Robot-based multi-sensor fusion measurement method and system

Abstract

The application belongs to the field of engineering surveying and robotics, and specifically proposes a multi-sensor fusion surveying method and system based on robots. The method first completes the alignment of sensor data timestamps and the calibration of external parameters. Based on SLAM and RTK fusion positioning and improved A* algorithm, the path is planned to drive the robot to the survey point. Then, the level, inertial navigation, laser radar, binocular vision and GNSS-RTK are used to synchronously collect elevation, attitude, environment and positioning data. After the gross error elimination and temperature and vibration compensation, the elevation and inertial navigation data are fused, and the adaptive weighting and Kalman filtering are used to correct the zero drift. The Helmert variance component estimation is used for adjustment, and the sliding window detection is combined to identify errors and generate a report. The application effectively solves the problems of low efficiency, single dimension, poor fusion accuracy and insufficient environmental stability of traditional manual measurement, meets the long-term automatic monitoring demand of water conservancy dam body, and realizes the whole process of unmanned operation, with the measurement efficiency improved by 3-5 times compared with the manual method.

Description

Technical Field

[0001] This invention belongs to the field of engineering surveying and robotics technology, specifically relating to a robot-based multi-sensor fusion measurement method and system for long-term first-order leveling automated monitoring of water conservancy dams, large slopes, and power plant structures. Background Technology

[0002] Engineering surveying is a key technical link in the construction and operation and maintenance of water conservancy projects, dam safety monitoring and other projects. Traditional surveying methods mainly rely on manual operation of single sensor equipment such as total stations and levels. It requires manual setup, leveling, aiming, reading and recording. The overall operation efficiency is low, the labor cost is high, and it is easily affected by human factors, resulting in measurement errors. It is difficult to meet the long-term needs of automated and high-precision measurement.

[0003] Existing automated measurement systems mostly use single-type sensors for data acquisition, resulting in limited measurement dimensions and an inability to meet the comprehensive measurement requirements of complex engineering scenarios. Existing multi-sensor integrated systems generally suffer from problems such as low spatiotemporal registration accuracy of heterogeneous sensors, poor data fusion algorithm performance, and low system integration. Furthermore, they exhibit poor stability under complex environments such as vibration, temperature changes, and wind disturbances, making it difficult to guarantee high accuracy and reliability of elevation measurements and to strictly meet the specifications for first-order leveling. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a robot-based multi-sensor fusion measurement method and system, which solves the problems of low efficiency, insufficient single-sensor dimension, poor accuracy of heterogeneous data fusion, and poor stability in complex environments in traditional manual measurement.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a multi-sensor fusion measurement method based on a robot, comprising the following steps: S1. System initialization and self-test; S2. Complete the timestamp alignment of data from each sensor and the calibration of the coordinate system extrinsic parameters; S3. Based on the SLAM and RTK fusion localization technology, the improved A* algorithm is used to plan the movement path and drive the robot to reach the target measurement point; S4. After adjusting the level to a horizontal position, trigger the synchronous acquisition command to collect elevation data using the level. Record the robot's attitude angles and acceleration in real time using the IMS inertial navigation module. Obtain environmental perception data by synchronously scanning the surrounding environment using LiDAR and binocular cameras, and use GNSS... The RTK module acquires location data; S5. Use the 3σ criterion to remove gross errors in the collected data and perform temperature and vibration compensation based on environmental parameters; S6. The elevation observations of the level instrument are fused with the attitude angle and acceleration data. The adaptive weighting algorithm and Kalman filter are used to correct the zero drift of the inertial navigation system and output high-precision pose results. S7. Use the Helmert variance component estimation method for indirect adjustment, and combine it with sliding window variance detection to identify random errors and generate a measurement quality report.

[0006] In the preferred embodiment, step S2 employs dynamic timestamp synchronization technology, using the edge computing box as the main clock source to synchronize the time of each sensor, and adopts feature-point-based vision. Inertial navigation calibration method to achieve unification of external and internal parameters of multiple sensors.

[0007] In the preferred embodiment, step S3, which involves planning the movement path using the improved A* algorithm, specifically includes: Construct a comprehensive cost function that includes distance cost, energy cost, and measurement point priority weights: f ( n )= g ( n )+ h ( n )+ w ( n ); in, g ( n ) represents the movement cost from the starting point to the current node. h ( n ) represents the heuristic distance from the current node to the target node. w ( n ) represents the priority weight of the measurement points; The system analyzes the GNSS-RTK positioning status in real time, determines the number of available satellites and the positioning PDOP value. When PDOP < 3 and the number of satellites ≥ 8, the RTK signal is deemed available, and the RTK centimeter-level coordinates are used as the global positioning reference. When satellites are blocked or PDOP > 5, the signal is deemed rejected, and the system switches to pure SLAM positioning mode, using lidar point cloud matching and inertial navigation calculation for positioning.

[0008] In the preferred embodiment, step S4, which involves collecting elevation data using a level instrument, specifically includes: The control of the rotating gimbal drives the level to rotate horizontally and tilt vertically. With the assistance of binocular camera vision recognition, the center of the crosshairs of the level objective lens is automatically aligned with the scale of the leveling rod, and the level outputs a digital elevation reading.

[0009] In a preferred embodiment, step S5, which involves performing temperature and vibration compensation based on environmental parameters, specifically includes: (1) Temperature compensation: real-time temperature of the environment is collected from the ambient temperature sensor. T0, reference measurement temperature T s Temperature difference ΔT=T 0 -T s Substitute into the thermal expansion compensation formula ΔL=αLΔT ,in, α The coefficient of linear expansion of the leveling rod material. L The original reading length of the level instrument, and the corrected elevation. L'=L-ΔL ; (2) Vibration compensation: Constructing a first-order linear vibration error model ΔH v = k 1 a x +k 2 a y +k 3 oh z In the formula a x 、a y For the horizontal vibration acceleration of the inertial navigation system, oh z The angular velocity of vibration about the vertical axis, k 1 、k 2 、k 3 represents the vibration coupling coefficient obtained from calibration; vibration deviation is calculated by inputting high-frequency vibration data from the inertial navigation system in real time. ΔH v Subtract from the original reading of the level instrument ΔH v It is used to suppress reading drift caused by track movement and chassis micro-vibration.

