A reliable positioning method and device for mobile robots

Abstract

The present invention relates to the field of mobile robot positioning and specifically provides a reliable mobile robot positioning method, characterized by comprising a reflector right-angle feature extraction module based on seed region growth, a global positioning module that integrates reflector right-angle features, and a local positioning module that integrates correlation scan matching. The reflector right-angle feature extraction module based on seed region growth extracts reflector right-angle features based on environmental point cloud information and light intensity information, and then constructs a feature dictionary database based on absolute position. The global positioning module that integrates reflector right-angle features matches to obtain robot pose particles with high pose confidence. The local positioning module that integrates correlation scan matching performs error minimization iteration to output high-precision real-time robot pose. Compared with existing technologies, this invention ensures real-time accuracy in robot operations, laying the foundation for multi-scenario applications of mobile robots.

Description

Technical Field

[0001] The present invention relates to the field of mobile robot positioning, and in particular provides a method and device for reliable positioning of a mobile robot. Background Art

[0002] With the rapid development of cutting-edge technologies such as computers, 5G communications, and sensors, robotics has leapt from its experimental stage to production applications, entering a new phase of development. At the same time, with current social trends such as the disappearance of the demographic dividend and the intensification of an aging population, the transformation of enterprises towards intelligent and flexible technologies is imperative. Demand for robots to replace humans in tasks that humans are unwilling to perform, cannot perform, or cannot perform well is growing.

[0003] Mobile robots, a key branch of robotics technology, are a relatively flexible type of intelligent robot within a robot cluster, capable of autonomous movement and environmental perception. They are used in a variety of fields, including welcoming and guiding guests, inspecting equipment rooms, performing sanitation and cleaning, and delivering goods. As demand for mobile robots continues to expand across multiple sectors, their operational environments are becoming increasingly complex and diverse. Simple trajectory tracking is no longer sufficient for these diverse applications. The high positioning accuracy and stability of mobile robots have become technical barriers to the development and commercialization of robots that can replace human operators.

[0004] In current practical applications, common and flexible robot positioning algorithms include: particle filter positioning algorithm based on laser point cloud information and positioning algorithm based on visual features.

[0005] The particle filtering method based on laser point cloud information is widely used due to its superiority in nonlinear and non-Gaussian systems. However, when it is actually applied in similar and relatively dynamic environments, there are certain particle degradation problems, resulting in low local positioning accuracy and poor global positioning effect. It cannot effectively solve unexpected events such as human "kidnapping" and sensor power failure, resulting in incorrect or sudden changes in the robot's posture, reduced positioning stability, and the inability to guarantee the reliability of operation positioning; the positioning algorithm based on visual features is easily affected by the environment, its application is relatively immature, and it is mostly used to assist lidar positioning. Summary of the Invention

[0006] The present invention aims to overcome the above-mentioned deficiencies in the prior art and provides a highly practical and reliable positioning method for a mobile robot.

[0007] A further technical task of the present invention is to provide a reliable positioning device for a mobile robot that is rationally designed, safe and applicable.

[0008] The technical solution adopted by the present invention to solve its technical problem is:

[0009] A reliable positioning method for a mobile robot includes a reflector right-angle feature extraction module based on seed region growth, a global positioning module integrating the reflector right-angle features, and a local positioning module integrating correlation scan matching.

[0010] The reflector right-angle feature extraction module based on seed region growth extracts the reflector right-angle features based on the environmental point cloud information and light intensity information, and then constructs a feature dictionary database under absolute position. When the mobile robot navigates, the identified reflector right-angle features are input into the global positioning module that integrates the reflector right-angle features.

[0011] The global positioning module integrating the right-angle features of the reflector scans and identifies the right-angle features of the reflector in the laser point cloud through the right-angle feature extraction module based on seed region growth, and then matches them with a given feature dictionary database to obtain robot pose particles with high pose confidence.

[0012] The local positioning module fused with correlation scan matching adopts the global positioning module fused with the right-angle features of the reflector to output the rough pose estimation of the robot, and the correlation scan matching uses the pose estimation as input and the environment map to iteratively minimize the error to output the high-precision real-time pose of the robot.

[0013] Furthermore, in the reflector right-angle feature extraction module based on seed region growth, the specific operation steps are as follows:

[0014] S1-1. Preliminary preparation of the mobile robot's operating environment;

[0015] S1-2, laser point cloud data preprocessing;

[0016] S1-3, line feature extraction based on seed region growth;

[0017] S1-4, extraction of right-angle features of reflector.

