Method and device for optimizing point cloud mapping quality, electronic equipment and medium

Abstract

The invention discloses a method and device for optimizing point cloud mapping quality, electronic equipment and a medium, and belongs to the technical field of image data processing. The method for optimizing the point cloud mapping quality comprises the steps of obtaining optimization factor information used for point cloud mapping; based on the optimization factor information, optimizing the point cloud trajectory; calculating a residual error of an optimization factor based on the optimized point cloud trajectory and the optimization factor information; performing inverse matrix weighting of a covariance matrix on the residual error of the optimization factor, and judging whether the weighted residual error is greater than a first threshold value or not; and if yes, processing the optimization factor information, and optimizing the point cloud trajectory based on the processed optimization factor information. According to the method and the device for optimizing the point cloud mapping quality, the electronic equipment and the medium provided by the invention, the multi-source sensor factors are introduced into a unified residual quantitative analysis framework, and abnormal factor information can be effectively detected in a graph optimization process.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular relates to a method, device, electronic equipment and medium for optimizing point cloud mapping quality. Background Art

[0002] Point cloud mapping is a core topic in the automatic localization and mapping (SLAM) technology. It generates three-dimensional point clouds through environmental data obtained by sensors to build environmental maps required by robots or unmanned driving systems. With the continuous development of SLAM technology, point cloud mapping has made significant progress in various application scenarios, especially in the fields of autonomous driving, drones, robot navigation, etc. Point cloud mapping technology has become an important means to support autonomous navigation.

[0003] At present, the process of point cloud mapping usually includes the optimization of local maps, and combines the optimization factors from different sensors (such as odometer, RTK satellite positioning, IMU, loop detection, etc.) for global optimization. In the framework of graph optimization, these optimization factors can be used to integrate the local map generated by odometer data into the global coordinate system, thereby obtaining an accurate global map that can be used for navigation. Through graph optimization technology, sensor errors can be eliminated and the accumulated errors in the map can be reduced, thereby improving the quality of mapping.

[0004] However, existing technologies still face some challenges in optimizing the quality of point cloud mapping. Although local maps can be effectively fused with global maps through graph optimization, conflicts between different optimization factors often occur during the optimization process in engineering applications. For example, there are varying degrees of errors between optimization factors such as odometer, IMU, and RTK, and they may not fully meet all optimization conditions, resulting in distortions in the optimization results. In this case, traditional layered detection methods often fail to effectively detect anomalies in point clouds because layered detection usually relies on the assumption that all optimization factors have been correctly integrated. However, conflicts between optimization factors can cause significant deviations between the optimization results and the actual scene, making layered detection unable to accurately reflect the quality of point cloud mapping.

[0005] At present, the evaluation of point cloud mapping quality in most engineering fields relies on manual inspection. When the optimization results are abnormal, it is often necessary to manually check the point cloud data to identify whether there are distortions or other abnormalities, and manually mark the abnormal points. This process is not only time-consuming, but also costly, which seriously restricts the demand for batch point cloud mapping in engineering. Especially when large-scale maps need to be built quickly and efficiently, manual intervention is obviously not efficient enough.

[0006] Therefore, it is necessary to provide a new solution to the above technical problems. Summary of the invention

[0007] The object of the present invention is to provide a method, device, electronic device and medium for optimizing the quality of point cloud mapping, which can monitor the abnormality of optimization factors in the process of point cloud mapping and improve the mapping quality.

[0008] To achieve the above purpose, the technical solution provided by the present invention is as follows: In a first aspect, the present invention provides a method for optimizing point cloud mapping quality, comprising: Obtain optimization factor information for point cloud mapping; optimize the point cloud trajectory based on the optimization factor information; calculate the residual of the optimization factor based on the optimized point cloud trajectory and the optimization factor information; weight the residual of the optimization factor by the inverse matrix of the covariance matrix to determine whether the weighted residual is greater than a first threshold; if so, process the optimization factor information, and optimize the point cloud trajectory based on the processed optimization factor information.

[0009] In one or more embodiments, the optimization factors include an odometer factor, an RTK satellite positioning factor, an IMU factor, and a loopback factor.

[0010] In one or more embodiments, before optimizing the point cloud trajectory based on the optimization factor information, the method also includes: calculating the residual of the loop factor based on the point cloud trajectory and loop factor information before optimization; weighting the residual of the loop factor with the inverse matrix of the covariance matrix to determine whether the weighted residual is greater than a second threshold; if so, deleting the corresponding loop factor.