[0010] In the preferred embodiment, step S6, which uses an adaptive weighting algorithm and Kalman filtering to correct inertial navigation zero drift, specifically includes: S601. Construct the Kalman filter state vector: X= [ f,i,s,a x ,a y ,a z ,b gx ,b gy ,b gz ]; in, f、θ、ψ These are roll angle, pitch angle, and yaw angle, respectively. ax 、a y 、a z These are triaxial accelerations. b gx 、b gy 、 b gz The parameters to be estimated for zero drift of the gyroscope; The state transition equation is: X k =F X k-1 +W k ; in, F Here is the state transition matrix for the inertial navigation strapdown system. W k This is process noise; S602, the measurement equation is: Z k =H X k +V k ; Among them, the measured values Z k For continuous sampling of elevation changes by a level instrument ΔH , H For the observation matrix, V k To observe noise; S603, Adaptive Weighted Calculation: Calculate the real-time signal-to-noise ratio (SNR) of inertial navigation and leveling instrument data respectively. IMU SNR LEVEL Then calculate the weighting coefficients respectively: w IMU =SNR IMU / ( SNR IMU +SNR LEVEL ); w LEVEL =SNR LEVEL / ( SNR IMU +SNRLEVEL ); The observation matrix is ​​dynamically updated based on two sets of weights. H Internal coefficients for each dimension; dynamically update the weights of each dimension of the H matrix. S604, Kalman Iterative Update: First, time prediction is completed using the state transition matrix, then measurement updates are completed using the adaptive weighted observation matrix, iteratively solving for gyroscope zero drift. b gx 、b gy 、b gz ; S605. Substitute the zero-drift compensation value obtained through iteration back into the original angular velocity data of the IMS inertial navigation system, and output the fused and corrected high-precision pose and elevation data.

[0011] In the preferred embodiment, step S7, which combines sliding window variance detection to identify random errors, specifically includes: Set the sliding window size and calculate the variance σ² of the elevation residuals within the window window one window at a time. Preset error threshold s 0, when the variance of the elevation residual σ² exceeds the preset threshold. s When the value is 0, it is determined that there is a random error in the current data segment, and a retest instruction is triggered. If the variance still exceeds the limit after two consecutive retests, the abnormal data segment is removed. After gross error elimination, the Helmert variance component estimation iterative calculation is used to calculate the prior variances of the three types of observations: leveling instrument, inertial navigation, and RTK. The indirect adjustment method equation is constructed to solve for the optimal elevation and measurement point coordinates. A standardized measurement quality report is output, which includes the positional error, the standard deviation of elevation per kilometer for round trip measurements, and the residual distribution.

[0012] In the preferred embodiment, step S8 is also included: transmitting the measurement quality report to the edge computing box carried by the robot, pushing it to the cloud server using the MQTT communication protocol, generating a visual measurement report on the cloud server, and overlaying and displaying the measurement point locations on the GIS map of the terminal device.

[0013] The present invention also provides a robot-based multi-sensor fusion measurement system for performing the above method, comprising: The robot mobile platform module provides autonomous movement, environmental perception, and precise positioning capabilities, and outputs real-time pose and obstacle information. The sensor integration module is used to integrate high-precision elevation measurement, automatic leveling, gimbal rotation, and attitude sensing units, and output multi-dimensional raw measurement data; The spatiotemporal registration and synchronization module is used to unify the timestamps of multiple sensors and calibrate the spatial coordinates, and to complete data preprocessing and gross error removal; The multi-source data fusion processing module is used to perform tight-coupled fusion, adaptive weighting and filtering correction on heterogeneous data, and output high-precision fusion results; The autonomous measurement and path planning module is used to automate the entire process of autonomous navigation, path optimization, automatic leveling, automatic aiming, and automatic reading. The measurement results output and quality assessment module is used to complete data adjustment, error correction, quality assessment and cloud transmission, and output standardized measurement reports.

[0014] In a preferred embodiment, the robot mobile platform module includes: Tracked chassis module, using a tracked chassis, has a climbing ability of ≥35°, and is used to provide stable movement and vibration isolation; The environmental perception submodule is equipped with a lidar and a binocular camera to output point cloud and image data; The navigation and positioning submodule is equipped with a GNSS-RTK module and combines SLAM and RTK fusion positioning algorithms. In outdoor environments, RTK positioning is used first, and when satellite signals are rejected, it automatically switches to pure SLAM mode to output centimeter-level real-time pose.

[0015] In a preferred embodiment, the sensor integration module includes: The elevation measurement submodule uses a level instrument and achieves automatic digital readings through the SDK interface. The automatic leveling submodule employs a dual-axis tilt sensor and closed-loop control. The gimbal alignment sub-module uses a rotating gimbal that supports 360° continuous horizontal rotation and ±45° vertical tilt. The inertial navigation attitude submodule uses the IMS inertial navigation module, which includes a three-axis gyroscope and accelerometer, to output the robot's attitude angle and acceleration data in real time.

[0016] In a preferred embodiment, the spatiotemporal registration and synchronization module includes: The time synchronization unit adopts dynamic timestamp synchronization technology, with the edge computing box as the main clock source. It uses the PTP protocol and PPS pulse signal to control the time synchronization accuracy of multiple sensors within 1ms. The spatial calibration unit adopts a feature-point-based vision-inertial navigation joint calibration method to calculate the spatial transformation matrix of the lidar, binocular camera and inertial navigation module relative to the robot base coordinate system. The data preprocessing unit uses the 3σ criterion to remove gross errors in the acquired data and performs temperature and vibration compensation based on ambient temperature sensor data and vibration data from the IMS inertial navigation module.

[0017] In a preferred embodiment, the autonomous measurement and path planning module includes: Path planning unit, improving the A* algorithm for planning movement paths; The automatic operation unit is used to control the robot to perform coarse leveling of the base, pre-turning of the gimbal, fine leveling, and automatic reading of the level instrument after it arrives at the measuring point. The obstacle avoidance unit uses an improved YOLOv5 algorithm to dynamically identify obstacles and adjust the path in real time.

[0018] In a preferred embodiment, the multi-source data fusion processing module includes: A tightly coupled fusion unit is used to construct a Kalman filter state equation, using the attitude angle and acceleration output by the IMS inertial navigation module as state prediction quantities and the elevation change observed by the level instrument as measurement update quantities. An adaptive weighting unit calculates the signal-to-noise ratio (SNR) of each sensor's data and dynamically adjusts the weighting coefficients of the observation matrix based on the SNR. The drift correction unit uses the elevation observations from the level instrument to correct the zero drift of the inertial navigation module in real time, and outputs high-precision pose results.

[0019] In a preferred embodiment, the measurement result output and quality assessment module includes: The adjustment calculation unit uses the Helmert variance component estimation method to improve the indirect adjustment algorithm; The error assessment unit is equipped with a sliding window variance detection mechanism to identify random errors and automatically remove outlier data points. The transmission and display unit connects to the edge computing box via an RS485 interface and uses the 5G / WiFi network and MQTT communication protocol to push the measurement quality report to the cloud server and the terminal GIS map.

