observation noise covariance matrix is ​​adjusted in real time according to the current operational attitude category and satellite signal strength. (1) When the satellite signal is good (carrier-to-noise ratio ≥32dBHz): Fusion weights (satellite: auxiliary: dead reckoning) = 0.9:0.1:0.

[0069] (2) When the satellite signal is weak / lost (carrier-to-noise ratio <32dBHz): If the attitude is "normal walking / cart movement": fusion weight (satellite: auxiliary: dead reckoning) = 0:0.5:0.5; If the posture is "bending over to sweep / swinging to pick up": Fusion weight (satellite: auxiliary: dead reckoning) = 0:0.8:0.2 (due to slow movement, reduce the dependence on reckoning); If the attitude is "Standing still / resting": Fusion weight (satellite: auxiliary: dead reckoning) = 0:1.0:0 (deduction disabled).

[0070] 4.4 Blind Spot Trajectory Filling Example For example, when a worker sweeps under the shade of a tree, the satellite signal is lost, and the current attitude recognition shows the worker as "bending over to sweep." The positioning fusion engine 400 immediately increases the weight of the auxiliary positioning data to 0.8 and decreases the dead reckoning weight to 0.2, and uses the directional trend bound to the attitude, reporting the fused position every 30 seconds. During this process, the system uses base station / Wi-Fi positioning to determine the approximate location range, while using low-weight dead reckoning to fill in short-term displacements, ensuring trajectory continuity. After 45 seconds, the worker leaves the blind spot, the satellite signal recovers to above 32dBHz, and the system smoothly switches back to satellite primary mode (satellite weight 0.9, auxiliary 0.1) in the next fusion cycle. The trajectory is complete and continuous on the platform side, without interruptions or jumps.

[0071] Step S5: Generate a job trajectory with job status tags based on the uploaded data.

[0072] The cloud platform receives the uploaded data and connects the location points in chronological order. Since each location point has a work posture label, the platform can generate a visual "work storyline" and calculate the effective cleaning time, movement time, and rest time to generate work reports.

[0073] Example 3: This embodiment uses an actual work segment of a sanitation worker as an example to fully present the workflow and effect of the present invention. The workflow is executed collaboratively by the attitude recognition engine 200, the strategy management engine 300, and the positioning fusion engine 400 in the main control unit 140.

[0074] Initial state: The worker starts working wearing terminal device 100, the system initializes, and the satellite positioning module 121 operates in low power mode, performing positioning at a frequency of once every minute.

[0075] Time Segment 1 (Road Sweeping): Workers begin sweeping the road, and the IMU collects data in real time. The attitude recognition engine 200 detects a Z-axis acceleration frequency of 1.2Hz, and an X-axis and Y-axis variance of 0.03m / s². 2 The gyroscope's X-axis exhibits alternating positive and negative peaks. If the conditions are met for four consecutive windows, it is determined to be "bending over for cleaning." The strategy management engine 300 queries the mapping table and triggers a time-integrated judgment: if the proportion of "bending over for cleaning" reaches 85% within 30 seconds (exceeding the 70% threshold), the system officially switches to high-frequency mode, and the satellite positioning module 121 operates at full power, reporting its location every 20 seconds and attaching a "cleaning" status label.

[0076] Time Period 2 (Entering the Blind Zone): The worker moves under the overpass, and the satellite signal drops to 28dBHz, below the preset signal threshold (32dBHz). The positioning fusion engine 400 immediately sets the satellite positioning data weight to 0, adjusts the auxiliary positioning data weight to 0.5, adjusts the dead reckoning weight to 0.5, and starts the attitude-aware step size model (the current attitude is normal walking, and the step size is calculated according to the formula L=L0×(1+a×(f-f0)), which is 0.65m). After dead reckoning continues for 50 meters, the satellite signal recovers, and the system uses satellite coordinates to calibrate the step size base value L0 (updating from 0.65m to 0.63m), and the reckoning error is corrected in time.