[0018] Furthermore, in step S1-1, a visible right-angle feature is searched in the working environment of the mobile robot, and reflectors of different sizes and lengths are installed to form a unique reflector right-angle feature;

[0019] In step S1-2, during the mapping and navigation process, the mobile robot obtains the original lidar point cloud data and completes the point cloud filtering operation according to the energy intensity threshold.

[0020] Furthermore, in step S1-3, continuous laser points are first selected for linear fitting based on the least squares method to complete the seed line segment extraction. The accuracy of seed line segment extraction directly affects the accuracy of the entire complete line segment extraction. To improve the accuracy of seed line segment extraction, the module sets the following extraction conditions:

[0021]

[0022] Where pl_dist i Represents the distance value from the i-th laser point to the seed segment used to fit the seed segment; ε pl_dist is the vertical distance threshold from the laser point to the seed segment, ε pl_dist The value is related to the accuracy of the laser radar itself. The seed segment extraction algorithm reduces the impact of discrete values ​​on the extraction accuracy based on this threshold; pp_dist i is the distance between the i-th laser point of the fitted seed segment and the predicted point based on the seed segment; δ pp_dist is the distance threshold between the laser point and the predicted point, δ pp_dist The value is selected based on empirical criteria. The seed segment extraction algorithm reduces the impact of the breakpoint value on the extraction accuracy based on the distance threshold. The calculation method of the predicted point based on the seed segment is as follows:

[0023]

[0024] Where, θ is the laser point (x i ,y i ) and the angle between the positive direction of the x-axis of the laser radar coordinate system. The above and below equations can be combined to solve the predicted point coordinates (x′ i ,y′ i );

[0025] Secondly, the region growing process checks the adjacent laser points of the seed point in the initial seed segment and determines whether the laser point has the same properties as the seed point. If they do, they are added to the same seed segment. The growth criterion of the seed region growth in the module is to judge the distance from the laser point in the neighborhood to the seed segment fitting line. If it is less than a given threshold, the laser point is added to the seed segment and the seed segment is refitted.

[0026] Finally, to improve the accuracy of line segment extraction, line segment merging is achieved based on the collinearity theory of continuous line segments and the distance difference between the laser point and the two line segments.

[0027] Furthermore, in step S1-4, based on the matching process of the rectangular model, the extracted continuous line features are traversed and the length of the line segment is compared with the actual reflector length value. If there are two or more continuous line segments in a frame of point cloud, it proves that the reflector rectangular feature exists;

[0028] Next, the processed line segments are matched by angle and distance. If the distance from one end point of a line segment to another line segment is less than dist thr 、The angle between the two line segments is 90°±θ thr These two right-angle model conditions, the length of the two line segments of the right-angle feature L i , Lj 、Corner point coordinates (x c ,y c ) is saved to the global reflector rectangular feature library, where the corner point coordinates (x c ,y c ) is the coordinate of the intersection of the two fitted lines in the world coordinate system, and the calculation formula is as follows:

[0029]

[0030] Furthermore, in the global positioning module that integrates the right-angle features of the reflector, the specific steps are as follows:

[0031] S2-1, extraction and matching of right-angle features of reflectors;

[0032] S2-2, global positioning initialization pose acquisition;

[0033] S2-3. Dealing with global positioning failure and robot kidnapping issues.

[0034] Furthermore, in step S2-1, when the mobile robot is in global pose initialization, robot abduction, and global positioning failure recovery, the robot obtains the laser point cloud information at the current moment and detects whether there are point clouds exceeding the energy threshold in the current frame point cloud and the number is greater than a given threshold;

[0035] If the conditions are met, the reflector right-angle feature extraction module based on seed region growth is used to extract the reflector right-angle feature, thereby completing the match with the reflector right-angle feature prior library and achieving the initial pose output; if the conditions are not met, the laser of this frame will not participate in the relevant calculations;

[0036] In step S2-2, the absolute pose value represented by the extracted right-angle feature of the reflector is used as the mean to expand the Gaussian distribution, and all particles after the Gaussian distribution are used as the input of the particle filter to complete the motion-free update, thereby obtaining a more accurate global positioning pose of the robot.