[0020] The present invention provides a robot-based multi-sensor fusion measurement method and system, which has the following beneficial effects: 1. From robot movement, automatic leveling, automatic aiming, data processing to cloud visualization, no human intervention is required. Measurement efficiency is increased by 3 to 5 times for the same number of measurement points, significantly reducing the cost of manual operation and maintenance in the field.

[0021] 2. By integrating the level instrument and the IMS inertial navigation module, and combining spatiotemporal registration, adaptive weighted fusion, and Kalman filtering algorithms, the elevation measurement accuracy reaches ±0.3mm / km, meeting the high-precision requirements of first-order leveling.

[0022] 3. Through inertial navigation compensation, vibration and interference resistance and dynamic leveling design, it effectively resists the environmental influences of vibration, wind disturbance, temperature change and other factors, and maintains stable and reliable measurement performance in complex engineering scenarios.

[0023] 4. It adopts a modular integrated architecture, is compatible with multiple types of sensor access, and can be widely used in the automated measurement of complex scenarios such as water conservancy dams and large-scale infrastructure, taking into account both versatility and future functional expansion. Attached Figure Description

[0024] The accompanying drawings, which are provided to further illustrate the invention and constitute a part of this invention, do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall structure of the robot of the present invention; Figure 2 This is a front view of the robot of the present invention; Figure 3 For Velodyne VLP-16 lidar and Basler binocular camera; Figure 4 It is an LS15 digital level. Figure 5 It is a Y200RA type rotary gimbal; Figure 6 A simplified schematic diagram of the ALP-01 automatic leveling base and its structure; Figure 7 To accurately identify the data interface diagram during the process; Figure 8 For dam surface map construction; Figure 9 A schematic diagram of robot measurement; Figure 10 This is a schematic diagram of the robot's on-site debugging. Figure 11 This is the overall flowchart for Example 2; Figure 12 This is a schematic diagram of laser SLAM matching for robots.

[0025] Among them, the tracked robot chassis 1, lidar 2, binocular camera 3, level 4, rotating gimbal 5, and leveling base 6. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0027] Example 1: A robot-based multi-sensor fusion measurement system includes: The robot mobile platform module provides autonomous movement, environmental perception, and precise positioning capabilities, and outputs real-time pose and obstacle information. The sensor integration module is used to integrate high-precision elevation measurement, automatic leveling, gimbal rotation, and attitude sensing units, and output multi-dimensional raw measurement data; The spatiotemporal registration and synchronization module is used to unify the timestamps of multiple sensors and calibrate the spatial coordinates, and to complete data preprocessing and gross error removal; The multi-source data fusion processing module is used to perform tight-coupled fusion, adaptive weighting and filtering correction on heterogeneous data, and output high-precision fusion results; The autonomous measurement and path planning module is used to automate the entire process of autonomous navigation, path optimization, automatic leveling, automatic aiming, and automatic reading. The measurement results output and quality assessment module is used to complete data adjustment, error correction, quality assessment and cloud transmission, and output standardized measurement reports.

[0028] In this invention, a tracked robot chassis is used as the carrier, such as... Figure 1 and 2 As shown, the lidar and binocular camera are fixed to the chassis with bolts, and power and data communication are achieved through an aviation plug; as Figure 4 As shown, an LS15 level instrument is selected. The level instrument, rotating head, automatic leveling base, and other sensors form a precision measurement system. Figure 4~6 As shown in Table 1.

[0029] The automatic leveling base is placed at the bottom of the system to achieve leveling. The rotating gimbal is located in the middle and can drive the rotation of the level instrument to align the level instrument objective with the leveling rod. The LS15 level instrument is placed at the top for precise leveling and reading measurement data. The advantage of this layout is that the rotation of the gimbal will not cause the entire measurement system to shift in the vertical direction, resulting in systematic measurement errors.

[0030]

[0031] The system utilizes sensors such as LiDAR and binocular cameras to achieve autonomous navigation and path planning. After arriving at the measurement point, the system is leveled using a leveling base, centered with the leveling instrument and leveling rod using a rotating gimbal, and precisely leveled using an LS15 level. The leveling program then automatically takes measurements, generates reports, and completes the alignment process as follows: Figure 7 As shown.

[0032] The detailed descriptions of each module are as follows: (1) The robot mobile platform module includes: Tracked chassis module, using a tracked chassis, has a climbing ability of ≥35°, and is used to provide stable movement and vibration isolation; The environmental perception submodule is equipped with a lidar and a binocular camera to output point cloud and image data; The navigation and positioning submodule is equipped with a GNSS-RTK module and combines SLAM and RTK fusion positioning algorithms. In outdoor environments, RTK positioning is used first, and when satellite signals are rejected, it automatically switches to pure SLAM mode to output centimeter-level real-time pose.

[0033] The hardware configuration is as follows: Tracked robot chassis (IP68 protection rating, climbing ability ≥35°). LiDAR (Velodyne VLP-16, scanning frequency 20Hz); Dual-lens camera (Basler ace acA2000-50gc, 2 megapixels); Built-in GNSS-RTK module (HuaCe i70, horizontal accuracy 8mm+1ppm).

[0034] The technology innovatively integrates mainstream SLAM navigation technology and RTK navigation technology, and simulates and models the actual situation of the dam body to achieve autonomous navigation and positioning and intelligent planning functions.

[0035] (2) The sensor integration module includes: The elevation measurement submodule uses a level instrument and achieves automatic digital readings through the SDK interface. The automatic leveling submodule uses a dual-axis tilt sensor and closed-loop control. In this embodiment, the dual-axis tilt sensor is Murata SCL3300, with a leveling range of ±11° (expandable to ±15° via mechanical limit release); closed-loop control response time: <2 seconds to achieve ±3" accuracy.

[0036] Three-level precision control is adopted, and the precision control parameters are shown in Table 2.

[0037]

[0038] The maximum horizontal adjustment range of the two axes of the automatic leveling base is ±11°, and the horizontal accuracy after the two axes are adjusted is ≤±5". It can be used with the LS15 automatic level to achieve a horizontal accuracy of <±3", which meets the requirements of first-order leveling measurement.

[0039] The gimbal alignment sub-module uses a rotating gimbal that supports 360° continuous horizontal rotation and ±45° vertical tilt.

[0040] The technical parameters of the rotating gimbal are: repeatability: ±0.005°; maximum rotation speed: 60° / s (S-curve acceleration and deceleration control).

[0041] The combination of a rotating pan-tilt head and a binocular camera ensures the hardware foundation for the automatic leveling technology, while the SURF and FLANN scale positioning algorithms ensure the software foundation for the automatic leveling technology.

