A body intelligent robot pose recognition method and device, electronic equipment, and storage medium

Abstract

This invention relates to the field of mobile robot positioning and navigation technology, and in particular to a method, device, electronic device, and storage medium for pose recognition of an embodied intelligent robot. By adaptively extracting highly reflective feature point clouds from the original point cloud and performing rigorous single-label clustering and dual-label pairing verification, it effectively filters out interfering reflective points that occasionally appear in dynamic environments, reducing the dependence on static environments and fixed geometric layouts compared to traditional methods. By using principal component analysis combined with sensor radial observation constraints, the center and direction of each label are accurately calculated, and the fixed geometric relationship of dual labels is fused to solve for high-precision intermediate poses. Finally, by performing spatiotemporal consistency filtering and fusion on intermediate poses of multiple consecutive frames, single-frame measurement noise and instantaneous occlusion interference are effectively suppressed, significantly improving the stability and reliability of pose output.

Description

Technical Field

[0001] This invention relates to the field of mobile robot positioning and navigation technology, and in particular to a method, device, electronic device, and storage medium for pose recognition of an embodied intelligent robot. Background Technology

[0002] Embodied AI robots are robots possessing embodied intelligence capabilities, consisting of a robot body and an intelligent agent, enabling them to perform tasks in complex environments. When performing autonomous navigation and docking tasks, embodied AI robots need to acquire their relative pose with the docking device in real time and with high accuracy; this is a crucial prerequisite for stable and reliable autonomous operation. However, in real-world dynamic working environments, due to interference factors such as pedestrians and moving equipment, traditional pose recognition methods often struggle to simultaneously meet multiple requirements, including high precision, strong robustness, and good environmental adaptability, thus limiting the effectiveness of robot applications in complex scenarios.

[0003] Currently, common pose recognition methods mainly fall into two categories: one is laser-based SLAM (simultaneous localization and mapping) localization technology, which estimates robot pose by matching real-time point clouds with pre-built environmental maps. However, its performance is highly dependent on the static characteristics of the environment and the accuracy of the map, and it is prone to matching deviations in dynamically changing scenarios. The recognition accuracy is usually only at the centimeter level. The other category is to place highly reflective markers (such as reflective pillars) near the docking target and calculate the pose based on their preset spatial distribution. Although this method can improve the recognition stability in specific scenarios, it requires additional installation and long-term maintenance of these auxiliary markers, which not only occupies physical space and increases deployment costs, but also affects the overall integrity and aesthetics of the application scenario.

[0004] Overall, existing technical solutions still have significant shortcomings in practical applications: laser SLAM methods are greatly affected by environmental dynamics, have limited accuracy, and are not very adaptable; while methods based on fixed reflective aids can improve stability, they sacrifice deployment flexibility and convenience. Therefore, there is an urgent need for a pose recognition method that can achieve high accuracy and high stability in dynamic environments without requiring complex environmental modifications, in order to promote the autonomous docking and application of embodied intelligent robots in a wider range of scenarios. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method, device, electronic device, and storage medium for recognizing the pose of an embodied intelligent robot.

[0006] In a first aspect, embodiments of the present invention provide a pose recognition method for an embodied intelligent robot, the method comprising: Acquire point cloud data of the environment in which the embodied intelligent robot is located; Extract feature point clouds with preset reaction characteristics from point cloud data; Spatial clustering and pairing verification are performed on the feature point cloud to identify paired feature identifier point cloud clusters that conform to the preset spatial layout; Based on the geometric distribution characteristics of paired feature marker cloud clusters and sensor observation constraints, the intermediate pose of the embodied intelligent robot relative to the paired feature marker cloud clusters is calculated. The target pose is output by performing temporal filtering and fusion based on the intermediate poses of multiple consecutive frames.

[0007] In conjunction with the first aspect, the steps for extracting feature point clouds with preset reaction characteristics from point cloud data include: For each scan point in the point cloud data, the reflection intensity threshold corresponding to the measurement distance of the scan point is determined based on the preset dynamic reflection intensity threshold function relationship; If the actual reflection intensity of the scanned point is higher than the reflection intensity threshold, the scanned point is determined to be a feature point; By combining all the feature points, a feature point cloud is obtained.

[0008] In conjunction with the first aspect, the steps of spatial clustering and pairing verification of feature point clouds to identify paired feature-identified point cloud clusters that conform to a preset spatial layout include: Single-label clustering is performed on the feature point cloud to obtain several candidate point cloud clusters; Each candidate point cloud cluster is filtered by geometric size to select single-label point cloud clusters that meet the preset label size; Calculate the centroid of each single identifier point cloud cluster, and pair two single identifier point cloud clusters with matching centroid distances into a pair of feature identifier point cloud clusters according to the preset identifier spacing and tolerance range.

[0009] In conjunction with the first aspect, the steps for single-label clustering of feature point clouds include: Traverse each scan point in the feature point cloud, calculate the Euclidean distance between the current scan point and the next scan point, and the angle formed by the sensor origin, the current scan point, and the next scan point; If both the Euclidean distance and the angular size are less than the preset threshold, the next scan point will be assigned to the point cloud cluster where the current scan point is located. Otherwise, create a new point cloud cluster starting from the next scan point.