[0011] In one or more embodiments, the residual calculation formula of the optimization factor is: ,in, h ( x ) is the theoretical calculated value of the optimization factor measurement equation, Z is the true measured value of the optimization factor.

[0012] In one or more embodiments, the formula for weighting the residual of the optimization factor by the inverse matrix of the covariance matrix is ​​expressed as: ,in, Var is the covariance matrix of the optimization factor information.

[0013] In one or more embodiments, when the optimization factor is a relative constraint factor, ,in, X 1 represents the pose variable of the first trajectory point, X 2 represents the pose variable of the second trajectory point; when the optimization factor is a unilateral constraint factor, ,in, X i Represents the sensor input at a trajectory point.

[0014] In one or more embodiments, the optimization factor information is processed, including: when the optimization factor information is loop factor information, deleting the optimization factor information; when the optimization factor information is odometer factor information, RTK satellite positioning factor information or IMU factor information, increasing the noise weight of the optimization factor information.

[0015] In a second aspect, the present invention provides a device for optimizing the quality of point cloud mapping, comprising: a sensor, an optimization module, a calculation module, a weighting module and a processing module; the sensor is used to obtain optimization factor information for point cloud mapping; the optimization module is used to optimize the point cloud trajectory based on the optimization factor information; the calculation module is used to calculate the residual of the optimization factor based on the optimized point cloud trajectory and the optimization factor information; the weighting module is used to weight the residual of the optimization factor by the inverse matrix of the covariance matrix, and determine whether the weighted residual is greater than a first threshold; the processing module processes the optimization factor information, and optimizes the point cloud trajectory based on the processed optimization factor information.

[0016] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for optimizing the point cloud mapping quality as described above is implemented.

[0017] In a fourth aspect, the present invention provides a computer-readable medium, which carries computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement the method for optimizing the point cloud mapping quality as described above.

[0018] Compared with the prior art, the method, device, electronic device and medium for optimizing the quality of point cloud mapping provided by the present invention introduce multi-source sensor factors into a unified residual quantitative analysis framework, and can effectively detect abnormal factor information before and after map optimization; by locking the index of the abnormal factor, it can be quickly eliminated or weighted corrected, reducing the distortion and flipping of the point cloud after optimization, and improving the stability and quality of point cloud mapping; this method realizes automated abnormality monitoring and correction in the point cloud mapping process, greatly reduces labor costs, and improves mapping quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A flowchart of a method for optimizing point cloud mapping quality in one embodiment of the present invention; Figure 2 Schematic diagram of a device for optimizing point cloud mapping quality in one embodiment of the present invention; Figure 3 FIG. 4 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0022] With the continuous development of fields such as autonomous driving, drones, and robots, point cloud mapping technology plays an increasingly important role in autonomous navigation and environmental modeling. In the process of point cloud mapping, sensors generate point clouds by acquiring three-dimensional data to build accurate environmental maps. Graph optimization technology globally optimizes point clouds by integrating the optimization factors of multiple sensors (such as odometers, IMUs, RTK, loopback factors, etc.) to improve the accuracy and robustness of maps. However, in practical applications, there are optimization distortion problems caused by sensor errors or conflicts in optimization factors in the optimization process, which makes it difficult to evaluate the quality of point clouds. Existing point cloud anomaly detection methods (such as layered detection) often cannot effectively cope with these challenges.

[0023] In the prior art, the quality assessment of point cloud mapping usually relies on traditional layered detection methods or manual inspection. When conflicts occur between optimization factors, traditional quality assessment methods cannot accurately capture anomalies or distortions in the optimization results, resulting in the need for manual intervention in the mapping process, which adds a lot of time and cost. In addition, the prior art lacks a quantitative and automated means to analyze the quality of point cloud mapping, and also lacks a strategy that can determine in real time whether anomalies occur in the point cloud optimization process.

[0024] In order to address these problems, the present invention proposes a method for optimizing the quality of point cloud mapping, which aims to detect and correct the optimization distortion that may exist in the point cloud mapping process through automated means. This method calculates the residual of the optimization factor in the point cloud trajectory, and combines the covariance matrix weighting method to determine whether the residual of the optimization factor exceeds the set threshold. If the residual exceeds the threshold, it means that there may be problems in the optimization process of the point cloud trajectory and further processing is required. This processing process can not only improve the quality assessment accuracy of point cloud mapping, but also effectively avoid the high cost of relying on manual inspection.

