Construction worker effective working time quantification method based on fusion of imu and gnss trajectories

Abstract

The application discloses a kind of construction personnel effective working time quantification method based on IMU and GNSS trajectory fusion, through for construction personnel wearing integrated inertial measurement unit and global navigation satellite system module smart bracelet, the instantaneous motion state data and trajectory data of construction personnel are collected;Adaptive least mean square error filter and low-pass filter cascade mode are used to restore pure human action signal;Based on dynamic geofencing and resident point extraction algorithm, spatial validity is determined, and deep learning network is used to identify and validity map construction action;Finally, the space validity vector and action validity vector are integrated in space-time coupling, the effective working time is accurately calculated to generate construction efficiency analysis report, realize the multi-dimensional, continuity, automated evaluation of construction personnel work validity, effectively solve the problem of spatial and action evaluation in traditional method, significantly improve the accuracy and intelligent level of water conservancy construction labor utilization rate analysis.

Description

Technical Field

[0001] This invention relates to the field of engineering construction management technology, and is particularly applicable to a method for quantifying the effective working time of construction personnel based on the fusion of IMU and GNSS trajectories. Background Technology

[0002] Water conservancy projects (such as reservoirs and dams, water diversion projects, etc.) are fundamental lifelines for national economic and social development, characterized by large-scale investment, long construction periods, complex construction environments, and high safety risks. Accompanying the main construction work, such as dam filling, tunnel excavation, and metal structure installation, are large numbers of personnel, materials, and machinery working collaboratively in high-intensity, dynamically changing environments. During water conservancy construction, factors such as fatigue, on-site waiting, unnecessary movement, and improper resource allocation can significantly reduce the effective working time of construction workers. How to scientifically and accurately evaluate the actual working efficiency of frontline construction workers, identify the key links in the loss of working hours, and thereby optimize resource allocation and improve construction management has always been a major challenge in the field of water conservancy construction.

[0003] However, existing attendance systems based on GPS or RFID positioning technologies can only address the spatial issue of whether personnel are on-site, failing to distinguish between workers actively engaged in work and those passively slacking off. Work hour statistics still primarily rely on manual on-site recording, construction log entries, and random checks of video surveillance using fixed cameras. On the complex surface of a hydraulic dam's pouring section, manual observation is not only costly but also impossible to achieve 24 / 7 tracking; while video surveillance is often limited by site obstructions, insufficient lighting at night, and the constantly changing elevation of the dam's working surface, resulting in numerous blind spots.

[0004] In recent years, with the rapid development of the Internet of Things (IoT), wearable sensing technology, BeiDou / GPS high-precision positioning technology, and deep learning algorithms in artificial intelligence, some studies have attempted to use inertial measurement units (IMUs) built into smart bracelets or safety helmets to identify workers' body movements, or to use GPS modules for personnel positioning. However, water conservancy construction sites are filled with complex environmental factors such as strong vibrations, strong electromagnetic interference, and multipath effects. For example, the broadband, high-intensity vibrations generated by equipment such as large vibratory rollers, hydraulic breakers, and high-frequency vibrators can severely contaminate the human motion signals collected by wearable sensors, making them easily misjudged by existing algorithms as high-frequency manual production actions, resulting in a seriously inflated effective working time statistics; satellite signal blockage in canyon areas can cause GPS trajectory drift and interruption. Existing technologies lack robust joint filtering and cleaning methods for handling such multi-source heterogeneous data noise.

[0005] In the specific scenario of water conservancy projects, which are characterized by strong interference and high dynamism, there is a lack of in-depth research on the spatiotemporal alignment and logical fusion of micro-action semantics and macro-spatial semantics, and there is still no accurate calculation model for the effective working time that combines the spatiotemporal coupling integration of spatial effectiveness and action effectiveness.

[0006] Therefore, how to overcome the interference of the complex physical environment unique to water conservancy projects, deeply integrate the macroscopic trajectory space dwell with the microscopic limb movement semantics, and propose a set of effective working time quantification algorithms that can accurately eliminate invalid dwell and pseudo-labor has become a core technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] The purpose of this invention is to provide a method for quantifying the effective working time of construction workers based on the fusion of IMU and GNSS trajectories, in order to solve the problems of large environmental interference, high single-dimensional misjudgment rate, and lack of real-time and accurate quantification methods in the analysis of construction efficiency of water conservancy projects.

