CN-121994258-A - Urban level crowd source collaborative mapping and real-time positioning method and system

CN121994258ACN 121994258 ACN121994258 ACN 121994258ACN-121994258-A

Abstract

The invention relates to the technical field of positioning and mapping related to intelligent traffic and automatic driving/road cooperation, in particular to a method and a system for urban mass-source cooperation mapping and real-time positioning, which are implemented through standardized multi-source observation data, space-time reference construction and cross-source registration, the stable fusion of urban mass source data is realized, and the stability and the expandability of heterogeneous sensors and data quality fluctuation are improved through dynamic regulation and control parameter optimization loop constraint and graph optimization. By adopting a multidimensional credibility characterization and dynamic gating mechanism, loop starting conditions are effectively controlled, the problem that a global map is polluted by wrong loops caused by the drift of a semantic anchor point is avoided, the risk that structured errors are solidified into global consistency is reduced, and the safety of the map is improved. Through layered map and dynamic regulation, a closed-loop updating mechanism can be formed according to positioning residual errors and regional risk statistics optimization parameters during real-time positioning.

Inventors

HU WEN
FANG XIAOLE
QI ZICHAO
YANG ZHENFEI
PAN JUNLIN
LV ZHENG
Wan Chuwen
ZHU SHAN

Assignees

广州宸境科技有限公司
韶关市数据产业研究院
韶关市空域科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260323

Claims (10)