[0037] Furthermore, in step S2-3, the mobile robot often experiences particle degradation or positioning failure due to environmental factors during the navigation process. When the short-term likelihood value ω in the particle filter is fast Less than the long-term likelihood ω slow When , the average measurement probability of each particle in the particle set decreases, and the robot may experience positioning failure or machine kidnapping problems;

[0038] If there is a right-angle feature of the reflector within the observable range of the laser radar, the particle filter resampling process changes the way of adding global random particles, but instead adds a Gaussian distributed robot posture particle swarm. Add it to the current particle group of AMCL, speed up the convergence and aggregation speed by adding high-confidence particle sets, and perform global positioning.

[0039] Furthermore, in the local positioning module of the fusion correlation scan matching, the specific operation steps are as follows:

[0040] S3-1, particle filter coarse positioning output;

[0041] When a mobile robot navigates with a known global initial pose, it inputs the robot's dead-reckoning absolute pose prediction value, the current frame's lidar point cloud, and the prior obstacle grid map into a particle filter to calculate and output a value near the actual pose. When the robot experiences particle degradation and positioning failure, the global positioning module, which incorporates the reflector's right-angle features, adds relatively reliable particles during the particle filter resampling process to ensure positioning robustness.

[0042] S3-2, correlation scan matching precise positioning output;

[0043] The mobile robot iteratively calculates and outputs high-precision pose based on the obtained pose prediction value, the current frame lidar point cloud, and the prior obstacle grid map, and updates the absolute pose measurement value of the mobile robot. The algorithm cyclically calculates and completes the local positioning of the robot.

[0044] A reliable positioning device for a mobile robot, comprising: at least one memory and at least one processor;

[0045] The at least one memory is configured to store a machine-readable program;

[0046] The at least one processor is configured to call the machine-readable program to execute a reliable positioning method for a mobile robot.

[0047] Compared with the prior art, the reliable positioning method and device of a mobile robot of the present invention has the following outstanding beneficial effects:

[0048] The reflector right-angle feature extraction module based on seed region growth in the present invention adopts the method of constructing enhanced features based on the high-energy intensity information of the reflector, thereby ensuring the feature extraction accuracy of the robot in actual application scenarios where the laser point cloud is noisy and irregular and the effective point cloud is sparse, providing prior information for faster and more accurate robot positioning, and enhancing the robot's perception ability in similar and dynamic environments.

[0049] The global positioning module that integrates the right-angle features of reflectors can improve the problems of traditional mobile robot positioning methods such as poor global initialization positioning effect in large-scale, similar, and dynamic environments, low application adaptability, and the inability to solve positioning timeliness and robot kidnapping. For the problem of global initialization positioning, the "prior reflector right-angle feature absolute pose estimation of the robot's global pose" is used instead of the "manual setting of the initial pose", which has a fast particle convergence speed and improves positioning efficiency and accuracy. For the problems of positioning failure and robot kidnapping, the algorithm resampling process is improved, and the high-confidence robot pose particles estimated based on the reflector right-angle feature are added to the existing particle set. The particles in the particle set are quickly and effectively updated, and the true pose of the robot is restored.

[0050] The local positioning module integrated with correlation scan matching uses a small-capacity particle set to converge and output a rough pose estimate of the robot. Correlation scan matching uses this pose estimate as input and iterates with the environment map to minimize the error and output a high-precision real-time pose of the robot, ensuring the real-time accuracy of the robot operation and laying the foundation for the multi-scenario application of mobile robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Attachment Figure 1 It is a flow chart of a reliable positioning method for a mobile robot;

[0053] Attachment Figure 2 This is a flow chart of a reflector right-angle feature extraction module based on seed region growth in a reliable positioning method for a mobile robot;

[0054] Attachment Figure 3 This is a flow chart of a global positioning module that integrates the right-angle feature of a reflector in a reliable positioning method for a mobile robot;

[0055] Attachment Figure 4 The present invention is a flowchart of a local positioning module integrating correlation scan matching in a reliable positioning method for a mobile robot. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solutions of the present invention, the present invention will be further described in detail below in conjunction with specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0057] A best embodiment is given below:

[0058] like Figure 1 As shown, a reliable positioning method for a mobile robot in this embodiment includes a reflector right-angle feature extraction module based on seed region growth, a global positioning module integrating the reflector right-angle features, and a local positioning module integrating correlation scan matching;

[0059] The reflector right-angle feature extraction module based on seed region growth extracts the reflector right-angle features based on the environmental point cloud information and light intensity information, and then constructs a feature dictionary database under absolute position. When the mobile robot navigates, the identified reflector right-angle features are input into the global positioning module that integrates the reflector right-angle features.

