High-altitude track AGV accurate positioning and map building method and system

Abstract

The invention discloses a high-altitude track AGV accurate positioning and map building method and system. The method comprises the following steps: collecting data through multiple sensors; constructing a fusion model, and dynamically distributing fusion weights of laser and visual SLAM based on visual image information entropy and laser point cloud geometric feature distribution; semantic features such as track planes and edge lines are extracted and matched with the prior model; and jointly inputting the weight fusion pose, the UWB constraint and the semantic constraint into a back-end graph for optimization, solving an optimal pose and updating the map. The system comprises corresponding modules. Through cooperation of dynamic weight distribution and semantic auxiliary optimization, the problems of low AGV positioning precision and poor robustness in a high-altitude orbit environment are solved, dynamic closed-loop updating of a map can be realized, and the automation and intelligence level of an intelligent detection system is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of automated guided vehicle navigation and mapping technology, and in particular to a method and system for achieving centimeter-level precise positioning and real-time high-precision map construction of AGVs in an indoor high-altitude track environment where GPS signals are absent. Background Technology

[0002] In large industrial plants, warehouses, and railway platforms, the safety inspection of elevated tracks (such as overhead crane tracks and material conveyor tracks) is crucial. Automated inspection using AGVs has become a trend, but it faces the following technical bottlenecks: GPS signal deficiency: Indoor environments completely block satellite signals; Unfavorable environmental characteristics: The repetitive structure and monotonous texture of elevated track areas make traditional single-vision or laser SLAM prone to feature matching failures or increased cumulative errors; High accuracy requirements: Inspection tasks require centimeter-level positioning accuracy to accurately mark defects; Dynamic map updates are necessary: ​​Infrastructure changes necessitate the system's ability to update maps online.

[0003] Existing solutions such as dead reckoning (accumulated error), single UWB positioning (insufficient accuracy and unable to map), and pre-scanning mapping (high cost and unable to adapt to changes) cannot meet the requirements of long-term, high-precision, and robust positioning and dynamic mapping in high-altitude orbit detection scenarios. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an improved method and system for precise positioning and map construction of high-altitude track AGVs in order to overcome the shortcomings of the existing technology.

[0005] The technical solution adopted by this invention to solve its technical problem is: a method for precise positioning and map construction of a high-altitude track AGV, comprising the following steps:

[0006] S1: Collect environmental data through a multi-sensor system mounted on the AGV;

[0007] S2: Construct a hierarchical multi-sensor fusion model based on the pose estimation results of front-end LiDAR SLAM and visual SLAM, and dynamically allocate the fusion weights of the two based on real-time environmental features.

[0008] S3: Extract the structural semantic features of the high-altitude track from sensor data in real time, and match these semantic features as strong constraints with a predefined prior model of the track structure;

[0009] S4: Input the front-end pose after weighted fusion, the absolute coordinate constraints provided by the UWB positioning module, and the semantic feature constraints obtained in step S3 into the back-end for graph optimization, solve for the optimal pose estimate of the AGV, and update the global map.

[0010] Preferably, the basis for dynamically allocating weights in step S2 includes: calculating the information entropy of the current visual image, and / or analyzing the geometric feature distribution of the current laser point cloud.

[0011] Preferably, the dynamic weight allocation in step S2 specifically involves: increasing the fusion weight of visual SLAM pose when the information entropy of the visual image is higher than the first threshold; and decreasing the fusion weight of visual SLAM and increasing the weight of UWB constraints in the backend optimization when the geometric feature distribution of the laser point cloud indicates that it is in a long straight corridor environment.

[0012] Preferably, the extraction of structural semantic features of the high-altitude track in step S3 specifically includes: segmenting the track plane and track edge lines from laser point cloud data; and / or extracting the track contour lines from visual images using an edge detection algorithm.