[0008] To achieve the above objectives, the method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion, as described in this invention, includes the following steps: The S1, a smart bracelet integrating a six-axis inertial measurement unit (IMU) and a global navigation satellite system (GNSS), synchronously collects instantaneous motion and trajectory data of construction workers via wireless transmission protocols and a unified timestamp reference. It collects trajectory data, including longitude, latitude, elevation, and horizontal accuracy factors, at relatively low frequencies (e.g., 1Hz) to track changes in the workers' overall spatial position on the construction site. Simultaneously, it collects kinematic time series data, including instantaneous three-axis acceleration and three-axis angular velocity of the workers' limbs, at high frequencies (e.g., 50-200Hz) to precisely characterize the instantaneous motion state of the human body.

[0009] S2 employs an adaptive minimum mean square error filter to adaptively adjust the filter weight coefficients from the instantaneous limb motion state data, eliminating vibration noise, and cascades a low-pass filter to filter out high-frequency residual noise, restoring the real human motion signal.

[0010] Due to the complex mechanical vibration interference in water conservancy projects caused by heavy vehicle engines (such as vibratory rollers and drilling rigs) and handheld high-frequency vibratory equipment transmitted to the wrist, this application adopts an adaptive minimum mean square error (LMS) filter to dynamically track and cancel the non-stationary interference frequency components generated by vehicles and machinery.

[0011] The dominant vibration frequency component is separated from the raw instantaneous limb motion data collected in step S1 through frequency domain analysis. This dominant vibration frequency component is used as the reference input signal of the adaptive minimum mean square error (LMS) filter. The filter weight coefficients are dynamically adjusted to track and cancel the dominant vibration frequency component.

[0012] Alternatively, a reference accelerometer signal deployed near the location of the currently acquired signal can be used as the reference input signal for the adaptive minimum mean square error (LMS) filter, and the filter weight coefficients can be dynamically adjusted to track and cancel the vibration noise.

[0013] Subsequently, considering that the frequency range of conscious human movements (such as walking, bending over, and operating tools) is mainly concentrated in the range of 0-20Hz, the signal after being denoised by the adaptive minimum mean square error (LMS) filter is input into a Butterworth low-pass filter with a cutoff frequency of 20Hz to further filter out high-frequency residual noise and restore a smooth and realistic human movement signal.

[0014] S3, for the trajectory data of construction workers collected in step S1, removes outliers with a horizontal precision factor (HDOP) > 5, and uses a Kalman filter to smooth the trajectory, thus completing the cleaning of the trajectory data.

[0015] After cleaning, the trajectory data is used to extract trajectory dwell points using a dual-constraint algorithm based on distance and time thresholds. These are clusters of trajectory points where the worker's spatial position changes minimally but lasts for a relatively long time. These clusters represent the worker's actual work area. Specifically, this is done by finding clusters of trajectory points where the maximum spatial distance between any two trajectory points is less than the distance threshold and the dwell time of the trajectory points is greater than the time threshold. Then, the extracted trajectory dwell point coordinates are compared with a pre-built and dynamically updated set of geofences for spatial collision detection. This geofence is dynamically delineated in a digital twin model based on the construction organization design, daily construction plan, and actual site conditions. Its boundaries precisely correspond to core work areas (such as dam filling areas, pouring surfaces, and tunnel faces), auxiliary work areas, and danger warning zones. When a worker's trajectory dwell point is determined to be within a pre-authorized core work area, the spatial state validity vector is assigned a value of 1 in the corresponding time interval; otherwise, it is 0.