[0025] Specifically, the present invention optimizes the quality of point cloud mapping through the following steps: first, obtain the optimization factor information used for point cloud mapping, and then optimize the point cloud trajectory based on this information; then, calculate the residual of the optimization factor, and evaluate the size of the residual by inverse weighting of the covariance matrix. If the weighted residual exceeds the predetermined threshold, it means that there is an abnormality in the optimization process, and the optimization factor needs to be processed (such as increasing the noise weight or deleting the abnormal factor), and re-optimized based on the processed factor. This process ensures high precision and high robustness of point cloud mapping, especially when facing optimization factor conflicts and abnormal points, it can intervene in real time, avoiding the blind spots existing in traditional methods.

[0026] Please refer to Figure 1 FIG. 1 is a flow chart of a method for optimizing point cloud mapping quality in an embodiment of the present invention. The method for optimizing point cloud mapping quality specifically includes the following steps: S101: Obtain optimization factor information for point cloud mapping.

[0027] In practical applications, point cloud mapping usually relies on multiple sensors to provide position information and environmental data. For example, the odometry factor provides displacement information between adjacent time points; the RTK factor can provide more accurate global positioning information; the IMU factor provides dynamic data such as acceleration and angular velocity to estimate the motion state of the device; and the loop closure factor corrects the accumulated errors by identifying the device returning to a known position. These factors provide various constraints required to optimize the point cloud trajectory in different ways.

[0028] There are many ways to obtain these optimization factor information. For example, real-time data collected by sensor modules (such as lidar, IMU, RTK positioning system, etc.) can be transmitted to the computing platform through communication protocols, and the data is processed and fused on the computing platform. Specifically, the sensor data may be cleaned and formatted through a series of preprocessing steps (such as filtering, denoising, etc.) before being input into the graph optimization system. The acquisition of loop factors may rely on vision-based loop detection methods to identify when the device returns to a previous location and generate constraints.

[0029] For example, in the point cloud mapping task of an autonomous vehicle, the point cloud data provided by the lidar measures the distance and angle of the object through lidar scanning, while the IMU provides the dynamic information of the vehicle and the RTK positioning system provides high-precision geographic coordinates. All of this information together constitutes the input data of the optimization factor, and the effective acquisition and integration of this data is the prerequisite for ensuring the smooth progress of subsequent optimization calculations.

[0030] Odometry factor is a common constraint used in factor graph optimization. It provides pose constraints between adjacent trajectory points based on sensors such as lidar or wheel speed meter. In the factor graph, the odometer factor is usually used as a relative constraint factor (between factor), which links the pose information between two trajectory points and represents the pose changes at adjacent moments. By adding these pose constraints to the factor graph, the error accumulation of the trajectory can be effectively limited to avoid long-term drift. For example, when a vehicle passes through a known path, the odometer factor measured by the lidar can help constrain the position and attitude of the vehicle to ensure that its trajectory in the global coordinate system does not deviate too much. Therefore, the laser odometer factor is crucial in the optimization process, especially in the absence of an external positioning system. It can provide reliable local constraints and reduce errors caused by sensor noise.

[0031] The loop factor is an optimization factor provided by the loop detection module, which corrects the trajectory by identifying when the device returns to a known position. When the device returns to a previously passed area, the loop factor generates new constraint information based on the pose constraints between the device's current pose and the historical trajectory points. This constraint information is introduced into the factor graph optimization as a between factor to help adjust the global consistency of the entire trajectory. The loop factor is particularly suitable for scenarios with long-term operation or large-scale mapping. For example, in an application scenario of an autonomous vehicle, if the vehicle has passed the same place during the mapping process, the loop detection module can identify this repeated path and generate a corresponding loop factor for the trajectory, forcing the optimization algorithm to constrain adjacent poses and correct the offset caused by error accumulation, thereby greatly improving the accuracy of the global map.

[0032] The RTK satellite positioning factor provides accurate position constraints for track points through a high-precision global positioning system (GPS). The RTK (Real-time Kinematic Differential Positioning) factor is a unilateral factor (GPS factor) in graph optimization that specifically constrains the position of specific track points. Because the RTK system has high accuracy (the error is usually only a few centimeters), it is able to provide extremely accurate position information for the optimization process. For example, on urban roads or open areas, RTK positioning can provide accurate latitude and longitude coordinates for each track point as a benchmark for global optimization. Since GPS signals may be unstable or lost in certain environments (such as underground or in urban centers with dense high-rise buildings), the RTK satellite positioning factor is particularly suitable for areas with good satellite signals, providing strong global constraints for the optimization process.

