CN-121982232-A - Laser point cloud map precision evaluation system and method based on multi-mode data fusion

CN121982232ACN 121982232 ACN121982232 ACN 121982232ACN-121982232-A

Abstract

The invention discloses a laser point cloud map precision evaluation system and method based on multi-mode data fusion, wherein the system comprises an MME evaluation module, an APE optimization module, a weight balance module and a precision evaluation index, wherein the MME evaluation module is used for constructing matrix elements through coding and layering processing principles and is used for removing dynamic points of a point cloud map and preserving static points of the point cloud map, calculating the proportion of the dynamic points, calculating the normal vector distribution entropy value of a local area and quantitatively evaluating microscopic quality on the basis of the static points, introducing intensity reflection consistency to detect and identify and calculate the intensity abnormal proportion of a special material area, comprehensively calculating an MME comprehensive score, correcting absolute pose errors and calculating APE errors and APE comprehensive scores through fusion of multi-sensor data, and the weight balance module is used for automatically adjusting the weight proportion of the MME and the APE and calculating the precision evaluation index of the MME and the APE under the definition of an environment adaptability weighting function. Therefore, the map evaluation result with high precision can be still provided in a complex environment.

Inventors

GUO XINYANG
WU ENGUANG
WANG HONGHUI

Assignees

城市之光(深圳)无人驾驶有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (10)