[0016] S4 employs an adaptive sliding window to segment the real human action signals reconstructed in step S2. The window length is adaptively set based on the periodic characteristics of the construction actions. The segmented real human action signal time-series data is input into a pre-trained deep learning action recognition network to identify the construction action categories. The deep learning action recognition network uses a deep learning model based on a bidirectional long short-term memory network combined with an attention mechanism (BiLSTM-Attention). The output layer uses the Softmax function to calculate the probability of the real human action signal corresponding to each preset construction action category within the current sliding time window, selecting the category with the highest probability as the recognition result. The construction action categories are preset according to the water conservancy engineering construction scenario, including walking, standing, bending over, squatting, climbing, operating a vibrator, tying rebar, driving vehicles, etc. Each construction action category is also classified and labeled as either a valid or invalid productive action. The identified construction action category is mapped to the corresponding time interval value in the action state validity vector based on its label. The action state validity vector uses 1 to represent a valid productive action and 0 to represent an invalid productive action.

[0017] S5 aligns the spatial state validity vector and the action state validity vector along the time axis and performs a logical AND operation to construct a spatiotemporally coupled integral model. This model calculates effective working time and total attendance time, generating a real-time construction efficiency analysis report, including construction efficiency indicators, time loss nodes and their durations, frequency of occurrence, and bottleneck location information. By analyzing intervals where the action state validity vector or spatial state validity vector on the time axis is zero, time loss nodes are automatically identified. These time loss nodes typically correspond to waiting for materials, equipment maintenance, unnecessary movement, and exceeding designated rest areas.

[0018] The spatiotemporal coupling integral model is as follows: ,in, Effective working hours; This is the spatial state validity vector; This is the action state validity vector. t represents the sampling time interval, and t represents the total attendance time.

[0019] Compared with the prior art, the present invention has the following beneficial effects: (1) In response to the interference environment of heavy machinery and vibrating equipment unique to water conservancy dams, an adaptive LMS and low-pass combined filtering technology was introduced to effectively solve the problem of false recognition of "pseudo-high frequency effective action" by smart bracelets in water conservancy sites, and greatly improve the purity of the underlying data.

[0020] (2) The semantics of human micro-movements and macro-spaces are spatiotemporally coupled and analyzed. The effective working time is precisely defined by logical AND operation, effectively eliminating the gray working time that cannot be distinguished by traditional methods, making the evaluation of labor utilization more scientific and objective.

[0021] (3) The action recognition based on deep learning and the spatial judgment based on dynamic geofence realize the automated and continuous monitoring of personnel's work status. No manual intervention is required throughout the process, avoiding data deviation caused by human factors.

[0022] (4) It can output efficiency analysis reports, which can not only provide macro utilization rate indicators, but also pinpoint the specific time and space nodes where the time loss occurs, providing clear and traceable data basis for managers to implement precise policies, optimize construction processes and reduce non-productive consumption. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method for quantifying the effective working time of construction workers based on the fusion of IMU and GNSS trajectories, as described in this invention.

[0024] Figure 2 This is a schematic diagram of triaxial accelerometer data under different states in this invention.

[0025] Figure 3 This is a clustering diagram of spatial dwelling points extracted by the time and distance dual threshold constraint algorithm of this invention.

[0026] Figure 4 This is a diagram of the BiLSTM-Attention network structure of the present invention.

[0027] Figure 5 This is a schematic diagram of the classification accuracy curve of the BiLSTM-Attention model of this invention.

[0028] Figure 6 This is a panel display diagram showing the effective working time data of construction workers according to the present invention. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1 Taking the compaction surface of a large roller-compacted concrete gravity dam as an application scenario, and selecting a vibratory roller compactor operator (hereinafter referred to as "worker A") as the monitoring object, the specific implementation process of the present invention is described in detail.

[0031] like Figure 1 As shown, the method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion described in this invention comprises the following steps: Step S1: Synchronous acquisition of multi-source heterogeneous data.

[0032] Before worker A enters the construction site, he must wear a smart bracelet that integrates an IMU and a high-precision GNSS module.

[0033] The core hardware of this bracelet includes: (1) Six-axis Inertial Measurement Unit (IMU): Integrates a three-axis accelerometer and a three-axis gyroscope. To accurately capture the operator's subtle limb movements in the cab and the body swaying caused by vehicle bumps, the IMU's data sampling frequency is set to 100Hz. This module simultaneously outputs three-axis acceleration. With triaxial angular velocity The triaxial accelerometer data collected under different states by this invention are as follows: Figure 2 As shown, Figure 2 a shows the working status signal; Figure 2 b shows the walking status signal. Figure 2 c shows the resting state signal.