[0033] The IMU factor is the constraint information provided by the inertial measurement unit (IMU). The IMU obtains the acceleration and angular velocity information of the device through the accelerometer and gyroscope to infer the motion state of the device. In graph optimization, the IMU factor is also used as a between factor to provide constraints for the position and posture between adjacent trajectory points. The IMU factor is particularly important in highly dynamic environments. It can infer the motion trajectory of the device through acceleration and angular velocity when the GPS signal is weak or completely unavailable. For example, in tunnels or indoor environments, the IMU factor provides reliable dynamic constraints for point cloud mapping by continuously tracking the motion state of the device. The IMU factor can effectively supplement the deficiencies of GPS and lidar and avoid error accumulation and drift.

[0034] S102: Optimizing the point cloud trajectory based on the optimization factor information.

[0035] In actual implementation, the steps of optimizing point cloud trajectories usually rely on graph optimization techniques. In the graph optimization framework, maps and trajectories are modeled as a graph, where each node represents a trajectory point or map element, and each edge represents a different type of constraint, such as pose constraint, position constraint, etc. The goal of graph optimization is to minimize the error of all edges in the graph so that the relationship between nodes is more consistent with the actual motion trajectory and environmental characteristics.

[0036] Specifically, based on the optimization factor information obtained in step S101, in step S102, these factors are first constructed as constraints in the graph in the form of a mathematical model. For example, the laser odometer factor provides constraints for the pose differences between adjacent track points, the IMU factor provides dynamic constraints for the track based on acceleration and angular velocity information, the RTK factor provides a high-precision reference for the absolute position of the track point, and the loopback factor provides a global position constraint based on the loopback detection information. After introducing this information as constraints into the factor graph, a graph optimization algorithm (such as the Levenberg-Marquardt algorithm, the Gauss-Newton algorithm, etc.) is used for calculation to solve the optimal track point pose.

[0037] For example, in an autonomous driving scenario, the vehicle obtains continuous point cloud data through the lidar, and obtains relevant trajectory information through the RTK positioning system and IMU sensor. During the driving process of the vehicle, the IMU factor provides constraints for the continuity of the trajectory by sensing the acceleration and angular velocity of the vehicle, and the odometer factor provides relative pose constraints between adjacent frames through radar data. As the vehicle drives, errors in RTK positioning may occur due to signal blocking and other reasons, causing the vehicle's trajectory to drift. At this time, the loop factor will play an important role. If the vehicle passes through the previous path, the loop detection module will identify and provide a constraint to correct the global position of the trajectory, thereby correcting the drift caused by the RTK positioning error. By integrating all this constraint information into the graph optimization, the optimization algorithm can adjust the trajectory so that the relative and absolute positions of all trajectory points are more in line with the actual situation.

[0038] The core function of this process is to minimize the error and noise of sensor data through the fusion of multi-source data, so as to make the point cloud trajectory more accurate. The optimized trajectory can significantly reduce the accumulation of errors, avoid the drift problem caused by long-term use of sensors, and improve the accuracy and reliability of point cloud mapping. For example, the optimized point cloud trajectory can more accurately reflect the geometric characteristics of the real environment, providing more accurate data support for subsequent map construction and environmental understanding.

[0039] In an exemplary embodiment, before optimizing the point cloud trajectory based on the optimization factor information, the method further includes: calculating the residual of the loop factor based on the point cloud trajectory and the loop factor information before optimization; weighting the residual of the loop factor with the inverse matrix of the covariance matrix to determine whether the weighted residual is greater than a second threshold; if so, deleting the corresponding loop factor.

[0040] The loop closure factor may produce errors in some cases, especially when the accuracy of loop closure detection is not high enough. The wrong loop closure factor may lead to deviations in the optimization results. Therefore, it is necessary to further verify and screen the loop closure factor before graph optimization.

[0041] The role of the loop factor is to provide a global position constraint by detecting whether a loop occurs in the point cloud trajectory. If the system detects that the current trajectory point is similar to a previous trajectory point, it can be considered that a loop has occurred, and a constraint is introduced for the relative pose between the two trajectory points. However, during the loop detection process, the loop factor sometimes produces errors due to the complexity of the environment or the noise of the sensor. For example, the sensor may incorrectly identify the loop due to short-term occlusion, signal loss, or calculation error, thereby introducing inappropriate loop constraints.