1. The city level crowd source collaborative mapping and real-time positioning method is characterized by comprising the following steps: Step 1, acquiring crowd source heterogeneous sensing data from a plurality of crowd source acquisition nodes, and preprocessing the crowd source heterogeneous sensing data to generate standardized multi-source observation data; Step 2, constructing a city level unified space-time reference based on time information and a space initial value of the standardized multi-source observation data, and executing cross-source time sequence alignment and space registration on the standardized multi-source observation data to obtain a candidate pose sequence and candidate observation association relation; step 3, calculating a multidimensional credibility characterization quantity according to the association relation between the candidate pose sequence and the candidate observation; generating loop constraint candidates based on the association relation between the candidate pose sequences and the candidate observation, and generating dynamic regulation parameters for fusing the building map and the loop constraint according to the multi-dimensional credibility characterization quantity; step 5, under the constraint of dynamic regulation parameters, constructing a layered city map and executing map optimization; And 6, performing multi-source matching and state estimation based on the layered urban map and the standardized multi-source observation data to realize real-time positioning, and updating dynamic regulation parameters and incremental updating of the layered urban map based on positioning residual errors, interlayer conflict information and regional risk statistics.
2. The method for collaborative mapping and real-time positioning according to claim 1, wherein the crowd source acquisition node at least comprises one or more of a vehicle-mounted terminal, a mobile terminal and a road side acquisition unit, and the crowd source heterogeneous sensing data at least comprises laser radar point cloud data, satellite navigation positioning track data and image data.
3. The method of urban mass-source collaborative mapping and real-time localization according to claim 1, wherein the preprocessing in step 1 comprises performing timestamp unification, coordinate system conversion, sensor calibration parameter unification and quality screening on mass-source heterogeneous sensing data to generate standardized multi-source observation data.
4. The method of urban level crowd-sourced collaborative mapping and real-time localization according to claim 1, wherein the multidimensional credibility characterizer in step 3 comprises at least a perceptual degradation strength, a semantic anchor mobility index, a loop information gain, an interlayer collision strength, and a crowd-sourced relevance penalty factor.
5. The method of urban level crowd-sourced collaborative mapping and real-time positioning according to claim 1, wherein the dynamic regulation parameters generated in step 4 at least comprise a gating condition and a loop constraint weight upper limit for enabling loop constraint, an anchor weight upper limit for participating in loop and positioning by a semantic anchor, a freezing condition for updating lane topology, a directivity gating rule of interlayer information flow, and a saturation or decremental rule of a crowd-sourced heterogeneous sensing data fusion weight.
6. The method for collaborative mapping and real-time positioning of urban level crowd source according to claim 1 is characterized in that a two-stage submitting mechanism is adopted for loop constraint candidates generated in the step 4, and the method comprises the steps of adding the loop constraint candidates into a local subgraph in a to-be-submitted state and executing consistency verification in the first stage, submitting the loop constraint candidates to a global map for updating a hierarchical urban map only when the loop constraint candidates meet preset cross-time window reproduction conditions and cross-modal consistency conditions and meet constraint conditions on lane layer topology influences, and writing variable elements which do not meet the submitted conditions into a variable information layer and setting validity periods, wherein the hierarchical urban map at least comprises a geometric layer, a topology layer, a lane layer, a semantic layer and the variable information layer.
7. The method for collaborative mapping and real-time positioning of urban mass sources according to claim 1, wherein the step of generating dynamic regulation parameters for merging mapping and loop-back constraints according to the magnitude of a multi-dimensional credibility characterization quantity comprises the steps of generating the dynamic regulation parameters for merging mapping and loop-back constraints according to the magnitude of the multi-dimensional credibility characterization quantity, normalizing perceived degradation intensity, semantic anchor point mobility index, loop-back information gain, interlayer conflict intensity and mass source relevance penalty factors, and performing robust statistical smoothing in a sliding time window to obtain credibility vectors for parameter generation.
8. The method of urban level crowd source collaborative mapping and real-time positioning according to claim 7, wherein generating an enabling gating condition and an upper limit of a loop constraint weight of the loop constraint based on the reliability vector comprises closing the loop constraint when the loop information gain is lower than a first threshold or the inter-layer conflict strength is higher than a second threshold, and setting the upper limit of the loop constraint weight as a function that monotonically increases with the loop information gain and monotonically decreases with the perceived degradation strength and the crowd source relevance penalty factor when the enabling gating condition is satisfied.
9. The method for collaborative mapping and real-time positioning according to claim 7, wherein generating a semantic anchor point weight upper limit, a lane topology update freezing condition and an interlayer information flow directivity gating rule based on a reliability vector includes that the semantic anchor point weight upper limit monotonically decreases with a semantic anchor point mobility index, and corresponding semantic anchor points are reduced or eliminated when the semantic anchor point mobility index is higher than a third threshold, lane topology update is frozen and change elements are written into a variable information layer when interlayer conflict intensity is higher than a fourth threshold or a loop constraint candidate exceeds a preset limit on lane layer topology, and reverse correction of a geometric layer, a topology layer and a lane layer by the semantic layer is limited when the interlayer conflict intensity is increased, and unidirectional constraint of the geometric layer, the topology layer and the lane layer on the semantic layer is preferentially adopted.
10. The system for collaborative mapping and real-time positioning of urban mass sources is applied to the method for collaborative mapping and real-time positioning of urban mass sources according to any one of claims 1-9, and is characterized by comprising the following steps: The crowd source acquisition module is used for acquiring crowd source heterogeneous sensing data from a plurality of crowd source acquisition nodes; the data preprocessing module is used for preprocessing the crowd-sourced heterogeneous sensing data to generate standardized multi-source observation data; The space-time alignment module is used for constructing a city-level unified space-time reference based on the time information and the space initial value of the standardized multi-source observation data, and executing cross-source time sequence alignment and space registration on the standardized multi-source observation data so as to obtain a candidate pose sequence and candidate observation association relation; the credibility calculation module is used for calculating a multidimensional credibility characterization quantity aiming at the association relation between the candidate pose sequence and the candidate observation; The loop regulation and control module is used for generating loop constraint candidates based on the association relation between the candidate pose sequences and the candidate observation, and generating dynamic regulation and control parameters for fusing the building map and the loop constraint according to the size of the multidimensional credibility characterization quantity; And the map construction positioning module is used for constructing a layered urban map and executing map optimization under the constraint of the dynamic regulation parameters, executing multi-source matching and state estimation based on the layered urban map and standardized multi-source observation data to realize real-time positioning, and updating the dynamic regulation parameters and incremental updating of the layered urban map based on positioning residual errors, interlayer conflict information and regional risk statistics.