[0034] (2) Global Navigation Satellite System (GNSS) Module: Supports joint positioning using GPS, BeiDou, and GLONASS systems. To balance positioning accuracy and power consumption, the positioning data output frequency is set to 1Hz. This module simultaneously outputs information such as longitude, latitude, altitude, horizontal accuracy factor, and number of satellites.

[0035] (3) Communication and timing module: The wristband has a built-in 4G / 5G communication module for uploading data to the cloud server in real time. To ensure the time consistency of all sensor data, the system uses the Network Time Protocol (NTP) to periodically synchronize with the server, ensuring that the time synchronization error is less than 10 milliseconds. Each frame of data is accompanied by a unified timestamp.

[0036] Step S2: IMU noise cancellation based on adaptive filtering.

[0037] In this embodiment, during operation, the vibratory roller's steel wheel violently collides with the concrete aggregate, generating a wide-frequency, high-intensity periodic vibration covering a frequency range of 10Hz to 60Hz. This vibration is transmitted through the mechanical structure to the wristband worn by worker A, severely overlapping with worker A's own arm movement signals, directly causing serious interference to subsequent action recognition.

[0038] To resolve this issue, the following steps are performed: (1) Adaptive LMS filter Construct an adaptive minimum mean square error (LMS) filter. The core idea of ​​this filter is to improve the performance of the original noisy signal. Subtract one from the reference noise The relevant noise is estimated to obtain a clean signal. Reference noise The signal is taken from the reference accelerometer signal deployed near the construction machinery, or from the dominant vibration frequency component separated from the original IMU signal through frequency domain analysis.

[0039] In this embodiment, since the frequency characteristics of the roller compactor vibration are relatively fixed and the energy is concentrated, the reference input signal of the filter is... Taken directly from the original acceleration signal In this process, the dominant vibrational frequency components (e.g., 30-40Hz) are analyzed in real time using Fast Fourier Transform (FFT), and a sine wave with these dominant frequencies is constructed as a... .

[0040] The iterative formula for the filter is as follows: in, This is the filter weight vector; The step size factor is dynamically adjusted based on the signal-to-noise ratio, ranging from 0.005 to 0.02; the LMS filter order is set to 32.

[0041] This step effectively cancels out the periodic noise associated with the vibration of the rolling mill.

[0042] (2) Butterworth low-pass filter Even after LMS filtering, some high-frequency harmonic noise may still remain in the signal. Considering that the frequency of conscious human actions is mainly distributed between 0-20Hz, the system will filter the LMS-filtered signal... A fourth-order Butterworth low-pass filter is applied. The filter's cutoff frequency is set to 20Hz, and its stopband attenuation is greater than 40dB. This step further smooths the signal, eliminates high-frequency residuals, and reconstructs a clear human motion waveform, providing high-quality input data for subsequent motion recognition.

[0043] Step S3: Determine the spatial validity based on dynamic geofencing.

[0044] In this embodiment, the spatial location change of worker A is reflected by GNSS trajectory. The specific processing flow is as follows: (1) GNSS data cleaning and smoothing The system receives worker A's raw GNSS trajectory points in real time. First, all positioning points with an HDOP > 5 are removed, as these points are usually caused by satellite signal obstruction or multipath effects, resulting in extremely poor positioning accuracy. Then, a Kalman filter is used to smooth the cleaned trajectory, making it more consistent with actual motion patterns.

[0045] (2) Detention point extraction A dual-constraint algorithm using distance and time thresholds is employed to extract the worker's trajectory dwell points. The smoothed trajectory points are traversed to find point clusters that satisfy the following condition: the maximum spatial distance between any two points within the cluster is less than a preset distance threshold. Furthermore, the duration of the trajectory points in this cluster of points is greater than a preset time threshold. .

[0046] In this embodiment, based on the spatial scale and operational characteristics of the dam filling area, the following settings are made: rice, Minutes. The extracted cluster of dwell points represents the area where worker A stays stably over a period of time.