[0042] In order to effectively avoid this situation, the present invention adds a step before performing trajectory optimization, that is, calculating the residual of the loop closure factor and screening it. Specifically, this step detects whether it is reasonable by calculating the residual of the loop closure factor. The residual of the loop closure factor refers to the difference between the pose constraint provided by the loop closure factor and the actual optimized trajectory point. If the difference is too large, it means that the loop closure factor may be wrong and should be eliminated.

[0043] When calculating the residual of the loop closure factor, the system first needs to consider the covariance matrix of the loop closure factor. The covariance matrix reflects the credibility of the loop closure factor data. The loop closure factor with higher covariance shows that its data is not reliable enough, which may have a greater impact on the optimization result. For this reason, the present invention adopts the inverse matrix weighting method of the covariance matrix to process the residual of the loop closure factor. The inverse matrix weighting can weight its residual according to the covariance of the loop closure factor, and the residual produced by the loop closure factor with lower weight has less influence, while the loop closure factor with higher weight will have a greater impact on the judgment of the residual.

[0044] When the weighted loop closure factor residual is greater than the set second threshold, the system will consider that the loop closure factor does not meet the requirements and may be erroneous loop closure data, and decide to delete it from the optimization. The purpose of this is to avoid the negative impact of erroneous loop closure factors on the point cloud trajectory optimization process and prevent errors from being further propagated and accumulated. If the weighted residual is less than or equal to the second threshold, it means that the constraints of the loop closure factor are reasonable and can continue to participate in the optimization process.

[0045] S103: Calculating the residual of the optimization factor based on the optimized point cloud trajectory and the optimization factor information.

[0046] In the process of point cloud trajectory optimization, the optimization factors provide constraints, such as odometer, loop factor, RTK satellite positioning factor, IMU factor, etc. These factors reflect the error of trajectory estimation through the difference between the actual measurement value and the theoretical value. In step S103, the residual of the optimization factor is calculated by comparing the optimized point cloud trajectory with the measurement equation of the relevant factor. In other words, whether the optimized point cloud trajectory meets the expectations of theoretical calculation under the constraints of various factors is judged by calculating the residual.

[0047] In the specific implementation, it is assumed that a set of optimized trajectory data is obtained through the graph optimization algorithm. After these trajectory data are updated, the optimization factors (such as laser odometer, loop closure factor, RTK positioning, etc.) still need to provide constraints on the trajectory. When calculating the residual, first calculate the measurement equation based on the optimized trajectory and related optimization factors (such as the pose constraint provided by the odometer factor, the loop constraint provided by the loop closure factor, etc.) to obtain the theoretical calculation value corresponding to each factor. h ( x Then, calculate these theoretical values ​​and the actual measured values. Z The difference between This residual quantifies the deviation between the optimization result and the actual measurement data, and is an important basis for evaluating the optimization effect.

[0048] For example, in a SLAM system using LiDAR, the laser odometry factor provides a constraint on the pose change between two frames. The optimized trajectory data may deviate from the actual measured data, and the residual is calculated to quantify this deviation. If the optimized trajectory is more accurate, the residual should be close to zero; if the residual is large, it indicates that there is a problem with the trajectory optimization, which may be caused by incomplete constraints during the optimization process or low data quality.

[0049] The calculation of residuals can not only help determine whether the optimization is successful, but also provide feedback for the subsequent optimization process. In actual SLAM systems, factors with large residuals may affect the accuracy of the global map, so the weights of the optimization factors can be adjusted or the factors can be eliminated according to the residual values. For example, when the residual of the loop factor is too large, it may mean that the constraint information of the loop factor is wrong, resulting in the accumulation of trajectory errors. In this case, the loop factor can be weighted, or in extreme cases, the factor can be eliminated to avoid its negative impact on the optimization results.

[0050] In the framework of optimization factors, factors provide information for the optimization of point cloud trajectories through different types of constraints. Specifically, optimization factors are usually divided into relative constraint factors (between factors) and unilateral constraint factors (singlefactors), which have different roles and mathematical expressions in the optimization graph.

[0051] Relative constraint factors (such as odometer factor, IMU factor, and loop closure factor) are used to describe the relative pose relationship between two track points. These factors provide constraints between adjacent track points. For example, the odometer factor usually represents the pose change between two consecutive frames, while the IMU factor can represent the relative motion measured by the accelerometer or gyroscope.

[0052] For this type of optimization factor, the calculation formula is ,in, X 1 represents the pose variable of the first trajectory point, X 2 represents the pose variable of the second trajectory point. This formula reflects the relationship between trajectory points in the optimization. Specifically, it represents the position variable of the second trajectory point. X 1 to the second track point X 2’s posture transformation.

