An Autonomous Navigation Method and System for Elderly Care Robots Based on Depth Vision

Abstract

This application provides an autonomous navigation method and system for elderly care robots based on depth vision. Specifically, this application utilizes a 3D vision sensor to acquire environmental point cloud data. By analyzing changes in ambient light intensity and incident angle, it dynamically switches the point cloud fusion mode to suppress light interference, resulting in an optimized point cloud. Subsequently, semantic analysis is performed to generate a 3D obstacle map with category labels. Based on this map, an object relationship topology graph is constructed to identify passable gaps between objects and determine whether there are any anomalies in the spatial relationships of objects that block the passage of high-priority objects. Finally, by integrating the 3D obstacle map, object relationship topology graph, passable gap information, semantic label priority, and anomaly judgment results, a navigation path for the elderly care robot is generated. This application constructs a complex environmental relationship model and intelligently identifies safe passages and abnormal object placement, generating a highly adaptable and reliable navigation path for the elderly care robot.

Description

Technical Field

[0001] This application relates to the field of intelligent robot navigation technology, and in particular to an autonomous navigation method and system for elderly care robots based on depth vision. Background Technology

[0002] With their enormous potential to assist seniors in their daily lives and provide care services, elderly care robots are gradually becoming a key technology for addressing the challenges of an aging population. Their autonomous navigation capabilities are central to achieving these functions, ensuring the robots move safely and efficiently in complex and ever-changing home environments, thereby significantly improving the quality of life and independence of seniors.

[0003] However, existing autonomous navigation methods for elderly care robots typically rely on traditional two-dimensional LiDAR or basic depth cameras for environmental perception and the construction of simple obstacle maps for path planning. While these methods can meet basic obstacle avoidance requirements, they often fall short when dealing with complex environmental information and struggle to provide a refined understanding of the environment.

[0004] Therefore, these existing methods exhibit numerous limitations in practical applications. For instance, their perception systems are susceptible to changes in ambient lighting, leading to inaccurate obstacle recognition or unstable map construction. Furthermore, due to a lack of semantic understanding and spatial relationship analysis capabilities for various objects in the environment, robots may be unable to distinguish between obstacles of varying importance, resulting in low navigation efficiency or potential safety hazards. Thus, existing technologies suffer from insufficient robustness in environmental perception and inadequate semantic understanding capabilities. Summary of the Invention

[0005] This application provides an autonomous navigation method and system for elderly care robots based on depth vision, which solves the problems of point cloud data distortion caused by ambient light interference and channel priority misjudgment and abnormal spatial relationship omission caused by lack of semantic understanding in the prior art.

[0006] Firstly, this application provides an autonomous navigation method for elderly care robots based on depth vision, including:

[0007] The system uses a 3D vision sensor to collect point cloud data of the environment, and based on the point cloud data, determines the intensity value of ambient light and the change in incident angle. According to the intensity value and the change in incident angle, the system switches the point cloud fusion mode to suppress ambient light interference and obtain optimized point cloud data.

[0008] Semantic analysis is performed on the optimized point cloud data to generate a 3D obstacle map with semantic labels of various categories;

[0009] Based on the spatial distribution of objects in the three-dimensional obstacle map, an object relationship topology map is constructed, and passable gaps between related objects are identified in the object relationship topology map.

[0010] In the object relationship topology diagram, the spatial relationship of items is determined. When it is determined that there is a situation where the spatial relationship of items blocks the passage of high-priority items, it is determined that there is an abnormal spatial relationship of items.

[0011] Based on the three-dimensional obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, combined with the determination results of the spatial relationships of abnormal items, a navigation path for the elderly care robot is generated.

[0012] Optionally, based on the 3D obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, combined with the determination results of the spatial relationships of abnormal items, a navigation path for the elderly care robot is generated, including:

[0013] When an abnormal spatial relationship is determined, the obstacle risk weight factor in the safety factor calculation is dynamically adjusted according to the semantic tags and priority weights involved in the abnormal spatial relationship.

[0014] In the object relationship topology map, identify alternative passable gaps near high-priority object channels blocked by the abnormal object spatial relationships;

[0015] Based on the adjusted obstacle risk weighting factor and the corresponding alternative passable gaps, the initial calculated path is revised or replanned to generate a navigation path that avoids abnormal congestion areas and meets a preset safety threshold; or,

[0016] When the spatial relationship of the abnormal items is not determined, a navigation path is generated based on the three-dimensional obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags.