[0047] (3) Dynamic geofencing In the digital twin platform, based on the daily construction organization design, the work area is dynamically divided into three sub-areas: Area A (the area currently being poured, which is the core work area), Area B (the area to be poured), and Area C (the area that has been poured and is undergoing curing). The fence boundary is composed of polygons defined by a set of latitude and longitude coordinates. The fence boundary is updated in real time according to the pouring progress.

[0048] The center coordinates of the extracted dwell point clusters are compared with the geofence polygons using point-polygon collision detection (e.g., ray tracing). When a dwell point is detected to be located within the core operational area A, the spatial state validity vector for that time interval is calculated. Assign a value of 1; otherwise, assign 0. The analysis results show that worker A's total stay in the core work area is 6.2 hours, and his total stay in the non-work area is 1.8 hours. For example... Figure 3 As shown, this is a schematic diagram of the extracted cluster of dwelling points. After spatial collision detection, the dwelling area can be intuitively seen by color.

[0049] Step S4: Determining the validity of action states based on deep learning This step employs a deep learning model based on a bidirectional long short-term memory network combined with an attention mechanism (BiLSTM-Attention) to identify and determine the validity of worker A's actions. Specifically: (1) Data preprocessing and window splitting The purified, realistic human motion signal processed in step S2, i.e., IMU data, including three-axis acceleration and three-axis angular velocity (6 channels in total), is used as input. An adaptive sliding window with a window length of 3 seconds and a sliding step size of 1.5 seconds is used to segment the data stream. The window length of 3 seconds is based on the spectral analysis of the periodic action of "driving a roller mill," the typical period of which is about 2-3 seconds. The sliding step size is half the window length to ensure the continuity of time series analysis and computational efficiency.

[0050] (2) Model structure and training The BiLSTM-Attention model structure used in this embodiment is as follows: Figure 4 As shown: ① Input layer: Receives a tensor of shape (window_length, n_features), where window_length = 300 (3 seconds × 100Hz) and n_features = 6.

[0051] ② Bidirectional LSTM layer: Two stacked BiLSTM layers are set up, with 128 hidden units in each layer. BiLSTM can capture temporal dependencies from both forward and backward directions, comprehensively understand the contextual information of actions, and avoid the recognition ambiguity caused by single-moment features.

[0052] ③ Attention Mechanism Layer: An attention mechanism is applied to the output of the BiLSTM. This layer learns a set of weights and automatically focuses on the key time step that is most discriminative for classifying the current action. The formula for calculating the attention weights is: in, It is BiLSTM at the time step The output; and These are learnable weights and biases; It is a context vector; These are normalized attention weights. The final output is a weighted sum. .

[0053] ④ Fully connected layer and output layer: The output of the attention layer is fed into a fully connected layer containing 64 neurons, and finally outputs the probability distribution of action categories through the Softmax layer.

[0054] The model was pre-trained on a large-scale labeled dataset containing various actions in hydraulic engineering (walking, standing, driving a roller, getting on and off a vehicle, resting inside the vehicle, bending over, etc.). Training used the Adam optimizer with a learning rate of 0.001 and the loss function was classification cross-entropy. Figure 5 The figure shows the classification accuracy curve of the BiLSTM-Attention model used in this embodiment.

[0055] (3) Action recognition and validity mapping For each input window, the model outputs the action label with the highest probability. The system maps the identified "driving the roller" as a valid productive action, and the other actions (such as "resting in the vehicle," "getting on and off the vehicle," and "walking") as invalid productive actions. When a window action is determined to be valid, the action state validity vector for the corresponding time interval of that window is calculated. Assign a value of 1 if the condition is met, otherwise assign a value of 0. The analysis results show that worker A, within 8 hours, The total duration is 5.2 hours.

[0056] Step S5: Construction efficiency evaluation and analysis based on spatiotemporal coupling integral Will and Perform timeline alignment and logical AND operations second by second to calculate and output the effective working time. The formula is as follows: Calculations show that worker A's total time was 4.8 hours. This means that although worker A spent 6.2 hours in the core work area and 5.2 hours operating the roller, only 4.8 hours were effective working time operating the roller within the work area.