[0053] Unilateral constraint factors (such as RTK satellite positioning factors) usually provide constraint information for a single track point. The RTK factor provides global position information of the track point through GPS, usually expressed as a position constraint. For this type of factor, the calculation formula is ,in, X i Represents the sensor input of a trajectory point (such as the specific location of RTK positioning). The RTK factor is used to provide a global position constraint to ensure that certain key positions in the global map have relatively accurate position information.

[0054] S104: performing weighting of the inverse matrix of the covariance matrix on the residual of the optimization factor, and determining whether the weighted residual is greater than a first threshold.

[0055] Since the sensors used in practical applications may have different measurement accuracy and noise characteristics, it is necessary to weight the residuals of each sensor. The weighting process uses the inverse matrix of the covariance matrix. This processing method is derived from the concept of Mahalanobis distance and can eliminate the noise differences in the measurements of different sensors, ensuring that all residuals can be compared under the same standard during the optimization process, thereby improving the accuracy and robustness of point cloud mapping.

[0056] Sensors of the same type, such as LiDAR, IMU, GPS, etc., have different noise characteristics, which can lead to significant differences in the measurement accuracy and uncertainty of the same sensor in different environments. In order to effectively compare and evaluate the residuals of all sensors at a unified scale, inspired by the Mahalanobis distance, the information matrix (i.e., the inverse matrix of the covariance matrix) is introduced for weighting. The weighting of the information matrix can eliminate the differences in noise levels between different sensors, so that the residual of each sensor plays a role in the optimization that is consistent with its measurement accuracy.

[0057] Mahalanobis distance is a metric used to measure the distance between data points. It differs from Euclidean distance in that Mahalanobis distance can take into account the covariance structure of the data. Traditional Euclidean distance calculation relies only on the difference between data points in each dimension, assuming that the scales of all dimensions are the same and the features are independent of each other. However, in real data, especially in multidimensional data, features are usually correlated, and the variance of each feature may also be different. In this case, directly using Euclidean distance may be misleading because it cannot take into account the correlation and scale differences between features. Mahalanobis distance solves this problem by introducing the covariance matrix and its inverse matrix, so that the distance calculation can be performed at a unified scale, eliminating the correlation between features and the influence of different scales.

[0058] In the specific implementation, for the residual generated by each sensor, we first need to obtain the covariance matrix of the sensor Var , which reflects the uncertainty and noise characteristics of the sensor measurement. The inverse matrix of the covariance matrix is Var −1 It is called the information matrix, which is used to weight the residuals generated by each sensor. The introduction of the information matrix allows different residuals to be weighted under the same standard, taking into account the accuracy changes of sensors under different conditions. For example, in open areas, the quality of GPS signals is good and the measurement error is small, while in urban canyons, the accuracy of GPS is affected by the occlusion of buildings and the measurement error is large. In this case, the residuals of GPS need to be weighted according to the noise level of the environment in which they are located to avoid GPS data with too large errors occupying too much weight in the optimization.

[0059] This weighting method allows the sensor residuals to be compared on a uniform scale, thereby determining whether a residual is greater than a set threshold. In the formula, the weighted residual expression is ,in, Varis the covariance matrix, which reflects the uncertainty of the measurement. By weighting the residual of each sensor factor, the contribution of each sensor to the optimization result can be more accurately evaluated, and the inaccurate optimization problem caused by excessive measurement noise can be effectively avoided.

[0060] For example, in a specific application, suppose a SLAM system uses LiDAR, IMU, and GPS to build point cloud maps. In this system, the measurement data provided by LiDAR may be very accurate, and IMU can also provide relatively reliable relative motion information, but the GPS signal in an urban environment is often blocked by buildings, resulting in large measurement errors. At this time, if the GPS residual is not weighted, it may affect the optimization results of the entire system, because the GPS measurement error in the urban canyon is large, and the unweighted residual may be greater than the set threshold, thus being mistakenly judged as abnormal data. By introducing the inverse matrix of the covariance matrix for weighting, we can effectively eliminate these noise effects in the GPS measurement residuals, so that their contribution to the optimization results matches their measurement accuracy, thereby ensuring the stability and accuracy of the optimization.

[0061] This weighted residual can also be used to determine whether there are outliers. When the weighted residual is greater than a certain threshold, the system can consider the data to be abnormal and take corresponding measures, such as deleting the loop factor or re-adjusting the weight of the sensor. This method can effectively eliminate unreliable data caused by sensor failure or extreme environmental conditions, preventing it from causing unnecessary interference to the optimization process.