[0013] Preferably, the method further includes step S5: during the AGV's execution of the detection task, using the optimal pose obtained in step S4, a defect detection algorithm is used to identify new defects or structural changes on the track; the three-dimensional coordinates and feature descriptors of the defect or change feature are defined as new semantic landmarks; the semantic landmarks are registered and updated in real time to the global map for loop closure detection and pose correction in subsequent tasks.

[0014] On the other hand, the present invention provides a system for implementing the above method, comprising:

[0015] The data acquisition module includes a lidar, a visual sensor, an IMU and an odometry, and a UWB receiver; the adaptive fusion localization module is configured to: receive data from the data acquisition module, run laser SLAM and visual SLAM algorithms, and dynamically allocate fusion weights of the laser SLAM and visual SLAM pose results based on the information entropy of the current visual image and / or the geometric feature distribution of the laser point cloud.

[0016] The semantic processing module is configured to: extract the structural semantic features of the high-altitude track from the data of the data acquisition module, including at least the track plane and track edge line segmented from the laser point cloud, and match them with the prior model of the track structure;

[0017] The backend map optimization module is configured to receive the fused pose output by the adaptive fusion positioning module, the semantic constraints matched by the semantic processing module, and the absolute coordinate constraints provided by UWB, perform joint optimization, output the precise positioning of the AGV, and update the global map.

[0018] Preferably, the visual sensor is a binocular camera or an RGB-D camera.

[0019] Preferably, the system further includes a dynamic map management module, configured to define track defects or structural change features identified based on precise positioning as new semantic landmarks and register and update them in the global map in real time.

[0020] The beneficial effects of this invention are as follows: Compared with the prior art, this invention exhibits the following direct and significant beneficial effects:

[0021] 1. When facing challenging environments such as abrupt texture changes and long straight tracks in the factory, the system effectively avoids the failure of a single sensor through dynamic weight allocation and semantic / UWB constraints. The positioning trajectory is smooth and continuous, without significant jumps or drifts, and the measured accuracy meets the detection requirements.

[0022] 2. Without the need for manual switching between different scenes, the system can automatically perceive environmental characteristics (texture, geometry) and adjust the core algorithm strategy, demonstrating a high degree of autonomy and adaptability.

[0023] 3. Maps are no longer static costs generated from a one-time investment, but rather, through a closed-loop update mechanism, they continuously accumulate expert knowledge (defect landmarks) in each inspection task, becoming a continuously value-added "living" digital asset that directly empowers predictive maintenance decisions. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the principle of the present invention; Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] A method for precise positioning and map construction of an aerial track AGV, comprising the following steps:

[0028] S1: Collect environmental data through a multi-sensor system mounted on the AGV;

[0029] S2: Construct a hierarchical multi-sensor fusion model based on the pose estimation results of front-end LiDAR SLAM and visual SLAM, and dynamically allocate the fusion weights of the two based on real-time environmental features.

[0030] S3: Extract the structural semantic features of the high-altitude track from sensor data in real time, and match these semantic features as strong constraints with a predefined prior model of the track structure;

[0031] S4: Input the front-end pose after weighted fusion, the absolute coordinate constraints provided by the UWB positioning module, and the semantic feature constraints obtained in step S3 into the back-end for graph optimization, solve for the optimal pose estimate of the AGV, and update the global map.

[0032] The basis for dynamically allocating weights in step S2 includes: calculating the information entropy of the current visual image and / or analyzing the geometric feature distribution of the current laser point cloud.

[0033] In step S2, the dynamic weight allocation is as follows: when the information entropy of the visual image is higher than the first threshold, the fusion weight of the visual SLAM pose is increased; when the geometric feature distribution of the laser point cloud indicates that it is in a long straight corridor environment, the fusion weight of the visual SLAM is decreased and the weight of the UWB constraint in the back-end optimization is increased.

[0034] Step S3, which extracts the structural semantic features of the high-altitude track, specifically includes: segmenting the track plane and track edge lines from laser point cloud data; and / or extracting the track contour lines from visual images using edge detection algorithms.