[0077] Period 3 (Rest): Workers enter the rest pavilion, and the IMU recognizes that they are "standing still" for 3 minutes. The system then triggers a low-frequency heartbeat mode: the satellite positioning module 121 goes into deep sleep and sends a heartbeat packet only every 10 minutes.

[0078] Period 4 (Resumption of Work): After 10 minutes, the worker gets up and continues working. The IMU detects the "normal walking" posture 3 times in a row (the interval between two adjacent detections is ≤3 seconds). The system wakes up the satellite positioning module 121 according to the hysteresis wake-up rule and resumes the intermediate frequency mode (reporting interval is 2 minutes, low power mode).

[0079] Track Generation: After the day's work is completed, the cloud platform receives a sequence of trajectory points with attitude tags and automatically generates a visual "work storyline." The report shows: effective cleaning time 3.2 hours, movement time 1.5 hours, and rest time 0.8 hours. Actual testing verified that the blind spot trajectory filling rate in this embodiment reaches 100%, and the terminal's average daily power consumption is reduced by 64% compared to traditional fixed high-frequency reporting solutions.

[0080] The specific embodiments of the present invention have been described in detail above, but they are merely examples, and the present invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, all equivalent transformations and modifications made without departing from the spirit and scope of the present invention should be covered within the scope of the present invention.

Claims

1. An adaptive acquisition method for the work trajectory of sanitation workers based on posture recognition, characterized in that, Includes the following steps: The inertial data from the inertial measurement unit (IMU) and the position data from the positioning module are collected synchronously through the terminal device. Based on the inertial data, real-time work posture recognition is performed to obtain the current work posture category, which includes at least one of bending over to sweep, swinging arm to pick up, pushing a cart, walking normally, and standing still / resting. Based on the current working posture category, the trajectory acquisition strategy is dynamically adjusted. The trajectory acquisition strategy includes at least the positioning data reporting frequency and the working power consumption status of the positioning module. The target location data is acquired according to the adjusted trajectory acquisition strategy, and then the target location data is associated with the current operation posture category and uploaded to the cloud platform. A job trajectory with job status tags is generated based on the uploaded data.

2. The method according to claim 1, characterized in that, The terminal device is a smart name tag or shoulder light terminal device worn on the torso or shoulder of sanitation workers. It integrates the inertial measurement unit (IMU), the positioning module, and the communication module. The positioning module includes a satellite positioning module and an auxiliary positioning module. The auxiliary positioning module is used to collect auxiliary positioning data from cellular networks and / or Wi-Fi.

3. The method according to claim 1, characterized in that, The real-time operation posture recognition based on the inertial data includes a step of extracting differentiated features for sanitation operation-specific movements, wherein the differentiated features include at least one of the following: Differential features used to identify bending over cleaning posture include: acceleration Z-axis dominant frequency, acceleration X-axis and Y-axis variance, and alternating peak values ​​of gyroscope X-axis angular velocity. Differential features used to identify the arm-swinging pickup posture include: short-time pulse characteristics of the gyroscope Z-axis angular velocity and sudden drop characteristics of acceleration magnitude; Differential features used to identify the trolley's movement posture include: Z-axis dominant frequency of acceleration, root mean square of acceleration magnitude, and root mean square of gyroscope Y-axis angular velocity; Differential features used to identify normal walking posture include: cadence, root mean square of acceleration magnitude, and variance of acceleration along the X and Y axes. Differential features used to identify static standing / resting postures include: acceleration magnitude variance and gyroscope three-axis angular velocity magnitude.