[0062] S105: If yes, process the optimization factor information, and optimize the point cloud trajectory based on the processed optimization factor information.

[0063] In the actual implementation process, the optimization factors may be affected by different factors, including the noise characteristics of the sensor, measurement errors, interference from the external environment, etc. These influences may cause some optimization factors to fail to accurately reflect the actual constraint relationship during the trajectory optimization process, and may even introduce abnormal data, thereby affecting the optimization effect. Therefore, in step S105, it is first necessary to properly process the optimization factors, exclude factors that do not meet the conditions, and ensure that only the factors with the most effective information are used for subsequent optimization calculations.

[0064] One possible approach is to optimize by weighting the optimization factors. In this process, each factor can be weighted according to its signal-to-noise ratio, measurement accuracy, and other information. For example, for the GPS sensor's posture factor, in open areas, the signal is strong and the measurement error is small, so a higher weight can be given; in urban canyons or other obstructed environments, the GPS signal is weak and the measurement error is large, so a lower weight can be given. This weighting method ensures that the system relies more on high-quality sensor data during optimization, reducing the impact of low-quality data on the optimization results.

[0065] Another way to deal with it is to filter or denoise the optimization factors. Sensor errors or abnormal data often cause some optimization factors to be inconsistent with the actual situation. Therefore, it is necessary to use some algorithms (such as Kalman filtering, mean filtering, etc.) to smooth the optimization factors and remove high-frequency noise and abnormal fluctuations. For example, if the residual of a loop factor is too large, it may indicate that the factor is affected by environmental interference or sensor failure. At this time, the system can determine whether the factor should be adjusted or eliminated by checking the trend of the residual change.

[0066] In addition, for some specific types of optimization factors, such as loop factors or IMU factors, these factors can be further optimized in combination with pose graph optimization algorithms. For example, through graph optimization methods, the system can convert the optimization factors into a pose graph, where each node represents a trajectory point and each edge represents an optimization factor (such as relative pose, GPS pose, or IMU measurement value). During the processing, by solving the minimization problem in the graph, the influence of the optimization factors can be effectively integrated into the optimization calculation of the point cloud trajectory, thereby improving the accuracy of the final optimization result.

[0067] In an exemplary embodiment, when the weighted residual is greater than a first threshold, the optimization factor information is processed, including: when the optimization factor information is loop factor information, deleting the optimization factor information; when the optimization factor information is odometer factor information, RTK satellite positioning factor information or IMU factor information, increasing the noise weight of the optimization factor information.

[0068] Regarding the processing of loop closure factor information, when the optimization factor is a loop closure factor, the system chooses to delete these optimization factors. This is because loop closure factors are usually used to correct the current trajectory through some known positions or map information to close the loop and optimize the trajectory. However, loop closure factors may produce large errors or inconsistencies due to factors such as environmental changes, untimely map updates, or sensor errors. If these errors are not handled correctly, they will have a negative impact on the optimization process of the point cloud trajectory. Therefore, deleting unreliable loop closure factors is an effective way to ensure the quality of optimization.

[0069] For other types of optimization factors such as odometer factors, RTK satellite positioning factors, and IMU factors, these factors usually have different degrees of noise in different environments and conditions, especially when the sensor works in a low-signal environment (such as urban canyons, tunnels, etc.), the noise will increase significantly. In order to avoid these factors from having too much influence on the final optimization results, the system reduces their influence in the optimization process by increasing the noise weight.

[0070] In summary, the method for optimizing the quality of point cloud mapping provided by the present invention introduces multi-source sensor factors into a unified residual quantitative analysis framework, which can effectively detect abnormal factor information before and after map optimization; by locking the index of the abnormal factor, it can be quickly eliminated or weighted corrected, reducing the distortion and flipping of the point cloud after optimization, and improving the stability and quality of point cloud mapping; this method realizes automated abnormality monitoring and correction in the point cloud mapping process, greatly reduces labor costs, and improves mapping quality and efficiency.

[0071] Please refer to Figure 2 As shown, based on the same inventive concept as the aforementioned method for optimizing point cloud mapping quality, the present invention provides a device 200 for optimizing point cloud mapping quality, which includes: a sensor 201, an optimization module 202, a calculation module 203, a weighting module 204 and a processing module 205.