[0042] The inertial navigation attitude submodule uses the IMS inertial navigation module, which includes a three-axis gyroscope and accelerometer, to output the robot's attitude angle and acceleration data in real time.

[0043] (3) The spatiotemporal registration and synchronization module includes: The time synchronization unit adopts dynamic timestamp synchronization technology, with the edge computing box as the main clock source. It uses the PTP protocol and PPS pulse signal to control the time synchronization accuracy of multiple sensors within 1ms. The spatial calibration unit adopts a feature-point-based vision-inertial navigation joint calibration method to calculate the spatial transformation matrix of the lidar, binocular camera and inertial navigation module relative to the robot base coordinate system. The data preprocessing unit uses the 3σ criterion to remove gross errors in the acquired data and performs temperature and vibration compensation based on ambient temperature sensor data and vibration data from the IMS inertial navigation module.

[0044] (4) The autonomous measurement and path planning module includes: Path planning unit, improving the A* algorithm for planning movement paths; The automatic operation unit is used to control the robot to perform coarse leveling of the base, pre-turning of the gimbal, fine leveling, and automatic reading of the level instrument after it arrives at the measuring point. The obstacle avoidance unit uses an improved YOLOv5 algorithm to dynamically identify obstacles and adjust the path in real time.

[0045] (5) The multi-source data fusion processing module includes: A tightly coupled fusion unit is used to construct a Kalman filter state equation, using the attitude angle and acceleration output by the IMS inertial navigation module as state prediction quantities and the elevation change observed by the level instrument as measurement update quantities. An adaptive weighting unit calculates the signal-to-noise ratio (SNR) of each sensor's data and dynamically adjusts the weighting coefficients of the observation matrix based on the SNR. The drift correction unit uses the elevation observations from the level instrument to correct the zero drift of the inertial navigation module in real time, and outputs high-precision pose results.

[0046] (6) The measurement result output and quality assessment module includes: The adjustment calculation unit uses the Helmert variance component estimation method to improve the indirect adjustment algorithm; The error assessment unit is equipped with a sliding window variance detection mechanism to identify random errors and automatically remove outlier data points. The transmission and display unit connects to the edge computing box via an RS485 interface and uses the 5G / WiFi network and MQTT communication protocol to push the measurement quality report to the cloud server and the terminal GIS map.

[0047] The workflow of the system of this invention is shown in Table 3.

[0048]

[0049] Example 2: This embodiment was implemented on the dam face of a concrete gravity dam on the main stream of the Yangtze River. 28 first-order leveling monitoring points were deployed in the monitoring area. The operating environment experienced a diurnal temperature range of -5℃ to 25℃, with strong winds affecting the dam crest, satellite signals blocked by the underground corridor walls, and GNSS-RTK lockout occurring frequently. Dynamic obstacles such as inspection vehicles and construction personnel were present on the ground, and the robot's tracked movement caused continuous micro-vibrations in its chassis. The terrain modeling is as follows: Figure 8 As shown, the measurement process is as follows: Figure 9 As shown, the on-site commissioning process is as follows: Figure 10 As shown in the figure. This embodiment provides a multi-sensor fusion measurement method based on a robot, and its flowchart is as follows. Figure 11 As shown, it includes the following steps: S1. System initialization and self-test: Completes self-test of robot chassis, sensors, and communication links.

[0050] Specifically, after power-on, the edge computing box traverses the tracked chassis, GNSS-RTK, lidar, binocular camera, IMS inertial navigation, level, automatic leveling base, rotating gimbal, temperature and vibration sensor, and 5G communication unit to verify power supply, data communication, and hardware limits. If a self-test is abnormal, a fault code is output and the machine stops. If everything is normal, it proceeds to the next step. S2. Complete the timestamp alignment of data from each sensor and the calibration of the coordinate system extrinsic parameters.

[0051] Dynamic timestamp synchronization technology is employed, with the edge computing box as the main clock source for time synchronization of various sensors, using feature point-based vision. Inertial navigation calibration method to achieve unification of external and internal parameters of multiple sensors.

[0052] Dynamic timestamp synchronization, visual The complete inertial navigation calibration operation procedure is as follows: (1) Dynamic timestamp synchronization: The robot edge computing box is used as the main clock source. An IEEE 1588 PTP master clock is deployed to send synchronization messages to all sensors. The edge computing box outputs a 1Hz PPS pulse hard synchronization signal to GNSS-RTK, IMS inertial navigation, lidar, binocular camera, and level. The local clock of each sensor corrects the offset according to the PTP message, and the PPS pulse triggers the synchronization sampling. The header of each data frame is embedded with a unified master clock timestamp. The asynchronously acquired data is interpolated and aligned to control the time synchronization error of all sensors within 1ms. (2) Feature point-based vision Joint Inertial Navigation Calibration: A checkerboard calibration board is placed in the robot's working area, and the gimbal is controlled to take pictures from multiple angles; the binocular camera extracts the sub-pixel corner features of the checkerboard, the lidar extracts the planar point cloud features of the calibration board, and the IMS inertial navigation system synchronously acquires the three-axis attitude at the corresponding time; the BA bundle adjustment method is used to jointly optimize the camera intrinsic parameters, camera-inertial navigation rotation and translation extrinsic parameters, and lidar-base transformation matrix, unify all sensors to the robot chassis world coordinate system, and output a unified spatial transformation matrix to complete the coordinate system registration.

[0053] This step is divided into two main sub-processes: dynamic timestamp hard synchronization and joint spatial calibration of vision-inertial navigation-LiDAR. It solves the problems of timing misalignment among multiple sensors and inconsistent spatial references. This is different from the simple solutions of existing technologies that only use software for simple timestamp alignment and separate calibration for each device. The details are as follows: Sub-process 1: Dynamic timestamp synchronization 1. The robot's internal edge computing box serves as the globally unique master clock, equipped with the IEEE1588 PTP master clock program, to synchronously send synchronization messages with a period of 100ms to GNSS-RTK, IMS inertial navigation, lidar, binocular camera, and level. 2. The edge computing box outputs a 1Hz standard PPS second pulse hard synchronization signal line from its hardware I / O port, which is connected to the synchronization trigger pins of all sensors respectively. The hardware pulse synchronization has a higher priority than the software PTP message, eliminating timing deviations caused by software delays. 3. Each sensor's local clock calculates the clock offset in real time based on the PTP message and corrects the local sampling time frame by frame; each round of sampling is uniformly triggered by the PPS pulse for synchronous acquisition, and all sensors synchronously output a frame of raw data; 4. For a small amount of asynchronous delayed data, a linear interpolation algorithm is used for timestamp alignment, and the timestamp of the edge computing box master clock is uniformly embedded. Finally, the time synchronization error of all sensors is stably controlled within 1ms, ensuring the timing consistency of subsequent multi-source data fusion.