[0060] The global positioning module, which integrates the right-angle features of reflectors, scans and identifies the right-angle features of reflectors in the laser point cloud through the right-angle feature extraction module based on seed region growth. It then matches the features with a given feature dictionary database to obtain robot pose particles with high pose confidence.

[0061] Among them, the local positioning module that integrates correlation scan matching uses the global positioning module that integrates the right-angle features of the reflector to output the robot's rough pose estimation, and the correlation scan matching uses the pose estimation as input and minimizes the error with the environment map to iteratively output the high-precision robot's real-time pose.

[0062] The reflector right-angle feature extraction module based on seed region growth takes into account that the structured environment in indoor scenes is more obvious, with more point and line features, and mobile robots often operate in complex and chaotic scenes. There are factors such as noisy and irregular laser point clouds, lidar measurement noise, and sparse effective point clouds that affect the accuracy of feature extraction, which in turn accumulates the positioning error of the mobile robot based on feature matching. To address the above problems, this module uses reflector enhancement features to extract right-angle key features with multi-scale uniqueness while reducing the amount of environmental features, thereby accurately and robustly constructing a global key feature library, thereby improving the anti-interference ability of mobile robot positioning. The reflector right-angle feature extraction module based on seed region growth mainly consists of two main parts: line feature extraction and reflector right-angle feature extraction. First, the module receives the original point cloud information fed back from the lidar, and then eliminates the point cloud with low energy intensity value according to the set energy intensity threshold, reducing the interference of invalid feature point cloud on the algorithm and improving the computational efficiency of the feature extraction process; the remaining valid point cloud obtains the coordinate information based on the lidar coordinate system through coordinate transformation, and inputs it into the seed region growth line feature extraction algorithm to complete the line feature extraction; finally, the right-angle feature extraction and recording of the reflector are completed through the right-angle feature verification part.

[0063] The reflector right-angle feature extraction module based on seed region growth includes the following steps:

[0064] S1-1. Preliminary preparation of the mobile robot's operating environment;

[0065] Visible right-angle features are found in the working environment of the mobile robot, and reflectors of different sizes and lengths are installed to form unique reflector right-angle features.

[0066] S1-2, laser point cloud data preprocessing;

[0067] During the mapping and navigation process, the mobile robot obtains the original lidar point cloud data and completes the point cloud filtering operation according to the energy intensity threshold.

[0068] S1-3, line feature extraction based on seed region growth;

[0069] First, continuous laser points are selected for linear fitting based on the least squares method to complete the seed line segment extraction. The accuracy of seed line segment extraction directly affects the accuracy of the entire complete line segment extraction. To improve the accuracy of seed line segment extraction, the module sets the following extraction conditions:

[0070]

[0071] Where pl_dist i Represents the distance value from the i-th laser point to the seed segment used to fit the seed segment; ε pl_distis the vertical distance threshold from the laser point to the seed segment, ε pl_dist The value is related to the accuracy of the laser radar itself. The seed segment extraction algorithm reduces the impact of discrete values ​​on the extraction accuracy based on this threshold; pp_dist i is the distance between the i-th laser point of the fitted seed segment and the predicted point based on the seed segment; δ pp_dist is the distance threshold between the laser point and the predicted point, δ pp_dist The value is selected based on empirical criteria. The seed segment extraction algorithm reduces the impact of the breakpoint value on the extraction accuracy based on the distance threshold. The calculation method of the predicted point based on the seed segment is as follows:

[0072]

[0073] Where, θ is the laser point (x i ,y i ) and the angle between the positive direction of the x-axis of the laser radar coordinate system. The above and below equations can be combined to solve the predicted point coordinates (x′ i ,y′ i );

[0074] Secondly, the region growing process checks the adjacent laser points of the seed point in the initial seed segment and determines whether the laser point has the same properties as the seed point. If they do, they are added to the same seed segment. The growth criterion of the seed region growth in the module is to judge the distance from the laser point in the neighborhood to the seed segment fitting line. If it is less than a given threshold, the laser point is added to the seed segment and the seed segment is refitted.