[0072] The sensor 201 is used to obtain the optimization factor information of the point cloud mapping. The optimization module 202 is used to optimize the point cloud trajectory based on the optimization factor information. The calculation module 203 is used to calculate the residual of the optimization factor based on the optimized point cloud trajectory and the optimization factor information. The weighting module 204 is used to weight the residual of the optimization factor by the inverse matrix of the covariance matrix, and determine whether the weighted residual is greater than a first threshold. The processing module 205 processes the optimization factor information and optimizes the point cloud trajectory based on the processed optimization factor information.

[0073] Please refer to Figure 3 As shown, an embodiment of the present invention further provides an electronic device 300, which includes at least one processor 301, a memory 302 (such as a non-volatile memory), a memory 303, and a communication interface 304, and the at least one processor 301, the memory 302, the memory 303, and the communication interface 304 are connected together via an internal bus 305. At least one processor 301 is used to call at least one program instruction stored or encoded in the memory 302, so that the at least one processor 301 performs various operations and functions of the method for optimizing the point cloud mapping quality described in various embodiments of this specification.

[0074] In the embodiments of the present specification, the electronic device 300 may include, but is not limited to, personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile electronic devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDAs), handheld devices, messaging devices, wearable electronic devices, consumer electronic devices, and the like.

[0075] An embodiment of the present invention further provides a computer-readable medium carrying computer-executable instructions. When the computer-executable instructions are executed by a processor, they can be used to implement various operations and functions of the method for optimizing the point cloud mapping quality described in various embodiments of this specification.

[0076] The computer-readable medium in the present invention can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0077] In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0078] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatuses, systems, and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0080] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0081] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for optimizing point cloud mapping quality, characterized in that: include: Get the optimization factor information for point cloud mapping; Based on the optimization factor information, optimizing the point cloud trajectory; Based on the optimized point cloud trajectory and the optimization factor information, calculating the residual of the optimization factor; Performing weighting of the inverse matrix of the covariance matrix on the residual of the optimization factor, and determining whether the weighted residual is greater than a first threshold; If so, the optimization factor information is processed, and the point cloud trajectory is optimized based on the processed optimization factor information.

2. The method for optimizing point cloud mapping quality according to claim 1, characterized in that: The optimization factors include odometer factor, RTK satellite positioning factor, IMU factor and loopback factor.

3. The method for optimizing point cloud mapping quality according to claim 2, characterized in that: Before optimizing the point cloud trajectory based on the optimization factor information, the method further includes: Based on the point cloud trajectory and loop closure factor information before optimization, calculate the residual of the loop closure factor; Performing weighting of the inverse matrix of the covariance matrix on the residual of the loop closure factor, and determining whether the weighted residual is greater than a second threshold; If so, delete the corresponding loop factor.

4. The method for optimizing point cloud mapping quality according to claim 1, characterized in that: The residual calculation formula of the optimization factor is: ,in, h ( x ) is the theoretical calculated value of the optimization factor measurement equation, Z is the true measured value of the optimization factor.

5. The method for optimizing point cloud mapping quality according to claim 4, characterized in that: The formula for weighting the residual of the optimization factor by the inverse matrix of the covariance matrix is ​​expressed as: ,in, Var is the covariance matrix of the optimization factor information.

6. The method for optimizing point cloud mapping quality according to claim 4, characterized in that: When the optimization factor is a relative constraint factor, ,in, X 1 represents the pose variable of the first trajectory point, X 2 represents the pose variable of the second trajectory point; When the optimization factor is a unilateral constraint factor, ,in, X i Represents the sensor input at a trajectory point.

7. The method for optimizing point cloud mapping quality according to claim 1, characterized in that: Processing the optimization factor information includes: When the optimization factor information is loop factor information, deleting the optimization factor information; When the optimization factor information is odometer factor information, RTK satellite positioning factor information or IMU factor information, the noise weight of the optimization factor information is increased.

8. A device for optimizing the quality of point cloud mapping, characterized in that: include: Sensors, used to obtain optimization factor information for point cloud mapping; An optimization module, used for optimizing the point cloud trajectory based on the optimization factor information; A calculation module, used for calculating the residual of the optimization factor based on the optimized point cloud trajectory and the optimization factor information; A weighting module, used for performing weighting of the inverse matrix of the covariance matrix on the residual of the optimization factor, and determining whether the weighted residual is greater than a first threshold; The processing module processes the optimization factor information and optimizes the point cloud trajectory based on the processed optimization factor information.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for optimizing point cloud mapping quality as described in any one of claims 1 to 7 is implemented.

10. A computer-readable medium, characterized in that The computer-readable medium carries computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method for optimizing point cloud mapping quality according to any one of claims 1 to 7.

Images