[0054] Sub-process 2: Visual-Inertial Joint Spatial Calibration Based on Feature Points 1. Place a standard 12×9 checkerboard calibration board within 1.5 to 3 meters in front of the robot on site, with the plane of the calibration board perpendicular to the robot's central axis; the program automatically controls the rotating gimbal to take images of the calibration board at 12 different pitch and horizontal angles; 2. After the binocular camera acquires images, a sub-pixel corner extraction algorithm is used to extract all the internal corner features of the checkerboard; the lidar synchronously acquires the planar point cloud of the calibration board at the corresponding time, and extracts the spatial planar features of the calibration board through planar fitting; the IMS inertial navigation system synchronously acquires three-axis roll, pitch, and heading attitude data at each shooting time. 3. The BA bundle adjustment optimization algorithm is used to jointly solve three sets of parameters: the internal distortion parameters of the binocular camera, the rotation and translation extrinsic parameter matrix of the binocular camera relative to the IMS inertial navigation system, and the spatial transformation matrix of the lidar relative to the robot chassis base. 4. Transform all coordinate systems of the lidar, binocular camera, IMS inertial navigation, and level to the tracked chassis world coordinate system, output a globally unified spatial transformation matrix and store it locally in the edge computing box to complete spatial registration; all subsequent acquired data are output based on the same coordinate system to eliminate multi-sensor spatial offset errors.

[0055] S3. Based on SLAM and RTK fusion localization technology, an improved A* algorithm is used to plan the movement path and drive the robot to reach the target measurement point, specifically including: 1. Load the global map of the dam body and the parameters of the measuring points. The edge computing box reads a pre-built 3D laser point cloud map of the dam body. After preprocessing the raw LiDAR data, matching is performed, which is a crucial step. This matching primarily involves locating the corresponding position of a point cloud data point in the local environment within the existing map. During SLAM, the point cloud data currently acquired by the LiDAR (red portion) needs to be matched and stitched into the existing map, such as... Figure 12 As shown. Then import the list of monitoring points for this inspection task, and bind the monitoring point priority weight w(n) to each monitoring point. High-risk dam cracks and settlement monitoring points are assigned higher weights and priority is given to planning the passage route.

[0056] 2. Plan the movement path using the improved A* algorithm. Construct a comprehensive cost function that includes distance cost, energy cost, and measurement point priority weights: f ( n )= g ( n )+ h ( n )+ w ( n ); in, g ( n) represents the movement cost from the starting point to the current node. h ( n ) represents the heuristic distance from the current node to the target node. w ( n ) represents the priority weight of the measurement point.

[0057] 3. Real-time GNSS signal quality assessment and automatic switching between dual-mode positioning. The edge computing box analyzes the GNSS-RTK positioning solution status in real time, and determines the number of available satellites and the positioning PDOP value. When PDOP < 3 and the number of satellites ≥ 8, the RTK signal is deemed available, and the RTK centimeter-level coordinates are used as the global positioning reference. When satellites are blocked or PDOP > 5, the signal is deemed rejected, and the pure SLAM positioning mode is switched.

[0058] Open dam crest area: ≥8 visible satellites and PDOP<3 are used to determine the validity of the RTK signal. The centimeter-level plane and elevation coordinates output by RTK are used as the global positioning reference, and the position is corrected by superimposing the lidar point cloud. Underground corridors and dam wall obstruction areas: ≤4 visible satellites, PDOP>5, satellite signal rejection is determined, automatically switching to pure laser SLAM positioning mode. Complete operation process: ① The IMS inertial navigation system outputs three-axis acceleration and angular velocity, which are used to calculate the robot's short-time pose via strapdown, serving as a pose prior. ②The Velodyne VLP-16 lidar outputs 16-line point clouds in real time. The ICP iterative nearest point algorithm is used to match the current frame point cloud with the pre-built dam body point cloud map and solve the rotation and translation transformation matrix between point clouds. ③ Use point cloud matching constraints as observation values ​​to correct the inertial navigation recursive cumulative error, and fuse the output to achieve decimeter-level continuous positioning results in satellite-free scenarios.

[0059] 4. Real-time dynamic obstacle recognition and local path replanning The binocular camera inputs images in real time to the improved YOLOv5 lightweight recognition model to identify dynamic obstacles such as pedestrians in the corridor, inspection vehicles, and piled-up debris. If an obstacle appears at a preset path node, the program automatically increases the distance cost g(n) of that grid node, triggers the improved A* algorithm for local replanning, generates a detour path, and the chassis adjusts its direction of travel in real time to avoid the obstacle and return to the original inspection route. 5. The robot autonomously travels along the planned path, updating and fusing the positioning coordinates in real time. When the deviation between the positioning coordinates and the target measurement point coordinates is less than 3cm, it is determined that the measurement point has been reached, the chassis brakes and locks, and the robot enters the S4 automatic measurement process.

[0060] S4. After adjusting the level to a horizontal position, trigger the synchronous acquisition command to collect elevation data using the level. Record the robot's attitude angles and acceleration in real time using the IMS inertial navigation module. Obtain environmental perception data by synchronously scanning the surrounding environment using LiDAR and binocular cameras, and use GNSS... The RTK module acquires location data.

[0061] The specific operation of the three-stage automatic leveling and stabilization of the base is as follows: 1. Coarse leveling: Dual-axis tilt sensors collect the tilt angle of the base in real time, and the motor quickly drives the leveling platform to adjust the tilt of the base to within ±1, which takes less than 5 seconds. 2. Fine leveling: Activate the piezoelectric ceramic micro-motion device to slightly compensate for the tilt of the two axes, achieving a leveling accuracy of ±10″; 3. Dynamic voltage stabilization: PID adaptive closed-loop control is activated to continuously compensate for minor chassis tilt in real time, maintaining the static level accuracy of the level instrument at ±3″ / 10min, meeting the requirements of the first-order leveling benchmark.

[0062] The complete process of automatic elevation acquisition by a leveling instrument is as follows: The system controls the pan-tilt head to rotate 360° horizontally and tilt ±45° vertically, while the binocular camera acquires images of the leveling rod in real time. The SURF algorithm extracts the scale features, and the FLANN matching library matches the standard scale template to achieve sub-pixel-level scale positioning. Based on the crosshair and scale deviation, a closed-loop adjustment command for the pan-tilt head is generated, iteratively correcting the pan-tilt head attitude until the crosshairs are precisely aligned with the leveling rod graduations. The level instrument reads the digital scale values ​​through the built-in SDK and outputs the raw elevation observation data.