Description

Urban level crowd source collaborative mapping and real-time positioning method and system Technical Field The invention relates to the technical field of positioning and mapping related to intelligent traffic and automatic driving/road cooperation, in particular to a method and a system for city level crowd source cooperation mapping and real-time positioning. Background In urban level high-precision map construction and real-time positioning, a multi-sensor fusion (such as GNSS/IMU, vision/laser/millimeter wave and the like) and map optimization framework is commonly adopted in the industry, global accuracy of a map is improved through cross-period data fusion, loop detection and factor map consistency constraint, repeated coverage of multiple acquisition nodes is also commonly used for enhancing coverage rate and robustness in a crowd source cooperative scene, and semantic elements (crossing devices, signboards, rods, building facades and the like) are gradually introduced as long-term stable constraint sources so as to make up positioning capability when geometric characteristics are insufficient. The route can achieve better mapping and positioning effects in most conventional weather and structurally stable areas. However, under extremely degraded environments such as snow storm, water accumulation mirror reflection, fallen leaf coverage, and strong haze scattering, key elements of geometric/lane layers such as lane lines, curbs and the like are often weak or invisible, the system generally improves the influence of semantic and appearance anchor points on matching, loop and state estimation in engineering so as to maintain usability, and meanwhile, short-period appearance or position change (such as seasonal advertisement replacement, flash shop recruitment, temporary facility establishment, temporary sign migration and the like) exists in part of semantic landmarks in urban environments, so that the stability and uniqueness of the semantic anchor points are reduced in certain time periods. The method can generate a chained risk that a degradation environment promotes the lane layer to be complemented by a structured but wrong drivable boundary caused by snow edges, reflection, textures and the like, and the drift and confusion of a semantic anchor point can improve the probability of similar non-co-sited matching so as to trigger wrong loop constraint, and the optimization of the graph can lead local structuring errors to be diffused into globally consistent but physically wrong topological rewrites (such as road folding, self-intersecting and lane connection being rewritten) through loop priori in the process of pursuing the whole residual descent and closed loop consistency, so that a hidden risk state of mathematical self-consistency and physical distortion is formed. The problems are not negative to the effectiveness of the existing method, but when urban mass sources, extreme degradation and semantic short-term drift are overlapped, higher requirements are put on loop starting time, constraint weight upper limit and cross-layer consistency check, namely interpretable and controllable dynamic balance needs to be realized between loop income and conflict risk, and unreliable priori solidification into a global map in a degradation scene is avoided. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a method and a system for collaborative mapping and real-time positioning of urban mass sources. In order to achieve the above purpose, the present invention provides the following technical solutions: a method for collaborative mapping and real-time positioning of urban mass sources comprises the following steps: Step 1, acquiring crowd source heterogeneous sensing data from a plurality of crowd source acquisition nodes, and preprocessing the crowd source heterogeneous sensing data to generate standardized multi-source observation data; Step 2, constructing a city level unified space-time reference based on time information and a space initial value of the standardized multi-source observation data, and executing cross-source time sequence alignment and space registration on the standardized multi-source observation data to obtain a candidate pose sequence and candidate observation association relation; step 3, calculating a multidimensional credibility characterization quantity according to the association relation between the candidate pose sequence and the candidate observation; generating loop constraint candidates based on the association relation between the candidate pose sequences and the candidate observation, and generating dynamic regulation parameters for fusing the building map and the loop constraint according to the multi-dimensional credibility characterization quantity; step 5, under the constraint of dynamic regulation parameters, constructing a layered city map and executing map optimization; And 6, performing multi-source matching