[0017] Optionally, when the existence of the abnormal object spatial relationship is not determined, a navigation path is generated based on the three-dimensional obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, including:

[0018] When the existence of the abnormal object spatial relationship is not determined, the three-dimensional coordinates of the current position and target position of the elderly care robot are determined in the three-dimensional obstacle map, and all marked passable gaps in the object relationship topology map are extracted to form a gap set.

[0019] For each gap in the gap set, the extended boundaries of the object surfaces on both sides of the gap are connected to generate a virtual channel space;

[0020] Discrete sampling points are set within the virtual channel space. The minimum distance from each sampling point to the nearest obstacle surface is calculated. Based on the minimum distance and the semantic label priority weights of the objects on both sides of the corresponding gap, the channel safety index is calculated.

[0021] Establish a path search space between the current location and the target location, consisting of all connected virtual channel spaces;

[0022] In the path search space, a connection path is selected that optimizes the accumulated channel safety index and satisfies a preset minimum safety threshold. The coordinate point sequence of the connection path is then transformed into a navigation path for the elderly care robot.

[0023] Optionally, based on the point cloud data, the intensity value of ambient light and the change in incident angle are determined, and the point cloud fusion mode is switched according to the intensity value and the change in incident angle to suppress ambient light interference, thereby obtaining optimized point cloud data, including:

[0024] The intensity value of ambient light is obtained by calculating the average brightness of the color components of all points in the point cloud data.

[0025] Based on the geometric distribution and brightness variation of the point cloud data, the change in the incident angle of ambient light is determined by identifying the direction with the largest brightness gradient in the point cloud and calculating the change in this direction between consecutive frames.

[0026] When the intensity value changes or the incident angle changes by more than a preset threshold, the point cloud fusion mode is switched. The point cloud fusion mode includes a first fusion mode and a second fusion mode. The first fusion mode is enabled when the intensity value of the ambient light is higher than the threshold, and the second fusion mode is enabled when the intensity value of the ambient light is lower than the threshold.

[0027] The point cloud data is processed according to the switched point cloud fusion mode. In the first fusion mode, the filtering weight of high brightness points is increased during the point cloud fusion process. In the second fusion mode, the degree of ignoring low brightness points during the point cloud fusion process is reduced to suppress ambient light interference and output optimized point cloud data.

[0028] Optionally, semantic analysis is performed on the optimized point cloud data to generate a 3D obstacle map with semantic labels of various categories, including:

[0029] A three-dimensional mesh structure is constructed, which contains multiple units, and the multiple units can cover the spatial range described by the optimized point cloud data;

[0030] Extract the geometric features and surface texture features of the point cloud data within each unit. The geometric features include the spatial density distribution and curvature variation of the point cloud data within the unit, and the surface texture features include the color pattern distribution of the point cloud data within the unit.

[0031] The geometric features and surface texture features are input into a pre-trained semantic recognition model, and the semantic recognition model outputs the object category identifier corresponding to each unit. The object category identifier includes ground, wall, furniture, medical equipment, and movable objects.

[0032] Regions with the same object category identifier in adjacent units are merged to form continuous object blocks, and each continuous object block is labeled with a unified semantic tag.

[0033] In the three-dimensional mesh structure, object blocks with a height higher than a preset passage height and object blocks of the impassable category are marked as obstacles.

[0034] Integrate all units with semantic labels and obstacle markers to generate a 3D obstacle map that includes spatial location, object category identifiers, and obstacle markers.

[0035] Optionally, based on the spatial distribution of objects in the three-dimensional obstacle map, an object relationship topology map is constructed, and passable gaps between related objects are identified in the object relationship topology map, including:

[0036] Extract the spatial location and boundary extent of all semantically labeled continuous object blocks in a 3D obstacle map;

[0037] The minimum distance between any two object blocks is calculated based on the spatial location and boundary range. When the minimum distance is less than the preset proximity judgment distance, it is determined that the two object blocks have a spatial relationship.