[0075] Finally, to improve the accuracy of line segment extraction, line segment merging is achieved based on the collinearity theory of continuous line segments and the distance difference between the laser point and the two line segments to solve the overlapping phenomenon caused by noise points and breakpoints.

[0076] S1-4, reflector right angle feature extraction;

[0077] First, based on the matching process of the rectangular model, the calculation results of S1-1 to S1-3 are used as known information, that is, the linear features of the reflector extracted in multiple segments. Then, based on this information, it is determined whether there is a rectangular feature in a frame of point cloud that meets the following matching model. If it meets the requirements, it is extracted.

[0078] Specifically, we traverse and extract continuous line features, and compare the length of the line segment with the actual length of the reflector. If there are two or more continuous line segments in a frame of point cloud, it proves that there is a right-angle feature of the reflector.

[0079] Next, the processed line segments are matched by angle and distance. If the distance from one end point of a line segment to another line segment is less than dist thr 、The angle between the two line segments is 90°±θthr These two right-angle model conditions, the length of the two line segments of the right-angle feature L i , L j 、Corner point coordinates (x c ,y c ) is saved to the global reflector rectangular feature library, where the corner point coordinates (x c ,y c ) is the coordinate of the intersection of the two fitted lines in the world coordinate system, and the calculation formula is as follows:

[0080]

[0081]

[0082] The global positioning module that integrates the right-angle features of reflectors uses the "prior reflector right-angle feature absolute pose estimation of the robot's global pose" instead of "manually setting the initial pose" to achieve more robust and accurate global positioning. When the mobile robot is performing global positioning or encounters global positioning failure or robot kidnapping problems, if the reflector features are complete within the LiDAR observation range, the reflector right-angle features can be extracted through the reflector right-angle feature extraction module based on seed region growth, and the position coordinates of the current feature corner point are obtained using the length of the two right-angle sides as the matching criterion. Finally, the global coordinate estimation value of the mobile robot (x r ′,y r ′,θ r ′), but there are still interferences such as measurement noise in the matching process. Therefore, this module generates a Gaussian pose sequence based on the estimated pose and uses it as the initial input particles of the particle filter or the random particles in the resampling after particle degradation. At this time, the particle set changes from uniform distribution to Gaussian distribution, the algorithm convergence time is reduced, and the global positioning efficiency is improved.

[0083] The global positioning module that integrates the right-angle features of the reflector includes the following steps:

[0084] S2-1, extraction and matching of right-angle features of reflectors;

[0085] When the mobile robot is initializing its global posture, kidnapping, or recovering from global positioning failure, the robot obtains the laser point cloud information at the current moment and detects whether there are point clouds in the current frame that exceed the energy threshold and the number is greater than a given threshold;

[0086] If the conditions are met, the reflector right-angle features are extracted through the reflector right-angle feature extraction module based on seed region growth, thereby completing the matching with the reflector right-angle feature prior library and realizing the initial pose output; if the conditions are not met, the laser of this frame does not participate in the relevant calculations.

[0087] S2-2, global positioning initialization pose acquisition;

[0088] The absolute pose value represented by the extracted right-angle feature of the reflector is used as the mean to expand the Gaussian distribution, and all particles after the Gaussian distribution are used as the input of the particle filter to complete the motion-free update, so as to quickly obtain a more accurate global positioning pose of the robot.

[0089] S2-3, handling of global positioning failure and robot kidnapping issues;

[0090] Global positioning failure and robot kidnapping problem processing. Mobile robots often experience particle degradation or positioning failure due to environmental factors during navigation. When the short-term likelihood value ω in the particle filter is fast Less than the long-term likelihood ω slow When the average measurement probability of each particle in the particle set decreases, the robot may experience positioning failure or machine kidnapping problems. If there is a right-angle feature of the reflector within the observable range of the lidar at this time, the particle filter resampling process changes the way of adding global random particles, but instead increases the robot posture particle swarm with Gaussian distribution. By adding it to the current particle group of AMCL, the convergence speed is accelerated by increasing the high-confidence particle set, and the global positioning efficiency and global positioning accuracy are improved.

[0091] The local positioning module integrated with correlation scan matching divides the robot's local positioning process into two parts: particle filter coarse positioning output and correlation scan matching fine positioning. Particle filter uses a small-capacity particle set to converge and output a rough pose estimate of the robot. Correlation scan matching uses this pose estimate as input and performs iterative matching with the environment map to minimize the error and output a high-precision pose. When the robot starts local positioning, the global initial pose of the robot and the current wheel speed v of the two driving wheels of the wheel odometer are known. l 、v r and the gyroscope angle increment Δθ, and the absolute pose measurement value is calculated by the odometry track dead reckoning method.