[0010] In conjunction with the first aspect, the step of pairing two single-marker point cloud clusters with matching centroid distances into paired feature marker point cloud clusters, based on preset marker spacing and tolerance range, includes: Calculate the Euclidean distance between the centroids of any two single-identifier point cloud clusters; Determine whether the Euclidean distance falls within the preset spacing matching range. The spacing matching range is determined based on the standard double-marker spacing and the preset tolerance coefficient. If so, the two corresponding single-identifier point cloud clusters will be paired into a pair of feature identifier point cloud clusters.

[0011] In conjunction with the first aspect, based on the geometric distribution characteristics of paired feature-labeled point cloud clusters and sensor observation constraints, the steps for calculating the intermediate pose of the embodied intelligent robot relative to the paired feature-labeled point cloud clusters include: For each single-identifier point cloud cluster in a pair of feature-identifier point cloud clusters, principal component analysis is performed on the two-dimensional coordinates of the single-identifier point cloud cluster to obtain the feature vector representing the main direction of the point cloud distribution. A fitted straight line is determined based on the feature vector and the center point of the single-identifier point cloud cluster; Select the start and end scanning points of the paired feature marker cloud clusters in the scanning sequence; Construct a first radial constraint line passing through the starting scan point and the sensor origin, and a second radial constraint line passing through the ending scan point and the sensor origin, respectively. Calculate the first intersection point of the fitted line corresponding to the pair of feature marker cloud clusters with the first radial constraint line, and the second intersection point of the fitted line with the second radial constraint line; Based on the first intersection point and the second intersection point, determine the coordinates of the center point of the identifier corresponding to the single identifier cloud cluster; Based on the center point coordinates and the main direction, determine the attitude yaw angle of the corresponding markers of the paired feature marker cloud clusters; Based on the attitude yaw angle, the intermediate pose of the embodied intelligent robot relative to the paired feature markers is determined, thus obtaining the intermediate pose of the embodied intelligent robot relative to the paired feature markers.

[0012] In conjunction with the first aspect, the steps of performing temporal filtering and fusion based on intermediate poses of multiple consecutive frames to output the target pose include: Among the multiple pose recognition results corresponding to the current point cloud frame, the one closest to the embodied intelligent robot is selected as the valid pose of the current frame. Multiple valid poses are obtained from multiple consecutive point cloud frames; Cluster analysis based on spatiotemporal consistency is performed on multiple valid poses to eliminate outlier poses caused by transient disturbances; The target pose is obtained by fusing the pose set retained after cluster analysis.

[0013] Secondly, this application provides a pose recognition device for an embodied intelligent robot, the device comprising: The acquisition module is used to acquire point cloud data of the environment in which the embodied intelligent robot is located; The extraction module is used to extract feature point clouds with preset reaction characteristics from point cloud data; The identification module is used to perform spatial clustering and pairing verification on the feature point cloud, and identify the pair of feature identifier point cloud clusters that conform to the preset spatial layout; The calculation module is used to calculate the intermediate pose of the embodied intelligent robot relative to the pair of feature marker cloud clusters based on the geometric distribution characteristics of the pair of feature marker cloud clusters and the sensor observation constraints. The fusion module is used to perform temporal filtering and fusion based on the intermediate poses of multiple consecutive frames, and output the target pose.

[0014] Thirdly, this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor runs the computer program to cause the electronic device to perform the above-described method.

[0015] Fourthly, this application provides a storage medium storing computer program instructions, which are read and executed by a processor to perform the above-described method.

[0016] The embodiments of the present invention bring the following beneficial effects: The present application provides a method, device, electronic device, and storage medium for embodied intelligent robot pose recognition. The method includes: acquiring point cloud data of the environment in which the embodied intelligent robot is located; extracting feature point clouds with preset response characteristics from the point cloud data; performing spatial clustering and pairing verification on the feature point clouds to identify paired feature marker point cloud clusters that conform to a preset spatial layout; calculating the intermediate pose of the embodied intelligent robot relative to the paired feature marker point cloud clusters based on the geometric distribution characteristics of the paired feature marker point cloud clusters and sensor observation constraints; and performing temporal filtering and fusion based on the intermediate poses of multiple consecutive frames to output the target pose.

[0017] This invention adaptively extracts highly reflective feature points from the original point cloud and performs rigorous single-label clustering and dual-label pairing verification, effectively filtering out occasional interfering reflective points in dynamic environments. Compared to traditional methods, it reduces dependence on static environments and fixed geometric layouts. By utilizing principal component analysis combined with sensor radial observation constraints, the center and orientation of each label are accurately calculated, and the fixed geometric relationship of dual labels is fused to solve for high-precision intermediate poses. Finally, by performing spatiotemporal consistency filtering and fusion on intermediate poses across multiple consecutive frames, single-frame measurement noise and transient occlusion interference are effectively suppressed, significantly improving the stability and reliability of pose output.

