High-dynamic scene long-term positioning method based on multi-layer map regional scoring mechanism

Abstract

The invention provides a high-dynamic scene long-term positioning method based on a multi-layer map regional scoring mechanism, and the method comprises the steps: A, loading 2D global map data, and obtaining an initial pose; step B, performing sub-graph segmentation on the 2D global map data to obtain 2D sub-graph units; step C, acquiring real-time scanning point cloud data, applying confidence to the 2D subgraph units based on the real-time scanning point cloud data, executing a continuous gain strategy, summarizing global confidence, executing dynamic and static matching positioning through the global confidence, and issuing a correction pose; and step D, continuously correcting the initial pose based on the step B to the step C. The invention provides the positioning method which is high in dynamic scene adaptability, low in computing power consumption and high in positioning reliability.

Description

Technical Field

[0001] This invention relates to the field of 2D mobile robot localization technology, and more specifically, to a long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism. Background Technology

[0002] In long-term operation, traditional positioning methods for 2D mobile robots often face two core problems.

[0003] On the one hand, static maps are difficult to adapt to dynamic changes in the environment, such as moving shelves in warehouses or pedestrians moving around in supermarkets, which leads to a decrease in the accuracy of real-time scanning and map matching, and the positioning is prone to drift.

[0004] On the other hand, continuous operation of robots will accumulate a large amount of repetitive or outdated map data, which will consume computing resources and further reduce positioning efficiency.

[0005] Existing solutions often fail to balance dynamic environment adaptation with low computing power requirements, or lack accurate judgment on the reliability of matching results, making it difficult to meet the long-term, stable, and high-precision positioning requirements in 2D scenarios.

[0006] In summary, there is an urgent need for a positioning method that is highly adaptable to dynamic scenarios, consumes little computing power, and has high positioning reliability. Summary of the Invention

[0007] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism, the method comprising:

[0008] A: Load 2D global map data and obtain the initial pose;

[0009] B: Perform sub-map segmentation on the 2D global map data to obtain 2D sub-map cells;

[0010] C: Acquire real-time scanned point cloud data, apply confidence to 2D sub-graph units based on the real-time scanned point cloud data, execute a continuous gain strategy, summarize the global confidence, perform dynamic and static matching localization through the global confidence, and publish the corrected pose.

[0011] D: Continuously correct the initial pose based on steps B to C.

[0012] As a further aspect of the present invention, loading 2D global map data includes:

[0013] The 2D global map data adopts a dual-map architecture, including a static base map and a dynamic local map;

[0014] The static reference map stores fixed features in the 2D scene.

[0015] The dynamic local graph updates the temporary dynamic features of the current region in real time.

[0016] As a further aspect of the present invention, sub-graph segmentation is performed on the 2D global map data to obtain 2D sub-graph units, including:

[0017] The 2D global map data is split to obtain multiple 2D sub-map units, wherein the 2D sub-map units are represented as an N*N grid and include dynamic sub-maps and static sub-maps.

[0018] As a further aspect of the present invention, real-time scanned point cloud data is acquired, confidence levels are applied to 2D sub-graph units based on the real-time scanned point cloud data, a continuous gain strategy is executed, global confidence levels are aggregated, dynamic-static matching localization is performed using the global confidence levels, and a corrected pose is published, including:

[0019] Perform region partitioning operations on 2D subgraph cells to obtain local regions;

[0020] A matching operation is performed based on local regions and real-time scanned point cloud data to apply regional confidence to the corresponding local regions.

[0021] If the confidence score of any consecutive local region exceeds the preset confidence threshold, then the confidence score of that consecutive local region is multiplied by a consecutive high-score gain coefficient of 1-1.2.

[0022] Summarize the global confidence scores and retain only the matching results with local confidence scores greater than or equal to the confidence threshold and without consecutive low confidence regions as the matching evaluation criteria;

[0023] The judgment and positioning operation is performed based on the matching evaluation criteria, and the static reference map or dynamic local output is selected to correct the pose based on the judgment and positioning operation.

[0024] As a further aspect of the present invention, a judgment operation is performed based on a matching evaluation criterion, and a corrected pose is output based on the judgment operation, including:

[0025] If the matching evaluation criteria are greater than the localization threshold, a new dynamic sub-graph is created, saved, and the sub-graph is cropped. The corrected pose is then output based on the static reference graph.

[0026] If the matching evaluation criteria are less than the localization threshold, read the dynamic local map and rematch it, and output the corrected pose based on the localization of the dynamic sub-map.

[0027] As a further aspect of the present invention, the method further includes:

[0028] If any 2D sub-graph cell has not been matched with real-time scanned point cloud data for an extended period of time, it is marked as an outdated sub-graph and is then cropped.

[0029] If any 2D subgraph cell has a repetition rate exceeding a preset repetition rate threshold with other 2D subgraph cells, it will be cropped.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. Strong adaptability to dynamic scenes: The dynamic and static map localization architecture enables 2D robots to maintain a stable localization error of ±6mm-±10mm in dynamic scenes such as warehouses and supermarkets (tested on datasets such as LiLocBench), which is 50% lower than traditional solutions.