[0038] Construct an object relationship topology graph, treat each object block as a node, establish a connection line between two nodes that have the spatial relationship, and record the corresponding minimum interval distance value between the corresponding object blocks on the connection line;

[0039] Identify the passable gaps between related objects in the object relationship topology graph, traverse all node pairs with connecting lines, and perform a passability judgment based on the minimum interval distance value recorded by the connecting lines;

[0040] When the minimum interval distance value is greater than the passage width threshold of the elderly care robot body, it is marked as a passable gap. When the minimum interval distance value is less than or equal to the passage width threshold and the associated object block contains a movable object category, dynamic passage feasibility marking is performed, and an object relationship topology map with passable gap markings is output.

[0041] Optionally, in the object relationship topology graph, the spatial relationships of items are determined. When it is determined that there is a situation where the spatial relationships of items block the passage of high-priority items, it is determined that there are abnormal spatial relationships of items, including:

[0042] Obtain the semantic label priority weights in the object relationship topology map with passable gap markers and the 3D obstacle map, wherein the priority weights are pre-associated with the semantic labels;

[0043] Locate the continuous object block with the highest priority weight in the object relationship topology graph, and extract all passable gaps pointing to the corresponding block as necessary passage routes.

[0044] Traverse all connecting lines on necessary passage routes. When the minimum interval distance recorded by the connecting line is less than the passage width threshold of the elderly care robot and no dynamic passage feasibility marker is set, the corresponding gap is determined to be blocked.

[0045] When any necessary route is detected to be blocked, the system outputs a determination result indicating the presence of abnormal spatial relationships between items.

[0046] Secondly, this application provides an autonomous navigation system for elderly care robots based on depth vision, including:

[0047] The acquisition module is used to acquire point cloud data of the environment using a three-dimensional vision sensor, and based on the point cloud data, determine the intensity value of ambient light and the change in incident angle, and switch the point cloud fusion mode according to the intensity value and the change in incident angle to suppress ambient light interference and obtain optimized point cloud data.

[0048] The generation module is used to perform semantic analysis on the optimized point cloud data and generate a 3D obstacle map with semantic labels of various categories.

[0049] The identification module is used to construct an object relationship topology map based on the spatial distribution of objects in the three-dimensional obstacle map, and to identify passable gaps between related objects in the object relationship topology map;

[0050] The judgment module is used to judge the spatial relationship of items in the object relationship topology diagram. When it is judged that there is a situation where the spatial relationship of items blocks the channel of high-priority items, it is determined that there is an abnormal spatial relationship of items.

[0051] The generation module is further configured to generate a navigation path for the elderly care robot based on the three-dimensional obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, combined with the determination results of the spatial relationships of abnormal items.

[0052] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement an autonomous navigation method for an elderly care robot based on depth vision as described in the first aspect above.

[0053] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an autonomous navigation method for an elderly care robot based on depth vision as described in the first aspect.

[0054] This application effectively suppresses depth data noise caused by strong light interference and improves the robustness of point cloud data by dynamically analyzing changes in ambient light intensity and incident angle to switch point cloud fusion modes. It generates a 3D obstacle map with category labels through semantic analysis, enabling the identification of high-priority objects such as medical equipment and emergency access routes. By constructing an object relationship topology map and identifying passable gaps, it can quantitatively analyze the spatial relationships between objects and safe passage areas. By identifying abnormal spatial relationships that block passageways for high-priority items, it can proactively detect risks specific to elderly care scenarios, such as wheelchairs blocking emergency access routes. Finally, it generates paths based on semantic priority, passable gaps, and anomaly detection results, achieving navigation decisions that balance safety standards and scenario adaptability.

[0055] Furthermore, when abnormal spatial relationships exist between items, the obstacle risk weight factor is dynamically adjusted to identify alternative passage gaps near blocked channels and correct the path to avoid abnormal areas; when there are no abnormalities, a path is directly generated. This design significantly improves the adaptability of the elderly care robot to sudden risks. By dynamically adjusting the safety factor through semantic weights, it ensures that the obstacle avoidance distance is expanded in high-risk areas; by quickly responding to channel blockage events through alternative gaps, it avoids path interruption; and finally, the generated path ensures passage efficiency while meeting the safety threshold requirements of the elderly care environment, systematically reducing the risk of collisions or delays caused by temporary placement of items.