[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a schematic flowchart of a pose recognition method for an embodied intelligent robot provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the dynamic reflection intensity threshold function provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the implementation process of the pose recognition method for an embodied intelligent robot provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a pose recognition device for an embodied intelligent robot provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the electronic device structure provided in an embodiment of the present invention.

[0022] Figure label: 10 - Acquisition module, 20 - Extraction module, 30 - Recognition module, 40 - Calculation module, 50 - Fusion module; 130 - Processor, 131 - Memory, 132 - Bus, 133 - Communication interface. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To facilitate understanding of this embodiment, the technical terms used in this application will be briefly introduced below.

[0025] SLAM (Simultaneous Localization and Mapping) is a technology that enables embodied intelligent robots to build a map of their surroundings in real time using onboard sensors (such as LiDAR and cameras) in unknown environments, and to estimate their own position and orientation using the map.

[0026] After introducing the technical terms used in this application, the application scenarios and design concepts of the embodiments of this application will be briefly described below.

[0027] Embossed intelligent robots require high-precision relative pose for navigation and docking. However, traditional laser SLAM methods are highly dependent on the environment and have limited accuracy under dynamic conditions. Adding fixed reflective posts, on the other hand, presents problems such as space occupation and inconvenient installation and maintenance. Existing technologies still have shortcomings in terms of dynamic environment adaptability, recognition accuracy, and ease of deployment, which limits the autonomous docking capabilities of embossed intelligent robots.

[0028] Based on this, this application provides a method, device, electronic device, and storage medium for embodying intelligent robot pose recognition, so as to improve the operational efficiency and reliability of autonomous docking tasks.

[0029] Example 1 This application provides a method for pose recognition of an embodied intelligent robot, combined with... Figure 1 As shown, the method includes: S110 acquires point cloud data of the environment in which the embodied intelligent robot is located.

[0030] S120, extract feature point clouds with preset reaction characteristics from point cloud data.

[0031] S130, spatial clustering and pairing verification are performed on the feature point cloud to identify paired feature identifier point cloud clusters that conform to the preset spatial layout.

[0032] S140, based on the geometric distribution characteristics of the paired feature marker cloud clusters and sensor observation constraints, calculates the intermediate pose of the embodied intelligent robot relative to the paired feature marker cloud clusters.

[0033] S150 performs temporal filtering and fusion based on the intermediate poses of multiple consecutive frames to output the target pose.

[0034] This application adaptively extracts highly reflective feature point clouds from the original point cloud and performs rigorous single-label clustering and dual-label pairing verification, effectively filtering out occasional interfering reflective points in dynamic environments. Compared with traditional methods, it reduces the dependence on static environments and fixed geometric layouts. By using principal component analysis combined with sensor radial observation constraints, the center and direction of each label are accurately calculated, and the fixed geometric relationship of dual labels is fused to solve for high-precision intermediate poses. Finally, by performing spatiotemporal consistency filtering and fusion on intermediate poses of multiple consecutive frames, single-frame measurement noise and instantaneous occlusion interference are effectively suppressed, significantly improving the stability and reliability of pose output.

[0035] In step S110, the surrounding environment is periodically scanned and perceived using a lidar sensor mounted on an embodied intelligent robot (e.g., an AMR). After each scan, the raw data directly output by the lidar contains a set of scan points, where each scan point contains at least three raw measurement values: The distance measurement value refers to the straight-line distance from the scanning point to the lidar sensor; The heading angle is the horizontal angle of the scanning point relative to the zero-degree reference direction of the lidar sensor (usually directly in front of the robot). The laser intensity (reflection intensity) value is the signal intensity returned after the laser beam irradiates the surface of an object, and it is directly related to the reflection characteristics of the object's surface.

[0036] To facilitate subsequent processing, the raw measurement data based on polar coordinates (distance and angle) needs to be converted into two-dimensional or three-dimensional Cartesian coordinates in the lidar sensor's own coordinate system (i.e., the sensor coordinate system). For two-dimensional lidar, the typical calculation formula is:

[0037]

[0038] in, This represents the x-coordinate of the scan point in the sensor coordinate system. The vertical coordinate of the scanning point in the sensor coordinate system; This represents the straight-line distance from the scanning point to the lidar sensor. This is the heading angle.

[0039] This determines the coordinates of the scanning point on the sensor ( , Repeating the above steps yields point cloud data, which is a structured collection where each scan point contains its coordinates in the sensor coordinate system. , The original laser intensity value [z] and its corresponding value provide a data foundation for subsequent feature extraction.

[0040] In conjunction with the first aspect, step S120 includes: S121, for each scan point in the point cloud data, based on the preset dynamic reflection intensity threshold function relationship, determine the reflection intensity threshold corresponding to the measurement distance of the scan point.

[0041] Understandably, the echo intensity (reflection intensity) received by lidar will naturally decrease as the measurement distance increases. If a fixed global intensity threshold is used to distinguish highly reflective markers, they may be missed at long distances due to weak signals, while they may be falsely detected at close distances due to strong environmental stray reflections.