[0032] 2. Low computing power consumption: Real-time sub-map pruning reduces the amount of 2D map data by more than 30%, and the update time for a single sub-map (area 500㎡) is ≤0.4 seconds.

[0033] 3. High positioning reliability: Regional confidence calculation reduces invalid matching rate by 60%, and there is no positioning drift during long-term operation (e.g., 8 hours), meeting the long-term positioning requirements in 2D dynamic scenarios. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the steps of the long-term positioning method for high-dynamic scenes based on a multi-layer map regional scoring mechanism, as described in this invention. Detailed Implementation

[0035] The present invention will now be described in detail with reference to the accompanying drawings. Please refer to the drawings. Figure 1 As shown below, the long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism will be described in detail.

[0036] This embodiment provides a long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism, which relates to the field of 2D mobile robot localization technology. It is particularly suitable for indoor and outdoor 2D dynamic scenes, such as warehouse workshops, supermarket aisles, and park roads. It solves the problem of map information redundancy and positioning offset caused by dynamic environmental changes when robots work for a long time in 2D scenes. It is especially suitable for small and medium-sized 2D mobile robots with limited computing power, such as 2D warehouse AGVs and indoor inspection robots.

[0037] Furthermore, this method specifically includes steps A, B, C, and D.

[0038] Step A: Load 2D global map data and obtain the initial pose.

[0039] Specifically, the 2D global map data adopts a dual-map architecture, including a static reference map and a dynamic local map. The static reference map stores fixed features in the 2D scene, such as the 2D outlines of walls and permanent shelves. The dynamic local map updates the temporary dynamic features of the current area in real time, such as temporarily stacked goods and moving equipment. In subsequent positioning operations, the static reference map ensures that the global position does not shift, while the dynamic local map adapts to real-time changes in the environment. The two work together to correct the pose flow, avoiding the problem that a single map cannot cope with dynamic scenes.

[0040] Understandably, adopting a dual-map architecture to adapt to 2D dynamic scenes differs from using a single static map. By coordinating a static baseline map and a dynamic local map, it retains the global constraints of fixed features of the 2D scene while updating dynamic features in real time, thus solving the positioning drift problem in 2D dynamic environments.

[0041] Step B involves subdividing the 2D global map data into sub-map cells to obtain 2D sub-map units.

[0042] Specifically, the 2D global map data is split to obtain multiple 2D sub-map units, where each 2D sub-map unit is represented as an N*N grid and includes dynamic and static sub-maps. The validity of the sub-maps is judged in real time. If any 2D sub-map unit has not been matched with real-time scanned point cloud data for a long time, it is marked as an outdated sub-map and is cropped. If the repetition rate of any 2D sub-map unit with other 2D sub-map units exceeds a preset repetition rate threshold, it is also cropped.

[0043] Furthermore, the 2D global map is divided into multiple 2D sub-map units, such as a 2m*2m grid, and the validity of the sub-maps is judged in real time. If a sub-map unit has not been matched with newly scanned data for a long time, it is marked as an outdated sub-map and is immediately cropped. If a sub-map unit has not been matched with newly scanned real-time point cloud data for a long time, it is marked as an outdated sub-map and is immediately cropped. Or, if the overlap rate with other sub-maps exceeds, for example, 50%, 60%, or 70%, it indicates information redundancy, and the sub-map is immediately cropped. By cropping redundant and outdated sub-maps, the amount of map data is reduced, the computing power consumption is reduced, and the real-time positioning is guaranteed.

[0044] Step C: Acquire real-time scanned point cloud data, apply confidence to 2D sub-graph units based on the real-time scanned point cloud data, execute a continuous gain strategy, summarize the global confidence, perform dynamic-static matching localization based on the global confidence, and publish the corrected pose.

[0045] Understandably, the traditional matching logic of globally unified scoring is abandoned, and regional scoring statistics are performed in highly dynamic scenarios.

[0046] Furthermore, step C includes steps C-1, C-2, C-3, C-4, and C-5.

[0047] Step C-1: Perform a region division operation on the 2D subgraph cells to obtain local regions.

[0048] Specifically, when performing region partitioning operations on 2D subgraph units, manual pre-annotation or clustering algorithms can be used to partition the regions.

[0049] Step C-2: Perform a matching operation based on the local region and the real-time scanned point cloud data, and apply a region confidence score to the corresponding local region.

[0050] Specifically, for each local area, the laser scanning points (i.e., real-time scanned point cloud data) are first counted separately, and their matching with map features is used to obtain the confidence score for that area.

[0051] Step C-3: If the confidence level of any continuous local region exceeds the preset confidence threshold, then multiply the confidence level of that continuous local region by a continuous high-score gain coefficient of 1-1.2.

[0052] Specifically, if any continuous local region, for example, has a confidence level of more than 80% for three or more adjacent grids, then the confidence level of that region is multiplied by a continuous high-score gain coefficient of 1-1.2 to enhance the contribution of the stable matching region.