4. The method according to claim 3, characterized in that, The real-time operation posture recognition also includes multi-feature fusion discrimination of the differentiated features through preset temporal logic rules: The rules for identifying the bending-over cleaning posture include: determining whether the dominant frequency of the acceleration along the Z-axis is within the range of 0.8~1.5Hz, and whether the mean variance of the acceleration along the X-axis and Y-axis is less than 0.05m / s². 2 Furthermore, if the number of alternations between positive and negative peak values ​​of the gyroscope's X-axis angular velocity is ≥3 times per window, and if the above conditions are met consecutively for ≥3 sliding windows, it is determined to be a bending-over cleaning posture; The rules for identifying the arm-swinging pickup posture include: when a gyroscope Z-axis angular velocity pulse is detected that meets condition A: amplitude > 50° / s, main frequency > 3Hz, pulse width 0.3~0.8s, and also meets condition B or condition C, it is determined to be an arm-swinging pickup posture; where condition B is: the acceleration magnitude decreases by more than 40% in 0.5 seconds before and after the pulse, and condition C is: there is no periodic Z-axis fluctuation within 1 second after the pulse; The rules for identifying the trolley's movement include: determining whether the Z-axis dominant frequency of acceleration is within the range of 0.5~1.0Hz and whether the root mean square of the acceleration magnitude is within the range of 0.6~1.0m / s². 2 Within the range, and whether the root mean square of the gyroscope's Y-axis angular velocity is within the range of 2~8° / s, when the above conditions are met simultaneously, it is determined to be the trolley's moving posture; The rules for identifying normal walking posture include: determining whether the cadence is within the range of 1.5~2.5Hz and whether the root mean square magnitude of the acceleration is greater than 1.2m / s². 2 Furthermore, are the variances of the acceleration along both the X and Y axes greater than 0.1 m / s²? 2 When the above conditions are met simultaneously, and the conditions for judging the posture of pushing a cart, bending over to sweep, or waving an arm to pick up are not met, it is judged as a normal walking posture. The rules for identifying static standing / resting postures include: determining whether the variance of the acceleration magnitude is less than 0.02 m / s². 2 Furthermore, if the gyroscope's three-axis angular velocity magnitude is less than 5° / s, and the duration of the above conditions being met continuously reaches the first preset time threshold, it is determined to be a static standing / resting posture.

5. The method according to claim 1, characterized in that, The step of dynamically adjusting the trajectory acquisition strategy based on the current working posture category includes establishing a quantitative mapping relationship between posture and localization strategy: The high-frequency mode is defined as follows: the location data reporting frequency is adjusted to the first frequency, and the power consumption state of the location module is adjusted to full power mode. The intermediate frequency mode is defined as follows: the location data reporting frequency is adjusted to a second frequency lower than the first frequency, and the power consumption state of the location module is adjusted to a low power consumption mode. The low-frequency heartbeat mode is defined as follows: the location data reporting frequency is adjusted to a third frequency lower than the second frequency, and the working power consumption state of the location module is adjusted to a deep sleep mode. When the identified working posture is bending over to sweep or swinging an arm to pick up, switch to the high-frequency mode; When the identified working posture is normal walking or pushing a cart, switch to the mid-frequency mode; When the identified working posture is static standing / resting, switch to the low-frequency heartbeat mode.

6. The method according to claim 5, characterized in that, The method of dynamically adjusting the trajectory acquisition strategy according to the current working posture category also includes an anti-jitter and hysteresis wake-up mechanism, wherein the mechanism includes at least one of the following: Time integration judgment rules: The switch from the current mode to the high-frequency mode is triggered only when the cumulative proportion of bending over to sweep and / or swinging arms to pick up within a 30-second sliding window is ≥70%; the switch to the low-frequency heartbeat mode is triggered only when the continuous duration of a static standing / resting posture is ≥2 minutes. Strategy Oscillation Suppression Rule: When the number of attitude switching times is ≥4 within 10 seconds, the system is forced to enter the frequency stabilization lock state, maintaining the current acquisition strategy unchanged for 30 seconds, and does not respond to any attitude changes during the lock period; Delayed wake-up rule: When switching from the low-frequency heartbeat mode to the medium-frequency mode or high-frequency mode, the positioning module must be woken up only if ≥3 valid working postures are detected continuously and the time interval between two adjacent detections is ≤5 seconds. The positioning module maintains the high-frequency mode for the first 30 seconds after being woken up and does not perform frequency reduction operation.