[0057] Further analysis and The mismatched time periods can be identified as follows: Spacetime mismatch 1 (space valid, action invalid): Worker A spends 1.4 hours in the compaction area, but is in a "resting in the vehicle" state, which usually occurs during the intervals while waiting for the dump truck to supply materials. This is the main source of lost working hours.

[0058] Spacetime mismatch 2 (space invalid, action valid): Worker A has 0.3 hours of "driving the roller" action, which occurs on the transfer road from the compaction area to other areas. Although he is driving during this journey, he does not perform effective compaction in the core work area, so it is not counted.

[0059] The final efficiency analysis report indicated that worker A's labor utilization rate on that day was [missing information]. The main efficiency bottleneck lies in the excessively long non-productive waiting time for materials within the work area. The report recommends optimizing the material supply plan to reduce waiting time.

[0060] like Figure 6The image shows a display panel for the statistical data on the effective working time of construction workers as described in this invention. The data panel provides a clear overview of the distribution of construction workers' on-site time throughout the day, the percentage of effective working hours per person, and data on abnormal attendance.

[0061] Example 2 Using the excavation face of a hydraulic tunnel project as an application scenario, this study evaluates the operational effectiveness of a work team consisting of four drilling workers (B1-B4). Since there is no GNSS signal inside the tunnel, this embodiment uses ultra-wideband (UWB) positioning technology instead of a GNSS module. The specific steps are as follows: Step S1: Synchronous Acquisition of Multi-Source Heterogeneous Data The smart bracelets worn by the four workers also integrate IMUs, but the positioning modules have been replaced with UWB modules. UWB positioning base stations are deployed at 50-meter intervals inside the tunnel to achieve centimeter-level real-time positioning of the workers, with the positioning frequency set at 5Hz. All data is collected via a wired network to a base station at the tunnel entrance and then transmitted to a central server.

[0062] Step S2: IMU noise cancellation based on adaptive filtering During drilling operations, the pneumatic rock drills used by workers generate high-intensity, wide-bandwidth impact vibrations (energy mainly distributed in the 10-150Hz range). This embodiment employs an improved adaptive LMS filter. Since a reference sensor cannot be deployed, the filter's reference input signal... The original IMU signal is constructed as follows: A real-time short-time Fourier transform (STFT) is performed to identify the frequency band with the most concentrated energy (usually related to the rock drill's impact frequency and its harmonics), and then a sine wave with that dominant frequency is generated as... The filter order is set to 64, and the step size factor is... Dynamic adjustment. The amplitude of the impact vibration was successfully attenuated by about 70%, and the subsequent 20Hz low-pass filter further purified the signal, allowing subtle movements to be detected.

[0063] Step S3: Determining the spatial status validity based on dynamic geofencing Using UWB positioning data, geofences were dynamically delineated within a digital twin tunnel model. The core work area for the day was the tunnel face region from chainage K1+200 to K1+210, and the geofence boundary consisted of a series of three-dimensional coordinate points. A distance threshold was also used. meters, due to the confined space of the tunnel) and time threshold ( The algorithm employs a double-constraint approach (minutes) to extract worker dwell times. The results show that the average dwell time of the four workers in the core work area is 5.6 hours per person (total attendance of 8 hours).

[0064] Step S4: Action validity assessment based on deep learning The IMU data processed in step S2 is input into a pre-trained BiLSTM-Attention model. The network structure of this model is the same as in Example 1. The action label library includes categories such as "drilling operation," "equipment handling," "walking," "short rest," and "waiting." The model maps "drilling operation" and "equipment handling" to effective productive actions. The analysis results show that the average action validity time for the four workers is 5.2 hours per person.

[0065] Step S5: Evaluation and Analysis of Hydraulic Construction Efficiency Based on Spatiotemporal Coupling Spatiotemporal coupling calculations were performed to determine the average effective working time of the four workers. The average utilization rate is 4.1 hours per person. The average utilization rate is 4.1 / 8 = 51.25%.