[0042] Based on this, this application employs a dynamic threshold strategy. A dynamic reflection intensity threshold function (e.g., [function name missing]) is established and stored beforehand through calibration experiments. Figure 2 As shown in the figure, the horizontal axis represents distance in meters (m), and the vertical axis represents the reflection intensity threshold (intensity value). For each scan point, based on its measured distance d, a reasonable reflection intensity threshold z applicable to that specific distance can be obtained by querying or calculating this function. This ensures the consistency of the screening criteria across different distances.

[0043] S122, If the actual reflection intensity of the scan point is higher than the reflection intensity threshold, the scan point is determined as a feature point.

[0044] Specifically, the actual measured laser intensity value at each scanning point is compared with a dynamic threshold determined in the previous step that matches its distance. If the actual intensity value at a point is higher than the threshold, it is determined that the point is likely to originate from a high-reflectivity feature marker (such as a reflective sticker) and is marked as a feature point. Otherwise, it is considered a normal environmental point and filtered out.

[0045] S123, combining all feature points, yields the feature point cloud.

[0046] The entire frame of point cloud data is traversed through all scan points, and all feature points determined by S122 are collected. The collection of these feature points constitutes the feature point cloud. This feature point cloud mainly contains reflection points of the target reflective markers, but may still contain a small number of accidental environmental noise points that also meet the high reflectivity conditions (such as reflections from glass or smooth metal surfaces).

[0047] In conjunction with the first aspect, step S130 includes: S131, perform single-label clustering on the feature point cloud to obtain several candidate point cloud clusters.

[0048] S132, perform geometric size filtering on each candidate point cloud cluster to select single-identifier point cloud clusters that meet the preset identifier size.

[0049] S133, calculate the centroid of each single identifier point cloud cluster, and pair two single identifier point cloud clusters with matching centroid distances into a pair of feature identifier point cloud clusters according to the preset identifier spacing and tolerance range.

[0050] The feature point cloud contains both the target point cloud that is actually needed and individual noise points in the environment that meet the threshold conditions. Therefore, the filtered feature point cloud is first subjected to single-label clustering, single-label filtering and double-label pairing based on regular geometry.

[0051] In conjunction with the first aspect, step S131, which involves single-label clustering of the feature point cloud, specifically includes: S1311, traverse each scan point in the feature point cloud, calculate the Euclidean distance between the current scan point and the next scan point, and the angle formed by the sensor origin, the current scan point and the next scan point.

[0052] S1312, if both the Euclidean distance and the angular size are less than the preset threshold, then the next scan point will be assigned to the point cloud cluster where the current scan point is located.

[0053] S1313, otherwise, create a new point cloud cluster starting from the next scan point.

[0054] Because when a lidar scans a continuous, highly reflective planar object (such as a reflective sticker), the returned points are spatially continuous and densely distributed, step S1311 utilizes this spatial local continuity feature to segment the discrete point cloud data into different candidate objects based on geometric proximity. Specifically: Traverse all scanned points in the feature point cloud. For the current point... With the next scan point Calculate the Euclidean distance between two points, and calculate the distance from the origin of the lidar. , The angle formed by the three points.

[0055] If both the Euclidean distance and the subtended angle are less than their respective preset thresholds, then it is determined that... , If they belong to the same reflective marker (i.e., reflective sticker), they should be included in the current point cloud cluster.

[0056] If any condition is not met, then it is considered that , It may belong to different objects or reflective signs, Create a new point cloud cluster starting from the origin.

[0057] After this step, the feature point cloud is initially segmented into several independent candidate point cloud clusters, each of which may correspond to a single reflective marker, environmental noise, or a part of a reflective marker.

[0058] Understandably, while accidental highly reflective points in the environment (such as localized areas of glass or metal) may cluster together, the spatial distribution of these point cloud clusters is random. In contrast, genuine reflective markers possess known, fixed physical dimensions (such as length). Therefore, step S132 utilizes this prior knowledge as a strong constraint to filter out interfering targets that do not conform to the required dimensions. Specifically: S1321, For each candidate point cloud cluster, calculate the straight-line distance between the two farthest endpoints among all scan points of the candidate point cloud cluster, and use it as the apparent length of the candidate point cloud cluster.

[0059] S1322, The apparent length is compared with the actual size of the predetermined SLR mark.

[0060] S1323, if the length is within the allowable error range, then retain the candidate point cloud cluster as a single-identified point cloud cluster.

[0061] S1324, otherwise, filter out the candidate point cloud clusters.

[0062] That is, all candidate clusters are traversed, and the straight-line distance between the two farthest endpoints in each candidate cluster is calculated as the "observation length" of that cluster. This observation length is compared with the pre-stored true reflective sticker size (considering a certain measurement error). Only clusters with lengths within a reasonable range are retained; these clusters are identified as single-identifier point cloud clusters, i.e., reliable point cloud representations of an independent reflective sticker. Through single-identifier clustering in step S131 and single-identifier filtering in step S132, the point clouds corresponding to all single reflective stickers in the environment that meet the predetermined conditions can be segmented into individual clusters.

[0063] Understandably, even after size filtering in step S132, there may still be multiple DSLR signs that meet the criteria in the environment. In this case, step S133 is executed to identify and match the set of dual-reflective signs with a specific, known spacing from the multiple DSLR signs that may exist in the environment.