[0035] The method also includes step S5: during the AGV's execution of the detection task, the optimal pose obtained in step S4 is used to identify new defects or structural changes on the track through a defect detection algorithm; the three-dimensional coordinates and feature descriptors of the defect or change feature are defined as new semantic landmarks; the semantic landmarks are registered and updated in real time to the global map for loop closure detection and pose correction in subsequent tasks.

[0036] On the other hand, the present invention provides a system for implementing the above method, comprising:

[0037] The data acquisition module includes a lidar, a visual sensor, an IMU and an odometry, and a UWB receiver; the adaptive fusion localization module is configured to: receive data from the data acquisition module, run laser SLAM and visual SLAM algorithms, and dynamically allocate fusion weights of the laser SLAM and visual SLAM pose results based on the information entropy of the current visual image and / or the geometric feature distribution of the laser point cloud.

[0038] The semantic processing module is configured to: extract the structural semantic features of the high-altitude track from the data of the data acquisition module, including at least the track plane and track edge line segmented from the laser point cloud, and match them with the prior model of the track structure;

[0039] The backend map optimization module is configured to receive the fused pose output by the adaptive fusion positioning module, the semantic constraints matched by the semantic processing module, and the absolute coordinate constraints provided by UWB, perform joint optimization, output the precise positioning of the AGV, and update the global map.

[0040] The visual sensor is either a binocular camera or an RGB-D camera.

[0041] The system also includes a dynamic map management module, which is configured to define track defects or structural change features identified based on precise positioning as new semantic landmarks and register and update them in the global map in real time.

[0042] This embodiment uses the inspection of overhead crane tracks in an industrial plant as an application scenario to illustrate the specific implementation process of the present invention.

[0043] I. System Hardware Setup

[0044] AGV Platform: A traveling trolley that can stably run on I-beam rails; Sensor Configuration: LiDAR: A 16-line LiDAR, tilted forward, used to scan the track and surrounding structure; Binocular Camera: Mounted next to the LiDAR, used to acquire track images; IMU and Encoder: Measures the AGV's motion posture and travel distance; UWB Positioning Tag: Mounted on the AGV, working in conjunction with four UWB base stations deployed within the factory; Processing Host: An industrial-grade computer responsible for running all calculation programs.

[0045] II. Workflow (Reference) Figure 1 )

[0046] Step 1: Data Collection

[0047] All sensors start working simultaneously: the lidar acquires environmental contour data 10 times per second; the binocular camera acquires 15 pairs of stereo images per second; the IMU and encoder acquire 100 motion data points per second; and the UWB system provides 20 position reference coordinates per second.

[0048] Step 2: Intelligent Fusion Positioning

[0049] The system employs a unique weighted adaptive method: dual-path computation: the laser data processing program calculates one position estimate; the image processing program calculates another position estimate.

[0050] Dynamic weight adjustment: When the environment has rich texture (e.g., many image details), increase the weight of the image program (e.g., 40%); when in a long straight track (single feature), decrease the weight of the image program (e.g., 10%), while enhancing the correction effect of UWB; under normal circumstances, the weight of the laser program remains high (60%-90%).

[0051] Step 3: Track Feature Recognition

[0052] The feature extraction process works as follows: identifying the track plane and the two track edges from the laser data; using images to help identify the track contour; and matching these features with the track model to provide geometric constraints for positioning.

[0053] Step 4: Comprehensive Optimization Calculation

[0054] The system receives the results from the first three steps: fused position estimation; orbital feature constraints; and UWB absolute position reference. Through comprehensive calculation, it outputs the final position with an accuracy of ±2 cm and updates the environment map.

[0055] Step 5: Map Self-Learning

[0056] Based on precise positioning: the defect identification program discovers new problems on the track (such as loose bolts); the system saves the precise location and characteristics of the problem point as a "special marker"; this "special marker" is added to the map; during the next inspection, the system can quickly and accurately locate the problem by recognizing this "special marker".