[0063] The specific steps are as follows: 1. Image acquisition: The rotating gimbal drives the level to perform a wide-range horizontal and vertical scan, while the binocular camera continuously acquires high-definition images of the leveling rod in front; 2. SURF feature extraction: The image is converted to grayscale, and stable SURF feature points of the level scale and digital division are extracted, while irrelevant interference features of the wall and ground are filtered out; 3. FLAN template matching: The extracted scale features are quickly matched with the local pre-stored standard leveling scale template using FLANN to filter out the effective scale matching areas; 4. Subpixel-level scale positioning: Subpixel interpolation refines the successfully matched scale area to accurately locate the pixel coordinates of the leveling ruler's graduation scale. 5. Generation of gimbal closed-loop adjustment commands: Calculate the difference in horizontal and vertical offset between the center pixel coordinates of the level crosshair and the target scale coordinates, and convert it into gimbal horizontal and pitch angle adjustment commands; 6. Gimbal fine-tuning completes aiming: The gimbal receives the command and rotates slightly to correct its attitude, iterating 3 to 5 times until the crosshairs are accurately aligned with the standard graduations of the leveling ruler, with aiming repeatability accuracy of ±0.005°.

[0064] After the gimbal is aligned, the edge computing box issues a unified synchronous acquisition command, and all sensors synchronously sample 5 sets of continuous data based on the unified timestamps of S2: 1. The LS15 level instrument reads digital raw elevation readings through its built-in SDK; 2. The IMS inertial navigation system synchronously acquires high-frequency vibration and attitude data of three-axis acceleration and three-axis gyroscope angular velocity; 3. The lidar outputs a 3D point cloud of the surrounding environment of the current measurement point; 4. The binocular camera retains the original image for ruler recognition; 5. GNSS-RTK synchronously outputs the current measurement point positioning coordinates; 6. The environmental temperature and vibration sensor synchronously collects real-time temperature and chassis vibration amplitude data at the measuring points; All raw data are tagged with a unified timestamp and measurement point number, and cached in the local memory of the edge computing box.

[0065] S5. Use the 3σ criterion to remove gross errors in the collected data and perform temperature and vibration compensation based on environmental parameters.

[0066] Includes the following steps: Step 1: Outlier Removal Based on 3σ Criterion Statistical calculations were performed on 5 sets of raw elevation observations from continuous level instruments to solve for the sample mean μ and sample standard deviation σ. A threshold interval [μ-3σ, μ+3σ] was set, and elevation data exceeding the interval were judged as gross errors and directly discarded. If there were less than 3 sets of valid data after the discarding, the robot automatically triggered a second synchronous acquisition to supplement the valid observation samples.

[0067] Step 2: Compensation for static error caused by thermal expansion of the leveling rod 1. Read the real-time temperature from the environmental sensor. T 0, the system presets a standard reference temperature for measurement. T s =20℃ Calculate the temperature difference ΔT =T 0 -T s ; 2. Substitute into the thermal expansion compensation formula ΔL=αLΔT The linear expansion coefficient of steel leveling ruler α=1.2×10 -5 / ℃ , L This is the original reading of the level instrument; 3. Calculate the corrected elevation L'=L ΔL This eliminates system measurement errors caused by the expansion and contraction of the leveling rod due to temperature changes.

[0068] Step 3: Compensation of first-order linear error model for chassis micro-vibration Constructing a first-order linear vibration error model ΔH v = k 1 a x +k 2 a y +k 3 oh z In the formula a x 、a y For the horizontal vibration acceleration of the inertial navigation system, oh z The angular velocity of vibration about the vertical axis, k 1 、k 2 、k 3 represents the vibration coupling coefficient obtained from calibration; vibration deviation is calculated by inputting high-frequency vibration data from the inertial navigation system in real time. ΔH v Subtract from the original reading of the level instrument ΔH v It is used to suppress reading drift caused by track movement and chassis micro-vibration.

[0069] The specific steps are as follows: 1. Extract horizontal X and Y axis vibration accelerations synchronously acquired by IMS inertial navigation. a x , a angular velocity of vertical axis vibration ω_z ; 2. Use the three sets of vibration coupling coefficients obtained from the equipment's factory calibration. k 1 、k 2 、k 3; 3. Substitute into the vibration error model: ΔH v = k 1 a x +k 2 a y +k 3 oh zReal-time calculation of elevation deviation caused by chassis vibration ΔH v ; 4. Elevation values ​​after temperature compensation L' Subtracting vibration deviation ΔH v We obtained clean elevation observations that eliminated temperature and vibration interference, which were then used for subsequent data fusion.

[0070] S6. The elevation observations of the level instrument are fused with the attitude angle and acceleration data. The adaptive weighting algorithm and Kalman filter are used to iteratively correct the inertial navigation zero drift and output high-precision pose results.

[0071] Complete implementation steps for adaptive weighted Kalman filter correction of inertial navigation zero drift: S601. Construct the Kalman filter state vector X=[φ,θ,ψ,a x ,a y ,a z ,b gx ,b gy ,b gz ] ,in, f、θ、ψ These are the attitude angles along the roll, pitch, and yaw axes, respectively. a x 、a y 、a z These are triaxial accelerations. b gx 、b gy 、b gz The parameters to be estimated for zero drift of the gyroscope; state transition equation X k =F X k-1 +W k , F Here is the state transition matrix for the inertial navigation strapdown system. W k This is process noise.

[0072] S602, The measurement equation is constructed as follows: Z k =H X k +V kMeasurement value Z k For continuous sampling of elevation changes by a level instrument ΔH , H For the observation matrix, V k To observe noise.

[0073] S603, Adaptive Weighted Calculation: Calculate the real-time signal-to-noise ratio (SNR) of inertial navigation and leveling instrument data respectively. IMU SNR LEVEL The higher the signal-to-noise ratio (SNR), the larger the corresponding weight coefficient of the observation matrix. The formula for calculating the weight coefficient is: w = SNR / (SNR) IMU +SNR LEVEL The weights of each dimension of the H matrix are dynamically updated.

[0074] S604, Kalman Iterative Update: First, time prediction is completed using the state transition matrix, then measurement updates are completed using the adaptive weighted observation matrix, iteratively solving for gyroscope zero drift. b g .

[0075] The specific steps are as follows: 1. Time prediction: Utilizing the state at the previous moment X k-1 State transition matrix F Predict the prior state at the current moment. k Synchronously update the prior covariance matrix; 2. Measurement Update: Input the adaptively weighted observation matrix H Elevation observations Z k Calculate the Kalman gain; 3. Correction State: The prior state is corrected by combining the observation residuals, and the optimal posterior state estimate is output. The zero drift of the three-axis gyroscope is solved iteratively in parallel. b gx 、b gy 、b gz .