1. Laser point cloud map precision evaluation system based on multimodal data fusion, which is characterized by comprising: The MME evaluation module is used for constructing matrix elements of the point cloud map through globally consistent binary coding and layering processing principles, removing dynamic points of the point cloud map and static points of the reserved point cloud map, calculating the proportion of the dynamic points on the basis of the total number of the point clouds, calculating the normal vector distribution entropy value of a local area on the basis of the static points of the reserved point cloud map, carrying out quantitative evaluation on the microscopic quality of the point cloud map, introducing intensity reflection consistency detection, identifying and calculating the intensity anomaly proportion of the special material area through the intensity difference between the point pairs, and summing the intensity anomaly proportion according to the proportion of the dynamic points, the normal vector distribution entropy value and the intensity anomaly proportion to obtain an MME comprehensive score; The APE optimization module is used for correcting absolute pose errors by fusing GNSS, IMU and laser radar data and adopting a multi-sensor tight coupling algorithm, and calculating APE errors and APE comprehensive scores according to GNSS confidence level; The weight balance module automatically adjusts the weight proportion of the MME and the APE according to the GNSS quality factor by defining an environment adaptability weighting function, and fuses the weight proportion and the comprehensive score evaluation result of the MME and the APE into a unified precision evaluation index.
2. The multi-modal data fusion-based laser point cloud map accuracy assessment system of claim 1, wherein the MME assessment module comprises: When a point is reflected in a third unit cell in the upward direction of the matrix element, the back projection ray is carried out on the position of each point of the dynamic area matrix, if the position is not shielded, the position is marked as 1, otherwise, the position is marked as 0; the distributed entropy calculation unit extracts a local normal vector set of each point neighborhood through PCA, calculates normal vector distribution density on a unit sphere, and normalizes the normal vector distribution density after converting the normal vector distribution density into a histogram; the consistency detection unit is used for detecting the intensity difference between the point pairs to obtain the average intensity of each point, marking the average intensity value as an abnormal reflection point if the average intensity value is larger than a set threshold value, and then calculating the abnormal proportion of the intensity in the corresponding area; and the MME calculation unit is used for carrying out summation calculation according to the dynamic point proportion, the normal vector distribution entropy value and the intensity anomaly proportion to obtain an MME comprehensive score.
3. The laser point cloud map accuracy evaluation system based on multi-modal data fusion according to claim 2, wherein the calculation formula of the MME calculation unit is as follows: ; dynamic point ratio (dynamic point/total point); normal vector entropy; the abnormal proportion of the intensity.
4. The laser point cloud map accuracy assessment system based on multi-modal data fusion of claim 1, wherein the APE optimization module comprises: the data fusion unit is used for fusing GNSS, IMU and laser radar data, and performing weighted average by utilizing a covariance matrix of multiple sensors to obtain final pose estimation; And the APE calculating unit adopts different error calculating modes according to the GNSS confidence level, directly calculates Euclidean distance when the confidence level is high, introduces a GNSS covariance matrix as a weighted sum to calculate the Mahalanobis distance when the confidence level is middle, adopts punishment item processing when the confidence level is low, and dynamically sets weight and calculates APE comprehensive scores according to the change of the high, middle and low confidence levels.
5. The system for evaluating the accuracy of the laser point cloud map based on the multi-modal data fusion according to claim 1, wherein the weight balancing module comprises a weight function calculating unit, the weight function calculating unit dynamically adjusts the weight ratio of the MME to the APE according to the GNSS quality factor, the value range of the GNSS quality factor is 0 to 1, the MME is dominant when the GNSS quality factor is biased to 0, and the APE is dominant when the GNSS quality factor is biased to 1.
6. The system of any one of claims 1-5, further comprising an anomaly handling module configured to monitor GNSS quality factors of environmental features in real time to determine GNSS anomalies, adjust weight ratios of MME and APE in the accuracy assessment index according to GNSS confidence level correspondence, and detect and correct dynamic object residuals.
7. The system of claim 6, wherein the anomaly handling module comprises a GNSS anomaly detection unit for monitoring GNSS quality factors to determine GNSS anomalies, and correspondingly adjusting weights of MME and APE in APE calculation scale and accuracy assessment index according to GNSS confidence level.
8. The system for evaluating the precision of the laser point cloud map based on multi-modal data fusion according to claim 6, wherein the anomaly processing module comprises a dynamic point cloud correction unit for projecting a current frame and a global map to a two-dimensional occupancy matrix based on binary matrix coding to identify a point cloud inconsistent area caused by dynamic obstacles, detecting and eliminating dynamic object residues through motion analysis and time sequence comparison, repairing point cloud holes by adopting an edge maintenance algorithm, constructing a point cloud time sequence to identify a high entropy area caused by the dynamic obstacles by adopting space-time consistency detection, and correcting interference of dynamic points to MME calculation.
9. The multi-modal data fusion-based laser point cloud map accuracy assessment system of any of claims 1-5, further comprising a visual verification module for projecting three-dimensional point cloud data of a point cloud map onto an image plane, calculating a reprojection error and assessing a degree of matching of the point cloud with the image by matching the projected image with the actual image pixel by pixel, and real-time three-dimensional visualization of presentation, querying of the point cloud state, and correction of assessment parameters.
10. A laser point cloud map precision evaluation method based on multi-mode data fusion is characterized by comprising the following steps: S1, evaluating and calculating an MME, namely constructing matrix elements of a point cloud map through coding and layering processing principles of an MME evaluation module, removing dynamic points of the point cloud map and static points of a reserved point cloud map, calculating normal vector distribution entropy values of local areas on the basis of the static points of the reserved point cloud map to quantitatively evaluate microscopic quality of the point cloud map, identifying and calculating intensity anomaly proportions of special material areas through intensity difference among point pairs by introducing intensity reflection consistency detection; S2, optimizing and calculating APE, namely fusing GNSS, IMU and laser radar data through an APE optimizing module, correcting absolute pose errors by adopting a multi-sensor tight coupling algorithm, and calculating APE errors and APE comprehensive scores according to GNSS confidence level; S3, adjusting the weight proportion of the MME and the APE, namely defining an environment adaptability weighting function through a weight balance module, automatically adjusting the weight proportion of the MME and the APE according to GNSS quality factors, and fusing the weight proportion of each of the MME and the APE and a comprehensive score evaluation result to form a unified precision evaluation index; S4, GNSS anomaly processing and point cloud correction, namely monitoring GNSS quality factors through an anomaly processing module to judge GNSS anomaly conditions, correspondingly adjusting weight proportions of the MME and the APE in the precision evaluation index in the step 3 according to GNSS confidence level, and detecting and correcting dynamic object residues; and S5, visually verifying and displaying, namely projecting three-dimensional point cloud data of the point cloud map to an image plane through a visual verification module, calculating a reprojection error and evaluating the matching degree of the point cloud and the image by matching the projected image with the actual image pixel by pixel, and visually displaying real-time interaction, inquiring the state of the point cloud and correcting the evaluation parameters.