[0057] III. Implementation Results

[0058] Through the above solutions, this system achieves the following: maintaining centimeter-level positioning accuracy in complex factory environments; automatically adapting to different environmental conditions without manual adjustment; automatically improving the map with each use, becoming more accurate with each use; and having multiple sensors that serve as backups for each other, so the failure of a single sensor does not affect the operation of the system.

[0059] This embodiment demonstrates the complete workflow of the system in a practical application, proving the feasibility and advancement of the technical solution. Through a unique weight adjustment method and map self-learning mechanism, the system effectively solves the challenge of precise positioning in high-altitude orbital environments.

[0060] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-altitude rail AGV precise positioning and map construction method, characterized in that, The method comprises the following steps: S1: collecting environment data through a multi-sensor system mounted on the AGV, the multi-sensor system comprising a laser radar, a vision sensor, an inertial measurement unit (IMU) and an odometer, and an ultra-wideband (UWB) positioning receiving module; S2: constructing a hierarchical multi-sensor fusion model, taking the pose estimation results of the front-end laser SLAM and vision SLAM as the basis, and dynamically assigning the fusion weights of the two based on real-time environment features; wherein the basis for dynamically assigning the weights comprises: calculating the information entropy of the current vision image, and / or analyzing the geometric feature distribution of the current laser point cloud; S3: extracting structural semantic features of the high-altitude track from the sensor data in real time, the structural semantic features at least comprising track planes and track edge lines segmented from the laser point cloud; and matching these semantic features as strong constraints with a pre-defined track structure prior model; S4: inputting the front-end pose after weight fusion, the absolute coordinate constraint provided by the UWB positioning module, and the semantic feature constraint obtained in step S3 into the back-end for graph optimization to solve the optimal pose estimation of the AGV and update the global map.

2. The method of claim 1, wherein, The dynamic weight assignment in step S2 is specifically: when the information entropy of the vision image is higher than a first threshold, the fusion weight of the vision SLAM pose is increased; when the laser point cloud geometric feature distribution indicates that it is in a long straight corridor environment, the fusion weight of the vision SLAM is reduced and the weight of the UWB constraint in the back-end optimization is increased.

3. The method of claim 1, wherein, The method further comprises step S5: during the detection task performed by the AGV, using the optimal pose obtained in step S4 to identify new defects or structural changes on the track through a defect detection algorithm; defining the three-dimensional coordinates and feature descriptors of the defect or change features as new semantic landmarks; registering and updating the semantic landmarks in real time into the global map for subsequent task loop detection and pose correction.

4. A high-altitude rail AGV precise positioning and map construction system, characterized in that, The system is deployed on the AGV body and comprises: a data acquisition module comprising a laser radar, a vision sensor, an IMU and an odometer, and a UWB receiver; an adaptive fusion positioning module configured to receive data of the data acquisition module, run laser SLAM and vision SLAM algorithms, and dynamically assign fusion weights of laser SLAM and vision SLAM pose results based on the information entropy of the current vision image and / or the geometric feature distribution of the laser point cloud; a semantic processing module configured to extract structural semantic features of the high-altitude track from the data of the data acquisition module, at least comprising track planes and track edge lines segmented from the laser point cloud, and match them with a track structure prior model; a back-end graph optimization module configured to receive the fused pose output by the adaptive fusion positioning module, the semantic constraint matched by the semantic processing module, and the absolute coordinate constraint provided by the UWB, perform joint optimization, and output the accurate positioning of the AGV and update the global map.

5. The system of claim 4, wherein, The vision sensor is a binocular camera or an RGB-D camera.

6. The system of claim 4, wherein, The system further comprises: A dynamic map management module is configured to define the track defects or structural change features identified on the basis of accurate positioning as new semantic landmarks and register updates to the global map in real time.

Images