[0066] The collaborative efficiency analysis report generated by the system revealed key management information: ① Analysis of sequence coherence issues. Through analysis... and During this period, a common problem was found among the work teams: after each round of drilling operations, there was a collective "waiting" period of approximately 0.6 hours. Comparison with the construction log confirmed that this period corresponded to waiting for blasting materials to arrive. The report clearly identified the bottleneck in this process transition.

[0067] ② Identification of body differences. The report compared the spatiotemporal coupling efficiency of four workers. It was found that the effective utilization rate of worker B3 was only 38%, significantly lower than the average of the other three (approximately 55%). Further analysis revealed that... and The time-series data revealed that B3 personnel took "short breaks" far more frequently than average in the core work area, and spent a significant portion of their time "moving equipment." Based on this, the report recommends providing B3 personnel with enhanced skills training and monitoring their individual work performance.

Claims

1. A method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion, characterized in that, Includes the following steps: S1 uses a smart bracelet that integrates a six-axis inertial measurement unit and a global navigation satellite system to collect real-time data on the instantaneous movement status and trajectory of construction workers' limbs; S2 uses an adaptive minimum mean square error filter to adaptively adjust the filter weight coefficients from the instantaneous limb motion state data to eliminate vibration noise, and cascades a low-pass filter to filter out high-frequency residual noise and restore the real human motion signal. S3, clean the trajectory data, use the distance threshold and time threshold dual constraint algorithm to extract the trajectory dwell points, perform collision detection between the dwell point coordinates and the dynamically defined geofence, and generate a spatial state validity vector. S4. After segmenting the real human action time sequence signal using an adaptive sliding window, the signal is input into a pre-trained deep learning action recognition network to identify the construction action category and map it into an action state validity vector. S5 aligns the spatial state validity vector and the action state validity vector along the time axis and performs a logical AND operation to construct a spatiotemporal coupled integral model, calculates the effective working time, and generates a construction efficiency analysis report.

2. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: In step S1, the instantaneous motion state data of the limb includes the instantaneous triaxial acceleration and triaxial angular velocity of the limb; the trajectory data includes longitude, latitude, elevation and horizontal accuracy factor.

3. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The adaptive minimum mean square error filter described in step S2 uses the dominant vibration frequency component separated from the original instantaneous limb motion data collected in step S1 as the reference input signal, or the reference accelerometer signal deployed near the vibration source of the construction machinery as the reference input signal, to dynamically adjust the filter weight coefficients and adaptively estimate and cancel the noise component in the original instantaneous limb motion data.

4. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The low-pass filter mentioned in step S2 is a Butterworth low-pass filter with a cutoff frequency of 20Hz.

5. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The dual-constraint algorithm of distance threshold and time threshold described in step S3 is as follows: find the cluster of trajectory points where the maximum spatial distance between any two trajectory points is less than the distance threshold and the dwell time of the trajectory points is greater than the time threshold, and these are the trajectory dwell points.

6. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The geofence described in step S3 is defined in the digital twin model based on the construction organization design, daily construction plan, and actual site conditions.

7. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The deep learning action recognition network described in step S4 adopts a deep learning model based on a bidirectional long short-term memory network combined with an attention mechanism. It outputs the probability of the real human action signal of the current sliding window corresponding to each preset construction action category through the Softmax function, and selects the category with the highest probability as the recognition result.

8. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The construction action categories mentioned in step S4 are preset according to the construction scenario of water conservancy projects and divided into effective productive actions or ineffective productive actions; the action state validity vector uses 1 to represent effective productive actions and 0 to represent ineffective productive actions.

9. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The spatiotemporal coupling integral model described in step S5 is: ,in, Effective working hours; This is the spatial state validity vector; This is the action state validity vector. t represents the sampling time interval, and t represents the total attendance time.

10. The method for quantifying the effective working time of construction workers based on IMU and GNSS trajectory fusion according to claim 1, characterized in that: The construction efficiency analysis report mentioned in step S5 includes construction efficiency indicators, time loss nodes and their duration, frequency of occurrence, and bottleneck location information; the time loss nodes are automatically identified by analyzing the intervals where the action state validity vector or spatial state validity vector on the time axis is zero.

Images