[0064] In conjunction with the first aspect, step S133 includes: S1331, calculate the Euclidean distance between the centroids of any two single-identifier point cloud clusters.

[0065] First, for each single-identifier point cloud cluster retained after S132 filtering, calculate the geometric center (centroid) of all its points. ; Next, calculate the Euclidean distance between the centroids of any two single-identifier cloud clusters. .

[0066] S1332, determine whether the Euclidean distance falls within the preset spacing matching range, the spacing matching range is determined based on the standard double mark spacing and the preset tolerance coefficient.

[0067] Compare this distance with a preset pairing condition based on a known fixed spacing of the double reflective film. and a tolerance factor (For example, 0.1) is generated, meaning the spacing must meet the following requirements: .

[0068] If so, execute S1333.

[0069] S1333 then pairs the two corresponding single-identifier point cloud clusters into a pair of feature-identifier point cloud clusters.

[0070] Understandably, all clusters that meet the above spacing matching conditions are identified as the final required paired feature identifier point cloud clusters (i.e., the target point cloud clusters corresponding to the double reflectors). Subsequently, in step S140, the paired feature identifier point cloud clusters are used as input, and principal component analysis (PCA) is used to extract the main direction of the point cloud. The position (center point coordinates) and attitude (yaw angle) of the reflector structure in the sensor coordinate system are accurately calculated in combination with specific geometric constraints.

[0071] In conjunction with the first aspect, step S140 includes: S141, For each single-identifier point cloud cluster in the paired feature-identifier point cloud cluster, perform principal component analysis on the two-dimensional coordinates of the single-identifier point cloud cluster to obtain the feature vector representing the main direction of the point cloud distribution.

[0072] First, the validity of the input paired feature-labeled point cloud clusters is verified; if the number of points is less than a preset threshold, the fitting is deemed unsuccessful. Then, to improve robustness, a sliding window smoothing process can be applied to the point clouds. The two-dimensional coordinate matrix of each single-labeled point cloud cluster (ignoring the Z-axis) is then calculated. Calculate its mean As the geometric center of the point cloud cluster, centralized data is obtained. .

[0073] Calculate the covariance matrix using centralized data. And perform singular value decomposition (SVD) on it: For the singular vector matrix from the left Extract the feature vector representing the principal direction (corresponding to the largest eigenvalue) of the point cloud distribution. This vector represents the direction of maximum variance in the point cloud distribution, and the magnitude of the maximum eigenvalue can be considered as a measure of the point cloud's extension along the principal direction. By calculating the direction of maximum variance in the target point cloud, the influence of noise on the fitting results can be relatively accurately ignored, resulting in a stable and accurate estimate of the reflector yaw angle. Simultaneously, the eigenvectors of the secondary direction (corresponding to the second largest eigenvalue) are extracted. .

[0074] S142, determine a fitted straight line based on the feature vector and the center point of the single-identifier point cloud cluster.

[0075] Using the geometric center of the point cloud obtained in step S141 and principal direction eigenvectors Determine a fitted straight line that extends infinitely in a two-dimensional plane, whose equation is of the form:

[0076] This straight line represents the optimal orientation estimate line for the reflective sign in the plane.

[0077] The measurement accuracy of lidar is limited, and the variance of the point cloud representing the direction perpendicular to the reflective marker is basically constant. By adjusting the spacing between two reflective stickers, the dispersion of the point cloud parallel to the reflective marker direction can be effectively increased. Therefore, the characteristic of the PCA-based fitting method is that it transforms the reflective marker direction into the direction of maximum variance of the point cloud.

[0078] S143, select the starting and ending scanning points of the paired feature marker cloud clusters in the scanning sequence.

[0079] Based on the measurement principle of lidar, each scanning point is located on a radial ray originating from the sensor origin (0,0). To accurately fit the endpoints of the reflector point cloud line segments, this application introduces geometric constraints to determine the endpoints.

[0080] Select the starting scan point (first endpoint) of the paired feature marker cloud cluster. and the end scan point (end point) These two points approximately represent the start and end positions of the laser radar scanning beam as it sweeps across the reflective mark.

[0081] S144, construct a first radial constraint line passing through the starting scan point and the sensor origin, and a second radial constraint line passing through the ending scan point and the sensor origin.

[0082] Based on the measurement principle of lidar, each scanning point lies on a radial ray originating from the sensor origin (0,0). Therefore, the constraint lines passing through this starting point and the lidar origin are denoted as follows: (i.e., the first radial constraint line) and the constraint line passing through the end point and the radar origin are denoted as (i.e., the second radial constraint line). The first and second radial constraint lines are constructed based on the starting and ending scanning points of the paired feature marker cloud clusters in the scanning sequence, and respectively represent the first and last radial observation directions of the entire set of double reflective patches observed by the lidar.

[0083] S145, calculate the first intersection point of the fitted line corresponding to the pair of feature marker cloud clusters and the first radial constraint line, and the second intersection point of the fitted line and the second radial constraint line.