Description

Laser point cloud map precision evaluation system and method based on multi-mode data fusion Technical Field The invention relates to the technical field of automatic driving positioning, in particular to a laser point cloud map precision evaluation system and method based on multi-mode data fusion. Background With the rapid development of automatic driving technology, synchronous positioning and map building (SLAM) technology is increasingly used in the fields of automatic driving, mobile robots and the like. The SLAM technology based on the laser radar can provide centimeter-level positioning and high-precision point cloud maps for automatic driving automobiles without Global Navigation Satellite Systems (GNSS) or unstable signals. The conventional point cloud map precision evaluation mainly depends on Absolute Pose Errors (APEs) of the GNSS as core indexes, wherein the Absolute Pose Errors (APEs) are core indexes for evaluating the precision of the SLAM/VO algorithm, and are used for measuring global consistency errors between an estimated track and a real track, so that the overall precision of the point cloud map is indirectly reflected by comparing the deviation of a map building track and a GNSS reference track. However, this prior art still has the following problems: 1. The traditional point cloud map precision evaluation method has the defects of single evaluation dimension and missing micro precision, so that the precision of the point cloud map precision evaluation method can reach centimeter level in an open area, but in complex environments such as urban canyons or tunnels with serious multipath effects and signal blocking, errors can be suddenly increased to meter level, the requirement of fine quality control in complex scenes is difficult to meet, the local defects of the point cloud are covered, and the evaluation result is unhooked from the real quality of the point cloud. 2. The traditional evaluation system has the defects of poor anti-interference performance and the like, and particularly when the automatic driving cleaning robot uses a high-precision map to conduct path planning in specific application scenes, the traditional evaluation technology cannot effectively identify and correct the problems of dynamic obstacles, abnormal environmental characteristics and the like in the point cloud. 3. In addition, the traditional evaluation system lacks visual verification function, can not provide real-time and visual point cloud data and evaluation result display for field operators, and seriously influences the debugging efficiency and the accuracy of problem diagnosis. Disclosure of Invention In order to overcome the defects of the prior art, one of the purposes of the invention is to provide a laser point cloud map accuracy assessment system based on multi-mode data fusion. The invention discloses a laser point cloud map precision evaluation system based on multi-mode data fusion, which comprises the following technical scheme: The MME evaluation module is used for constructing matrix elements of the point cloud map through globally consistent binary coding and layering processing principles, removing dynamic points of the point cloud map and static points of the reserved point cloud map, calculating the proportion of the dynamic points on the basis of the total number of the point clouds, calculating the normal vector distribution entropy value of a local area on the basis of the static points of the reserved point cloud map, carrying out quantitative evaluation on the microscopic quality of the point cloud map, introducing intensity reflection consistency detection, identifying and calculating the intensity anomaly proportion of the special material area through the intensity difference between the point pairs, and summing the intensity anomaly proportion according to the proportion of the dynamic points, the normal vector distribution entropy value and the intensity anomaly proportion to obtain an MME comprehensive score; The APE optimization module is used for correcting absolute pose errors by fusing GNSS, IMU and laser radar data and adopting a multi-sensor tight coupling algorithm, and calculating APE errors and APE comprehensive scores according to GNSS confidence level; The weight balance module automatically adjusts the weight proportion of the MME and the APE according to the GNSS quality factor by defining an environment adaptability weighting function, and fuses the weight proportion and the comprehensive score evaluation result of the MME and the APE into a unified precision evaluation index. Further, the MME evaluation module includes: When a point is reflected in a third unit cell in the upward direction of the matrix element, the back projection ray is carried out on the position of each point of the dynamic area matrix, if the position is not shielded, the position is marked as 1, otherwise, the position is marked as 0; the distributed entropy calculation unit extract