[0053] Understandably, calculating confidence based on 2D local regions and introducing a gain mechanism can accurately distinguish reliable matching regions from interference regions, avoid the low confidence problem caused by global uniform scoring, and improve the reliability of matching results.

[0054] Step C-4: Summarize the global confidence scores and retain only the matching results with local confidence scores greater than or equal to the confidence threshold and without consecutive low confidence regions as the matching evaluation criteria.

[0055] For example, matching results with a global confidence level of 60% or higher and no consecutive low confidence regions (30%) can be retained as the point cloud matching evaluation criteria.

[0056] Step C-5: Perform a judgment and positioning operation based on the matching evaluation criteria, and select either a static reference map or a dynamic local output to correct the pose based on the judgment and positioning operation.

[0057] Furthermore, step C-5 includes steps C-5-1 and C-5-2.

[0058] Step C-5-1: If the matching evaluation criterion is greater than the positioning threshold, create a new dynamic sub-map, save it, and perform sub-map cropping. Then, output the corrected pose based on the static reference map.

[0059] Step C-5-2: If the matching evaluation criterion is less than the localization threshold, read the dynamic local map and rematch it, and output the corrected pose based on the localization of the dynamic sub-map.

[0060] Finally, the pose can be corrected based on GICP (Generalized Iterative Closest Point) and used as the initial pose.

[0061] Understandably, in 2D localization, the core is to optimize rigid body transformation (rotation + translation) by combining the local geometric uncertainty of the point cloud to achieve localization. First, for each point in the source point cloud (current frame) and the target point cloud (map / historical frame), the covariance matrix is ​​calculated using k-nearest neighbors to represent the geometric uncertainty. Based on the initial pose, the source point cloud is transformed to match the corresponding point in the target point cloud. A Mahalanobis distance cost function (fusing covariance weighted error) is constructed, and the transformation parameters are iteratively optimized using the Gauss-Newton method until convergence. The final output transformation parameters are the 2D pose, completing the localization. Compared with traditional ICP, it is more suitable for noisy and uneven density scenarios.

[0062] Step D involves continuously refining the initial pose based on steps B through C.

[0063] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via limited means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0064] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0065] It should be understood that, in the embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0066] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism, characterized in that: It includes the following steps: A: Load 2D global map data and obtain the initial pose; B: Perform sub-map segmentation on the 2D global map data to obtain 2D sub-map cells; C: Acquire real-time scanned point cloud data, apply confidence to 2D sub-graph units based on the real-time scanned point cloud data, execute a continuous gain strategy, summarize the global confidence, perform dynamic and static matching localization through the global confidence, and publish the corrected pose. D: Continuously correct the initial pose based on steps B to C.

2. The long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism according to claim 1, characterized in that, Load 2D global map data, including: The 2D global map data adopts a dual-map architecture, including a static base map and a dynamic local map; The static reference map stores fixed features in the 2D scene. The dynamic local graph updates the temporary dynamic features of the current region in real time.

3. The long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism according to claim 1, characterized in that, The 2D global map data is subdivided to obtain 2D sub-map cells, including: The 2D global map data is split to obtain multiple 2D sub-map units, wherein the 2D sub-map units are represented as an N*N grid and include dynamic sub-maps and static sub-maps.

4. The long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism according to claim 1, characterized in that, Acquire real-time scanned point cloud data, apply confidence levels to 2D sub-graph cells based on the real-time scanned point cloud data, execute a continuous gain strategy, summarize the global confidence levels, perform dynamic-static matching localization based on the global confidence levels, and publish corrected poses, including: Perform region partitioning operations on 2D subgraph cells to obtain local regions; A matching operation is performed based on local regions and real-time scanned point cloud data to apply regional confidence to the corresponding local regions. If the confidence score of any consecutive local region exceeds the preset confidence threshold, then the confidence score of that consecutive local region is multiplied by a consecutive high-score gain coefficient of 1-1.

2. Summarize the global confidence scores and retain only the matching results with local confidence scores greater than or equal to the confidence threshold and without consecutive low confidence regions as the matching evaluation criteria; The judgment and positioning operation is performed based on the matching evaluation criteria, and the static reference map or dynamic local output is selected to correct the pose based on the judgment and positioning operation.

5. The long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism according to claim 4, characterized in that, Based on the matching evaluation criteria, a judgment operation is performed, and based on the judgment operation, a corrected pose is output, including: If the matching evaluation criteria are greater than the localization threshold, a new dynamic sub-graph is created, saved, and the sub-graph is cropped. The corrected pose is then output based on the static reference graph. If the matching evaluation criteria are less than the localization threshold, read the dynamic local map and rematch it, and output the corrected pose based on the localization of the dynamic sub-map.

6. The long-term localization method for high-dynamic scenes based on a multi-layer map regional scoring mechanism according to claim 1, characterized in that, The method further includes: If any 2D sub-graph cell has not been matched with real-time scanned point cloud data for an extended period of time, it is marked as an outdated sub-graph and is then cropped. If any 2D subgraph cell has a repetition rate exceeding a preset repetition rate threshold with other 2D subgraph cells, it will be cropped.

Images