[0084] Find the intersection point of the infinitely long straight line fitted by PCA in step S142 and the two radial constraint lines in step S144. and These two intersection points truncate the fitted direction line, and their positions are precisely limited by the radial constraints of the first and last observations of the entire set of double reflective patches corresponding to the paired feature marker point cloud clusters. This allows us to obtain the precise endpoint positions of the paired feature marker point cloud clusters at the start and end times of the lidar scan, which are respectively the front point... With the end point Its accuracy is superior to using the original scan points directly. The coordinates of the two endpoints mentioned above can more realistically reflect the actual distribution range of the entire set of dual reflective dot clouds in space.

[0085] S146, Based on the first intersection point and the second intersection point, determine the coordinates of the center point of the corresponding identifier of the paired feature identifier cloud cluster.

[0086] The front end point calculated in step S145 With the end point The geometric midpoint is determined as the center point coordinate of the paired feature markers (i.e., the entire set of double reflective stickers). :

[0087] The center point coordinates integrate the overall direction information extracted by principal component analysis with the endpoint constraint information provided by global radial observation. Its positioning accuracy is better than that of directly using the geometric center of the original point cloud, and it can effectively suppress the influence of lidar measurement noise and local noise on the position solution.

[0088] S147, based on the center point coordinates and the main direction, determine the attitude yaw angle of the corresponding markers of the paired marker cloud clusters.

[0089] Based on the principal direction feature vector obtained in step S142 (i.e., the direction with the largest variance in the point cloud distribution), calculate its orthogonal vector as the normal vector of the reflector label. To ensure that the normal vector always points to the preset "front" of the reflective sticker, the calculation is performed. And from the sensor origin to the point cloud center ( The vector of ) (or equivalently, directly using The dot product of (). According to the formula: like Then keep Direction; otherwise take .Right now:

[0090] The normal vector after consistency adjustment The orientation angle (e.g., the angle with the X-axis) in the two-dimensional plane is calculated. This angle is the attitude yaw angle corresponding to the pair of feature point cloud clusters, representing the orientation of the reflective marker.

[0091] S148, Based on the attitude yaw angle, determine the intermediate pose of the embodied intelligent robot relative to the paired feature identifiers.

[0092] Due to the spacing between the two signs Known and fixed, and typically rigidly mounted on the same device (with a common orientation), the center point coordinates of the paired feature identifiers determined in step S146 are used. Combined with the yaw angle, the final calculated pose (X, Y, Yaw) of the embodied intelligent robot relative to a unified intermediate pose representing the entire set of dual reflective pads is obtained. This pose, with the sensor coordinate system as a reference, fully expresses the relative position (translation) and relative orientation (yaw angle) between the embodied intelligent robot and the target docking device. It can be directly input into the subsequent pose stabilization module for timing filtering and fusion, or directly used for robot navigation, docking and other control tasks.

[0093] In conjunction with the first aspect, step S150 includes: S151, among the multiple pose recognition results corresponding to the current point cloud frame, select the one closest to the embodied intelligent robot as the valid pose of the current frame.

[0094] S152, acquire multiple intermediate poses obtained from multiple consecutive point cloud frames.

[0095] S153 performs spatiotemporal consistency-based clustering analysis on multiple intermediate poses to eliminate outlier poses caused by transient disturbances.

[0096] S154 performs a fusion calculation on the pose set retained after cluster analysis to obtain the target pose.

[0097] First, multiple sets of dual reflective markers may be pre-arranged in the environment. Steps S130-S140 may identify multiple "paired feature marker point cloud clusters" in the same frame point cloud and calculate multiple corresponding intermediate poses. Therefore, step S141 selects the intermediate pose closest to the spatial position of the embodied intelligent robot (i.e., the sensor origin) from all the recognition results in the current frame as the only valid output of this frame, eliminating the interference of other irrelevant markers.

[0098] Subsequently, the lidar continuously scans at a fixed frequency (e.g., 10Hz). The system accumulates and caches the "current frame effective pose" of each frame selected through step S151 within a preset time window (e.g., the most recent 0.5 seconds, corresponding to 5 frames), forming a pose set in a time series. , ... },in, for Effective pose of time frame for Effective pose of time frame.

[0099] Subsequently, since the true relative poses of the embodied intelligent robot and the target identifier should change continuously and slowly within a short period of time, misidentified poses (outliers) caused by occlusion, point cloud jitter, etc., will significantly differ from the true, continuous pose sequence in space (position and angle). Therefore, a pose difference threshold (including position tolerance and angle tolerance) is set in step S153. Then, a distance-based clustering analysis is performed on the multiple cached intermediate poses. Specifically, the pose space (X, Y, Yaw) is divided into multiple "buckets," and samples with similar poses (difference less than the threshold) are grouped into the same "bucket" (i.e., the same class). Since the true poses are continuous and predominant, they will cluster in the "bucket" with the densest sample density. Isolated poses caused by transient interference that do not conform to the mainstream trend will be separated into other sparse "buckets" or identified as outliers, thus outputting the pose sample set belonging to the most stable and densest cluster (bucket) after removing outliers.