7. The method according to claim 1, characterized in that, The step of acquiring target location data according to the adjusted trajectory acquisition strategy includes the following sub-steps: multimodal positioning fusion and blind spot compensation. Detect satellite signal strength; When the satellite signal strength is greater than or equal to the signal threshold, the target location data is generated based on the satellite positioning data. When the satellite signal strength is below the signal threshold or is lost, the fusion weights of satellite positioning data, auxiliary positioning data and dead reckoning data are dynamically adjusted according to the current operational attitude category, and dead reckoning based on inertial data is triggered to fill the trajectory gap in the positioning blind spot. The step of dynamically adjusting the fusion weights of satellite positioning data, assisted positioning data, and dead reckoning data according to the current operational attitude category includes: When the satellite signal strength is below the signal threshold, the satellite positioning data weight is set to 0, and the weights of the auxiliary positioning data and dead reckoning data are assigned according to the current operational attitude category: If the current working posture category is normal walking or cart movement, set the fusion weight ratio of auxiliary positioning data and dead reckoning data to 0.5:0.5; If the current operation posture category is bending over to sweep or swinging arm to pick up, set the fusion weight ratio of auxiliary positioning data and dead reckoning data to 0.8:0.2; If the current working posture category is stationary standing / resting, set the auxiliary positioning data weight to 1.0 and disable dead reckoning.

8. The method according to claim 7, characterized in that, The dead reckoning step size parameters are dynamically adjusted according to the current operational attitude category, including: The step length calculation model is dynamically adjusted based on the real-time identified work posture category. When the walking posture is identified as normal, the stride length L = L0 × (1 + a × (f - f0)), where L0 is the historical statistical base stride length, f is the current stride frequency, f0 is the baseline stride frequency, and a is the adjustment coefficient. When the movement is identified as a pushcart, the step length L = L0 × 0.7; When the posture is identified as bending over to sweep, the step length L = L0 × 0.4; When identified as a stationary standing / resting posture, step length L=0.

9. The method according to claim 7 or 8, characterized in that, The dead reckoning also includes attitude-assisted smoothing of the bearing reckoning, including: When the posture is identified as static standing / resting, the static output of the gyroscope is collected as the zero bias value, which is then updated and used for temperature drift compensation in subsequent directional integration. When the posture is identified as bending over to sweep, the change in heading angle is linearly extrapolated using the historical directional trend of auxiliary positioning data to shield the short-term drift of the inertial measurement unit. When the movement is identified as normal walking or pushing a cart, the rate of change of heading angle is limited to ≤30° / second.

10. An adaptive data acquisition system for sanitation worker work trajectories based on posture recognition, characterized in that, The system for implementing the method as described in any one of claims 1-9 includes: Terminal equipment, including: An inertial measurement unit is used to collect inertial data from sanitation workers. The positioning module includes a satellite positioning module and an auxiliary positioning module. The power consumption state of the satellite positioning module switches between full power, low power and deep sleep according to control commands. The auxiliary positioning module is used to collect auxiliary positioning data from cellular networks and / or Wi-Fi. The communication module is used for data interaction with the cloud platform; The pose recognition engine, policy management engine, and localization fusion engine are deployed on the terminal device or cloud platform, wherein: An attitude recognition engine is used to identify the current work attitude category based on the inertial data. The current work attitude category includes at least one of bending over to sweep, swinging arm to pick up, pushing a cart, walking normally, and standing still / resting. The strategy management engine is used to dynamically adjust the trajectory acquisition strategy according to the current operation posture category. The strategy includes the positioning reporting frequency and the working power consumption status of the positioning module. A positioning fusion engine is used to acquire target location data according to the trajectory acquisition strategy and trigger dead reckoning based on the current operational attitude category in satellite signal blind zones. The cloud platform is used to receive uploaded target location data and corresponding job posture categories, and generate job trajectories with job status tags.

Images