[0092] The local positioning module integrating correlation scan matching includes the following steps:

[0093] S3-1, particle filter coarse positioning output;

[0094] When a mobile robot navigates with a known global initial pose, the robot's absolute pose prediction value calculated by dead reckoning, the current frame lidar point cloud, and the prior obstacle grid map are input into the particle filter to calculate and output values ​​near the true pose. When the robot experiences particle degradation and positioning failure, the global positioning module that integrates the right-angle features of the reflector adds relatively reliable particles during the particle filter resampling process to ensure positioning robustness.

[0095] S3-2, correlation scan matching precise positioning output;

[0096] The mobile robot iteratively calculates and outputs a high-precision pose based on the obtained rough true pose, the current frame laser point cloud, and the prior obstacle grid map, and updates the absolute pose measurement value of the mobile robot. The algorithm performs cyclic calculations and can complete the local positioning of the robot efficiently and accurately.

[0097] Based on the above method, a reliable positioning device for a mobile robot in this embodiment includes: at least one memory and at least one processor;

[0098] The at least one memory is configured to store a machine-readable program;

[0099] The at least one processor is configured to call the machine-readable program to execute a reliable positioning method for a mobile robot.

[0100] The above-mentioned specific implementation manner is only a specific case of the present invention, and the patent protection scope of the present invention includes but is not limited to the above-mentioned specific implementation manner.

[0101] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A reliable positioning method for a mobile robot, characterized in that , including a reflector right-angle feature extraction module based on seed region growth, a global positioning module integrating the reflector right-angle features, and a local positioning module integrating correlation scan matching; The reflector right-angle feature extraction module based on seed region growth extracts the reflector right-angle features based on the environmental point cloud information and light intensity information, and then constructs a feature dictionary database under absolute position. When the mobile robot navigates, the identified reflector right-angle features are input into the global positioning module that integrates the reflector right-angle features. The global positioning module integrating the right-angle features of the reflector scans and identifies the right-angle features of the reflector in the laser point cloud through the right-angle feature extraction module based on seed region growth, and then matches them with a given feature dictionary database to obtain robot pose particles with high pose confidence. The local positioning module fused with correlation scan matching uses the global positioning module fused with the right-angle features of the reflector to output a rough pose estimate of the robot, and the correlation scan matching uses the pose estimate as input and performs iterative minimization of the error with the environment map to output a high-precision real-time pose of the robot; In the reflector right-angle feature extraction module based on seed region growth, the specific operation steps are as follows: S1-1. Preliminary preparation of the mobile robot's operating environment; S1-2, laser point cloud data preprocessing; S1-3, line feature extraction based on seed region growth; S1-4, reflector right angle feature extraction; In the global positioning module that integrates the right-angle features of the reflector, the specific steps are as follows: S2-1, extraction and matching of right-angle features of reflectors; S2-2, global positioning initialization pose acquisition; S2-3, handling of global positioning failure and robot kidnapping issues; During the navigation process, mobile robots often experience particle degradation or positioning failure due to environmental factors. When the short-term likelihood value ω in the particle filter is fast Less than the long-term likelihood ω slow When , the average measurement probability of each particle in the particle set decreases, and the robot may experience positioning failure or machine kidnapping problems; If there is a right-angle feature of the reflector within the observable range of the laser radar, the particle filter resampling process changes the way of adding global random particles, but instead adds a Gaussian distributed robot posture particle swarm. Add to the current particle group of AMCL, speed up the convergence and aggregation speed by adding high-confidence particle sets, and perform global positioning; In the local positioning module of the fusion correlation scan matching, the specific operation steps are as follows: S3-1, particle filter coarse positioning output; When a mobile robot navigates with a known global initial pose, it inputs the robot's dead-reckoning absolute pose prediction value, the current frame's lidar point cloud, and the prior obstacle grid map into a particle filter to calculate and output a value near the true pose. When the robot experiences particle degradation and positioning failure, the global positioning module, which incorporates the reflector's right-angle features, adds relatively reliable particles during the particle filter resampling process to ensure positioning robustness. S3-2, correlation scan matching precise positioning output; The mobile robot iteratively calculates and outputs high-precision pose based on the obtained pose prediction value, the current frame lidar point cloud, and the prior obstacle grid map, and updates the absolute pose measurement value of the mobile robot. The algorithm cyclically calculates and completes the local positioning of the robot.