[0100] Finally, in step S154, all pose samples belonging to the most stable cluster (e.g., four stable poses from the past five frames) retained after filtering in S153 are fused. This fusion operation is equivalent to a low-pass filter, which not only smooths the measurement noise but also fundamentally resists serious misidentifications that may occur in a single frame by utilizing information from multiple frames, thereby outputting a highly reliable target pose after stabilization and jitter reduction.

[0101] The typical algorithm used for fusion is to calculate the arithmetic mean of these poses in the X coordinate, Y coordinate and yaw angle as the final output target pose. As another feasible method, a weighted average method can be used, for example, giving higher weights to more recent frames. This can be selected according to the actual application requirements. This is just an example and is not limited.

[0102] Combination Figure 3As shown, in practical applications, the embodied intelligent robot first scans the surrounding environment using its onboard LiDAR to obtain point cloud data (i.e., raw point cloud data) of the current frame, which includes information on distance, angle, and reflection intensity. This data is then converted to the sensor coordinate system to provide basic data for subsequent processing.

[0103] The process then proceeds to point cloud recognition and clustering. Specifically, firstly, the original point cloud is filtered according to a preset dynamic reflection intensity threshold function to extract all high-reflectivity feature point clouds. Secondly, the clustering results are compared and matched based on prior feature information (width of a single reflective marker and spacing between two reflective markers). Specifically, the feature point clouds are clustered according to geometric distance and angular resolution, and point cloud clusters that conform to the physical size of a single reflective marker are separated and retained. By calculating the centroid of each single marker cluster and performing pairing verification according to the preset double-tag spacing, the paired feature marker point cloud clusters corresponding to the target double reflective marker are uniquely identified from multiple possible reflectors.

[0104] Subsequently, precise pose calculation is performed on each individual cluster within the paired feature marker point cloud clusters. Specifically: First, Principal Component Analysis (PCA) is used to calculate the center of the feature marker point cloud cluster and the orientation of the reflective markers using the first principal component as the direction of the dual reflective markers and the second principal component as the direction of the recognition position, fitting a straight line. Next, a sensor radial constraint is introduced, and a constraint line is constructed using the start and end scan points of the point cloud. This constraint line is then intersected with the PCA fitted line to accurately calculate the coordinates of the endpoints and center point of the reflective markers. Finally, the yaw angle is calculated while ensuring the consistency of the normal vector direction. By combining the results for a pair of markers, the intermediate pose of the robot relative to the marker group can be obtained, ensuring high accuracy and strong robustness in pose calculation.

[0105] To improve the system's stability in dynamic environments, post-processing is performed on the poses in the time series. First, when multiple recognition results exist in a single frame, the closest result is selected as the valid value for that frame. Then, valid intermediate poses from multiple consecutive frames are accumulated, and outliers caused by momentary occlusion or noise are removed using spatiotemporal consistency-based clustering analysis (such as bucket sort). Finally, the selected stable pose set is fused (e.g., weighted averaging) to output a final smooth and reliable target pose, which can be directly used by the robot's localization and navigation system and can be packaged as a visual message output for debugging.

[0106] Secondly, embodiments of this application provide a pose recognition device for an embodied intelligent robot, combined with Figure 4 As shown, the device includes: an acquisition module 10, an extraction module 20, an identification module 30, a calculation module 40, and a fusion module 50.

[0107] The acquisition module 10 is used to acquire point cloud data of the environment in which the embodied intelligent robot is located.

[0108] The extraction module 20 is used to extract feature point clouds with preset reaction characteristics from the point cloud data.

[0109] The identification module 30 is used to perform spatial clustering and pairing verification on the feature point cloud, and identify paired feature identifier point cloud clusters that conform to the preset spatial layout.

[0110] The calculation module 40 is used to calculate the intermediate pose of the embodied intelligent robot relative to the pair of feature marker cloud clusters based on the geometric distribution characteristics of the pair of feature marker cloud clusters and sensor observation constraints.

[0111] The fusion module 50 is used to perform temporal filtering and fusion based on the intermediate poses of multiple consecutive frames, and output the target pose.

[0112] Thirdly, embodiments of this application provide an electronic device, combined with Figure 5 As shown, the electronic device includes a memory 131 and a processor 130. The memory 131 stores a computer program, and the processor 130 runs the computer program to make the electronic device perform the above-described method.

[0113] Furthermore, combined Figure 5 The electronic device shown also includes a bus 132 and a communication interface 133, with the processor 130, the communication interface 133 and the memory 131 connected via the bus 132.

[0114] The memory 131 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 133 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 132 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0115] Processor 130 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 130 or by instructions in software form. Processor 130 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 131, and processor 130 reads the information in memory 131 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0116] Fourthly, embodiments of this application provide a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the above-described method.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0118] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0119] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0120] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0121] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for pose recognition of an embodied intelligent robot, characterized in that, The method includes: Acquire point cloud data of the environment in which the embodied intelligent robot is located; Extract feature point clouds with preset reaction characteristics from the point cloud data; Spatial clustering and pairing verification are performed on the feature point cloud to identify paired feature identifier point cloud clusters that conform to a preset spatial layout; Based on the geometric distribution characteristics of the paired feature marker cloud clusters and sensor observation constraints, the intermediate pose of the embodied intelligent robot relative to the paired feature marker cloud clusters is calculated. The intermediate poses of multiple consecutive frames are used for temporal filtering and fusion to output the target pose.