[0076] S605. Substitute the zero-drift compensation value obtained through iteration back into the original angular velocity data of the IMS inertial navigation system to eliminate long-term accumulated drift, output the fused and corrected high-precision pose and elevation data, and store it in the local storage unit.

[0077] S7. Use the Helmert variance component estimation method for indirect adjustment, and combine it with sliding window variance detection to identify random errors and generate a measurement quality report.

[0078] Specifically, the following steps are included: Step 1: Random Error Detection in Sliding Window 1. Set a fixed sampling length of N=20 sets of fused elevation data for the sliding window, and slide the window frame by frame to capture continuous observation samples; 2. Calculate the variance of elevation residuals within each window. σ² The system has a preset error judgment threshold. s 0 = 0.1mm; 3. Decision logic: If σ²>σ If the variance is 0, it is determined that there is a random error in the current window, and the robot automatically performs a retest; if the variance still exceeds the limit after two consecutive retests, all abnormal observation data in that segment are directly removed; if the variance within the window is less than the threshold, the data is determined to be valid and enters the adjustment calculation.

[0079] Step 2: Iterative indirect adjustment of Helmert variance component estimation 1. Initialize the initial prior variances of the three types of observations: leveling instrument, IMS inertial navigation system, and GNSS-RTK. 2. Construct the indirect adjustment method equation and substitute it into the observation data from all valid measurement points; 3. Iteratively execute the Helmert variance component estimation algorithm, continuously correct the prior variance of the observations from the three types of sensors, balance the weights of the multi-source observation data, until the variance converges; 4. Solve for the optimal plane coordinates and optimal elevation estimates of the measuring points, and simultaneously calculate the elevation mean error, standard deviation per kilometer of round trip measurement, observation residual distribution, and positioning error index of each measuring point.

[0080] Sub-step 3: Generate a standardized measurement quality report The edge computing box automatically integrates all adjustment calculation results and generates a structured measurement quality report. The report includes: measurement point number, fused elevation value, plane coordinates, elevation accuracy index, original observation data, temperature and vibration compensation parameters, error elimination records, sliding window variance detection results, and positioning mode records (RTK / SLAM).

[0081] S8. Transmit the measurement quality report to the edge computing box on the robot, push it to the cloud server using the MQTT communication protocol, generate a visual measurement report on the cloud server, and overlay the measurement point location on the GIS map of the terminal device.

[0082] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-sensor fusion measurement method based on a robot, characterized in that, Includes the following steps: S1. System initialization and self-test; S2. Complete the timestamp alignment of data from each sensor and the calibration of the coordinate system extrinsic parameters; S3. Based on the SLAM and RTK fusion localization technology, the improved A* algorithm is used to plan the movement path and drive the robot to reach the target measurement point; S4. After adjusting the level to a horizontal position, trigger the synchronous acquisition command to collect elevation data using the level. Record the robot's attitude angles and acceleration in real time using the IMS inertial navigation module. Obtain environmental perception data by synchronously scanning the surrounding environment using LiDAR and binocular cameras, and use GNSS... The RTK module acquires location data; S5. Use the 3σ criterion to remove gross errors in the collected data and perform temperature and vibration compensation based on environmental parameters; S6. The elevation observations of the level instrument are fused with the attitude angle and acceleration data. The adaptive weighting algorithm and Kalman filter are used to correct the zero drift of the inertial navigation system and output high-precision pose results. S7. Indirect adjustment is performed using the Helmert variance component estimation method, combined with sliding window variance detection to identify random errors and generate a measurement quality report.

2. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, In step S2, dynamic timestamp synchronization technology is used, with the edge computing box as the main clock source, to synchronize the time of each sensor, and feature point-based vision is employed. Inertial navigation calibration method to achieve unification of external and internal parameters of multiple sensors.

3. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, In step S3, planning the movement path using the improved A* algorithm specifically includes: Construct a comprehensive cost function that includes distance cost, energy cost, and measurement point priority weights: f ( n )= g ( n )+ h ( n )+ w ( n )； in, g ( n ) represents the movement cost from the starting point to the current node. h ( n ) represents the heuristic distance from the current node to the target node. w ( n ) represents the priority weight of the measurement points; The system analyzes the GNSS-RTK positioning status in real time, determines the number of available satellites and the positioning PDOP value. When PDOP < 3 and the number of satellites ≥ 8, the RTK signal is deemed available, and the RTK centimeter-level coordinates are used as the global positioning reference. When satellites are blocked or PDOP > 5, the signal is deemed rejected, and the system switches to pure SLAM positioning mode, using lidar point cloud matching and inertial navigation calculation for positioning.

4. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, In step S4, collecting elevation data using a level specifically includes: The control of the rotating gimbal drives the level to rotate horizontally and tilt vertically. With the assistance of binocular camera vision recognition, the center of the crosshairs of the level objective lens is automatically aligned with the scale of the leveling rod, and the level outputs a digital elevation reading.

5. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, In step S5, performing temperature and vibration compensation based on environmental parameters specifically includes: (1) Temperature compensation: real-time temperature of the environment is collected from the ambient temperature sensor. T 0, reference measurement temperature T s Temperature difference ΔT=T 0 -T s Substitute into the thermal expansion compensation formula ΔL=αLΔT ,in, α The coefficient of linear expansion of the leveling rod material. L The original reading length of the level instrument, and the corrected elevation. L'=L-ΔL ; (2) Vibration compensation: Constructing a first-order linear vibration error model ΔH v = k 1 a x +k 2 a y +k 3 ω z In the formula a x 、a y For the horizontal vibration acceleration of the inertial navigation system, ω z The angular velocity of vibration about the vertical axis, k 1 、k 2 、k 3 represents the vibration coupling coefficient obtained from calibration; vibration deviation is calculated by inputting high-frequency vibration data from the inertial navigation system in real time. ΔH v Subtract from the original reading of the level instrument ΔH v It is used to suppress reading drift caused by track movement and chassis micro-vibration.

6. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, In step S6, the use of adaptive weighting algorithm and Kalman filter to correct inertial navigation zero drift specifically includes: S601. Construct the Kalman filter state vector: X= [ φ,θ,ψ,a x ,a y ,a z ,b gx ,b gy ,b gz ]； in, φ, θ, ψ These are roll angle, pitch angle, and yaw angle, respectively. a x 、a y 、a z These are triaxial accelerations. b gx 、b gy 、b gz The parameters to be estimated for zero drift of the gyroscope; The state transition equation is: X k =F X k-1 +W k ； in, F Here is the state transition matrix for the inertial navigation strapdown system. W k This is process noise; S602, the measurement equation is: Z k =H X k +V k ； Among them, the measured values Z k For continuous sampling of elevation changes by a level instrument ΔH , H For the observation matrix, V k To observe noise; S603, Adaptive Weighted Calculation: Calculate the real-time signal-to-noise ratio (SNR) of inertial navigation and leveling instrument data respectively. IMU SNR LEVEL Then calculate the weighting coefficients respectively: w IMU =SNR IMU / ( SNR IMU +SNR LEVEL )； w LEVEL =SNR LEVEL / ( SNR IMU +SNR LEVEL )； The observation matrix is ​​dynamically updated based on two sets of weights. H Internal coefficients for each dimension; dynamically update the weights of each dimension of the H matrix. S604, Kalman Iterative Update: First, time prediction is completed using the state transition matrix, then measurement updates are completed using the adaptive weighted observation matrix, iteratively solving for gyroscope zero drift. b gx 、b gy 、b gz ; S605. Substitute the zero-drift compensation value obtained through iteration back into the original angular velocity data of the IMS inertial navigation system, and output the fused and corrected high-precision pose and elevation data.

7. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, In step S7, the identification of random errors in conjunction with sliding window variance detection specifically includes: Set the sliding window size and calculate the variance σ² of the elevation residuals within the window window one window at a time. Preset error threshold σ 0, when the variance of the elevation residual σ² exceeds the preset threshold. σ When the value is 0, it is determined that there is a random error in the current data segment, and a retest instruction is triggered. If the variance still exceeds the limit after two consecutive retests, the abnormal data segment is removed. After gross error elimination, the Helmert variance component estimation iterative calculation is used to calculate the prior variances of the three types of observations: leveling instrument, inertial navigation, and RTK. The indirect adjustment method equation is constructed to solve for the optimal elevation and measurement point coordinates. A standardized measurement quality report is output, which includes the positional error, the standard deviation of elevation per kilometer for round trip measurements, and the residual distribution.

8. The multi-sensor fusion measurement method based on a robot according to claim 1, characterized in that, It also includes step S8, transmitting the measurement quality report to the edge computing box carried by the robot, pushing it to the cloud server using the MQTT communication protocol, generating a visual measurement report on the cloud server, and overlaying and displaying the measurement point locations on the GIS map of the terminal device.

9. A robot-based multi-sensor fusion measurement system, characterized in that, For performing the method according to any one of claims 1 to 8, comprising: The robot mobile platform module provides autonomous movement, environmental perception, and precise positioning capabilities, and outputs real-time pose and obstacle information. The sensor integration module is used to integrate high-precision elevation measurement, automatic leveling, gimbal rotation, and attitude sensing units, and output multi-dimensional raw measurement data; The spatiotemporal registration and synchronization module is used to unify the timestamps of multiple sensors and calibrate the spatial coordinates, and to complete data preprocessing and gross error removal; The multi-source data fusion processing module is used to perform tight-coupled fusion, adaptive weighting and filtering correction on heterogeneous data, and output high-precision fusion results; The autonomous measurement and path planning module is used to automate the entire process of autonomous navigation, path optimization, automatic leveling, automatic aiming, and automatic reading. The measurement results output and quality assessment module is used to complete data adjustment, error correction, quality assessment and cloud transmission, and output standardized measurement reports.

10. A robot-based multi-sensor fusion measurement system according to claim 9, characterized in that, The robot mobile platform module includes: Tracked chassis module, using a tracked chassis, has a climbing ability of ≥35°, and is used to provide stable movement and vibration isolation; The environmental perception submodule is equipped with a lidar and a binocular camera to output point cloud and image data; The navigation and positioning submodule is equipped with a GNSS-RTK module and combines SLAM and RTK fusion positioning algorithms. In outdoor environments, RTK positioning is used first, and when satellite signals are rejected, it automatically switches to pure SLAM mode to output centimeter-level real-time pose.

11. A robot-based multi-sensor fusion measurement system according to claim 9, characterized in that, The sensor integration module includes: The elevation measurement submodule uses a level instrument and achieves automatic digital readings through the SDK interface. The automatic leveling submodule employs a dual-axis tilt sensor and closed-loop control. The gimbal alignment sub-module uses a rotating gimbal that supports 360° continuous horizontal rotation and ±45° vertical tilt. The inertial navigation attitude submodule uses the IMS inertial navigation module, which includes a three-axis gyroscope and accelerometer, to output the robot's attitude angle and acceleration data in real time.

12. A robot-based multi-sensor fusion measurement system according to claim 9, characterized in that, The spatiotemporal registration and synchronization module includes: The time synchronization unit adopts dynamic timestamp synchronization technology, with the edge computing box as the main clock source. It uses the PTP protocol and PPS pulse signal to control the time synchronization accuracy of multiple sensors within 1ms. The spatial calibration unit adopts a feature-point-based vision-inertial navigation joint calibration method to calculate the spatial transformation matrix of the lidar, binocular camera and inertial navigation module relative to the robot base coordinate system. The data preprocessing unit uses the 3σ criterion to remove gross errors in the acquired data and performs temperature and vibration compensation based on ambient temperature sensor data and vibration data from the IMS inertial navigation module.

13. A robot-based multi-sensor fusion measurement system according to claim 9, characterized in that, The autonomous measurement and path planning module includes: Path planning unit, improving the A* algorithm for planning movement paths; The automatic operation unit is used to control the robot to perform coarse leveling of the base, pre-turning of the gimbal, fine leveling, and automatic reading of the level instrument after it arrives at the measuring point. The obstacle avoidance unit uses an improved YOLOv5 algorithm to dynamically identify obstacles and adjust the path in real time.

14. A robot-based multi-sensor fusion measurement system according to claim 9, characterized in that, The multi-source data fusion processing module includes: A tightly coupled fusion unit is used to construct a Kalman filter state equation, using the attitude angle and acceleration output by the IMS inertial navigation module as state prediction quantities and the elevation change observed by the level instrument as measurement update quantities. An adaptive weighting unit calculates the signal-to-noise ratio (SNR) of each sensor's data and dynamically adjusts the weighting coefficients of the observation matrix based on the SNR. The drift correction unit uses the elevation observations from the level instrument to correct the zero drift of the inertial navigation module in real time, and outputs high-precision pose results.

15. A robot-based multi-sensor fusion measurement system according to claim 9, characterized in that, The measurement result output and quality assessment module includes: The adjustment calculation unit uses the Helmert variance component estimation method to improve the indirect adjustment algorithm; The error assessment unit is equipped with a sliding window variance detection mechanism to identify random errors and automatically remove outlier data points. The transmission and display unit connects to the edge computing box via an RS485 interface and uses the 5G / WiFi network and MQTT communication protocol to push the measurement quality report to the cloud server and the terminal GIS map.

Images