2. A reliable positioning method for a mobile robot according to claim 1, characterized in that: In step S1-1, a visible right-angle feature is searched in the working environment of the mobile robot, and reflectors of different sizes and lengths are installed to form a unique reflector right-angle feature; In step S1-2, during the mapping and navigation process, the mobile robot obtains the original lidar point cloud data and completes the point cloud filtering operation according to the energy intensity threshold.

3. A reliable positioning method for a mobile robot according to claim 2, characterized in that: In step S1-3, continuous laser points are first selected for linear fitting based on the least squares method to complete the seed line segment extraction. The accuracy of seed line segment extraction directly affects the accuracy of the entire complete line segment extraction. To improve the accuracy of seed line segment extraction, the module sets the following extraction conditions: Where pl_dist i Represents the distance value from the i-th laser point to the seed segment used to fit the seed segment; ε pl_dist is the vertical distance threshold from the laser point to the seed segment, ε pl_dist The value is related to the accuracy of the laser radar itself. The seed segment extraction algorithm reduces the impact of discrete values ​​on the extraction accuracy based on this threshold; pp_dist i is the distance between the i-th laser point of the fitted seed segment and the predicted point based on the seed segment; δ pp_dist is the distance threshold between the laser point and the predicted point, δ pp_dist The value is selected based on empirical criteria. The seed segment extraction algorithm reduces the impact of the breakpoint value on the extraction accuracy based on the distance threshold. The calculation method of the predicted point based on the seed segment is as follows: Where, θ is the laser point (x i ,y i ) and the angle between the positive direction of the x-axis of the laser radar coordinate system. The above and below equations can be combined to solve the coordinates of the predicted point (x i ′,y i ′); Secondly, the region growing process checks the adjacent laser points of the seed point in the initial seed segment and determines whether the laser point has the same properties as the seed point. If so, it is added to the same seed segment. The growth criterion of the seed region growth in the module is to judge the distance from the laser point in the neighborhood to the seed segment fitting line. If it is less than a given threshold, the laser point is added to the seed segment and the seed segment is refitted. Finally, to improve the accuracy of line segment extraction, line segment merging is achieved based on the collinearity theory of continuous line segments and the distance difference between the laser point and the two line segments.

4. A reliable positioning method for a mobile robot according to claim 3, characterized in that: In step S1-4, based on the matching process of the rectangular model, the continuous line features extracted are traversed and the length of the line segment is compared with the actual length of the reflector. If there are two or more continuous line segments in a frame of point cloud, it proves that the reflector rectangular feature exists; Next, the processed line segments are matched by angle and distance. If the distance from one end point of a line segment to another line segment is less than dist thr 、The angle between the two line segments is 90°±θ thr These two right-angle model conditions, the length of the two line segments of the right-angle feature L i , L j 、Corner point coordinates (x c ,y c ) is saved to the global reflector rectangular feature library, where the corner point coordinates (x c ,y c ) is the coordinate of the intersection of the two fitted lines in the world coordinate system, and the calculation formula is as follows:

5. A reliable positioning method for a mobile robot according to claim 4, characterized in that: In step S2-1, when the mobile robot is initializing its global posture, abducting the robot, or recovering from global positioning failure, the robot obtains the laser point cloud information at the current moment and detects whether there are point clouds in the current frame point cloud that exceed the energy threshold and the number is greater than a given threshold; If the conditions are met, the reflector right-angle feature extraction module based on seed region growth is used to extract the reflector right-angle feature, thereby completing the match with the reflector right-angle feature prior library and achieving the initial pose output; if the conditions are not met, the laser of this frame will not participate in the relevant calculations; In step S2-2, the absolute pose value represented by the extracted right-angle feature of the reflector is used as the mean to expand the Gaussian distribution, and all particles after the Gaussian distribution are used as the input of the particle filter to complete the motion-free update, thereby obtaining a more accurate global positioning pose of the robot.

6. A reliable positioning device for a mobile robot, characterized in that: include: at least one memory and at least one processor; The at least one memory is configured to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method according to any one of claims 1 to 5.

Images