2. The method according to claim 1, characterized in that, The steps of extracting feature point clouds with preset reaction characteristics from the point cloud data include: For each scan point in the point cloud data, the reflection intensity threshold corresponding to the measurement distance of the scan point is determined based on a preset dynamic reflection intensity threshold function relationship; If the actual reflection intensity of the scanned point is higher than the reflection intensity threshold, the scanned point is determined to be a feature point; By combining all the aforementioned feature points, the feature point cloud is obtained.

3. The method according to claim 1, characterized in that, The steps of performing spatial clustering and pairing verification on the feature point cloud to identify paired feature identifier point cloud clusters that conform to a preset spatial layout include: The feature point cloud is clustered using a single identifier to obtain several candidate point cloud clusters; Geometric size filtering is performed on each candidate point cloud cluster to select single-identifier point cloud clusters that meet the preset identifier size; Calculate the centroid of each of the single identifier point cloud clusters, and pair two single identifier point cloud clusters with matching centroid distances into the paired feature identifier point cloud clusters according to the preset identifier spacing and tolerance range.

4. The method according to claim 3, characterized in that, The steps of performing single-identifier clustering on the feature point cloud include: Traverse each scan point in the feature point cloud, calculate the Euclidean distance between the current scan point and the next scan point, and the angle formed by the sensor origin, the current scan point, and the next scan point; If both the Euclidean distance and the angular subtraction are less than a preset threshold, then the next scan point is assigned to the point cloud cluster where the current scan point is located; Otherwise, create a new point cloud cluster starting from the next scan point.

5. The method according to claim 3, characterized in that, The step of pairing two single-marker point cloud clusters with matching centroid distances into the paired feature marker point cloud clusters according to the preset marker spacing and tolerance range includes: Calculate the Euclidean distance between the centroids of any two of the single-identifier point cloud clusters; Determine whether the Euclidean distance falls within a preset spacing matching range, the spacing matching range being determined based on the standard double-marker spacing and a preset tolerance coefficient; If so, the two corresponding single-identifier point cloud clusters are paired as the paired feature identifier point cloud clusters.

6. The method according to claim 1, characterized in that, The step of calculating the intermediate pose of the embodied intelligent robot relative to the paired feature marker point cloud clusters, based on the geometric distribution characteristics of the paired feature marker point cloud clusters and sensor observation constraints, includes: For each single-identifier point cloud cluster in the paired feature identifier point cloud clusters, principal component analysis is performed on the two-dimensional coordinates of the single-identifier point cloud cluster to obtain the feature vector characterizing the main direction of the point cloud distribution. A fitting straight line is determined based on the feature vector and the center point of the single-identifier point cloud cluster; Select the start and end scanning points of the paired feature identifier point cloud clusters in the scanning sequence; Construct a first radial constraint line passing through the starting scan point and the sensor origin, and a second radial constraint line passing through the ending scan point and the sensor origin, respectively. Calculate the first intersection point of the fitted line corresponding to the paired feature marker cloud cluster and the first radial constraint line, and the second intersection point of the fitted line and the second radial constraint line; Based on the first intersection point and the second intersection point, determine the coordinates of the center point of the corresponding identifier of the paired feature identifier cloud cluster; Based on the center point coordinates and the main direction, determine the attitude yaw angle of the paired feature marker cloud cluster; Based on the attitude yaw angle, the intermediate pose of the embodied intelligent robot relative to the paired feature markers is determined.

7. The method according to claim 1, characterized in that, The steps of performing temporal filtering and fusion based on the intermediate poses of multiple consecutive frames to output the target pose include: Among the multiple pose recognition results corresponding to the current point cloud frame, the one closest to the embodied intelligent robot is selected as the valid pose of the current frame. Multiple valid poses are obtained by acquiring multiple consecutive point cloud frames; Cluster analysis based on spatiotemporal consistency is performed on the multiple valid poses to eliminate outlier poses caused by transient disturbances; The target pose is obtained by performing a fusion calculation on the pose set retained after cluster analysis.

8. A pose recognition device for an embodied intelligent robot, characterized in that, The device includes: The acquisition module is used to acquire point cloud data of the environment in which the embodied intelligent robot is located; The extraction module is used to extract feature point clouds with preset reaction characteristics from the point cloud data; The identification module is used to perform spatial clustering and pairing verification on the feature point cloud, and identify paired feature identifier point cloud clusters that conform to the preset spatial layout; The calculation module is used to calculate the intermediate pose of the embodied intelligent robot relative to the pair of feature marker cloud clusters based on the geometric distribution characteristics of the pair of feature marker cloud clusters and sensor observation constraints. The fusion module is used to perform temporal filtering and fusion based on the intermediate poses of multiple consecutive frames, and output the target pose.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program and the processor running the computer program to cause the electronic device to perform the method of any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores computer program instructions, which, when read and executed by a processor, perform the method described in any one of claims 1 to 7.

Images