[0056] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1A flowchart of an autonomous navigation method for elderly care robots based on depth vision, provided in this application, is shown.

[0059] Figure 2 A schematic diagram of the structure of an autonomous navigation system for elderly care robots based on depth vision provided in this application is shown.

[0060] Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0061] Given that existing autonomous navigation methods for elderly care robots suffer from insufficient perceptual robustness and a lack of semantic understanding in complex environments, primarily manifested in the interference of ambient lighting changes on 3D visual perception, leading to decreased accuracy in obstacle recognition and map construction. Furthermore, traditional methods often fail to perform deep semantic analysis of various objects in the environment, making it difficult to identify spatial relationships between objects and their impact on the travel path. This limits the robot's ability to make intelligent decisions in dynamic, unstructured environments, potentially resulting in low navigation efficiency and even safety hazards.

[0062] To overcome the aforementioned technical deficiencies, this application proposes an innovative autonomous navigation method for elderly care robots. This method, by introducing an advanced 3D vision sensor, can intelligently perceive and suppress ambient light interference, thereby acquiring high-quality optimized point cloud data and significantly improving the accuracy and stability of environmental perception. Based on this, the method further performs semantic analysis on the optimized point cloud data, constructing a 3D obstacle map containing various semantic tags, and building an object relationship topology map based on the spatial distribution of objects. This allows for the identification of passable gaps and the determination of any abnormal spatial relationships that might obstruct the passage of high-priority objects. By comprehensively considering this information, the robot can generate a more intelligent, safe, and efficient navigation path, effectively solving the problems of easily interfered environmental perception, lack of semantic understanding, and imprecise navigation decisions in existing technologies.

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] Figure 1 A flowchart of an autonomous navigation method for elderly care robots based on depth vision is provided in this application embodiment, as shown below. Figure 1 As shown, the method includes:

[0065] 101. Use a 3D vision sensor to collect point cloud data of the environment, and based on the point cloud data, determine the intensity value and incident angle change of the ambient light, and switch the point cloud fusion mode according to the intensity value and incident angle change to suppress ambient light interference and obtain optimized point cloud data.

[0066] Optionally, step 101, "based on point cloud data, determining the intensity value and incident angle change of ambient light, and switching the point cloud fusion mode according to the intensity value and incident angle change to suppress ambient light interference and obtain optimized point cloud data," may specifically include the following steps:

[0067] 1011. Calculate the average brightness of the color components of all points in the point cloud data to obtain the intensity value of ambient light.

[0068] 1012. Based on the geometric distribution and brightness variation of point cloud data, the change in the incident angle of ambient light is determined by identifying the direction with the largest brightness gradient in the point cloud and calculating the change in this direction between consecutive frames.

[0069] 1013. When the intensity value changes or the incident angle changes by more than a preset change threshold, the point cloud fusion mode is switched. The point cloud fusion mode includes a first fusion mode and a second fusion mode. The first fusion mode is enabled when the intensity value of the ambient light is higher than the threshold, and the second fusion mode is enabled when the intensity value of the ambient light is lower than the threshold.

[0070] 1014. Process point cloud data according to the switched point cloud fusion mode. In the first fusion mode, increase the filtering weight of high brightness points during the point cloud fusion process. In the second fusion mode, reduce the degree of ignoring low brightness points during the point cloud fusion process to suppress ambient light interference and output optimized point cloud data.

[0071] Point cloud data refers to a set of discrete three-dimensional points collected by a three-dimensional vision sensor, representing the geometric shape and spatial distribution of an object's surface. Each point typically contains three-dimensional coordinate information, and sometimes also color or intensity information.

[0072] In this embodiment, the color information of all points in the point cloud data is first analyzed to calculate their average brightness value, thereby quantifying the overall intensity of the current ambient light. Further analysis is performed on the spatial arrangement (geometric distribution) and brightness variation patterns of the point cloud. It identifies the direction of the fastest brightness change in the point cloud and tracks the amount of change in this direction between continuously acquired frame data, thus determining the variation in the ambient light incidence angle.

[0073] Secondly, the system continuously monitors changes in ambient light intensity and incident angle. When these changes exceed the preset allowable range, the system automatically switches the point cloud fusion processing mode according to the current light intensity. For example, it activates the "first fusion mode" in environments with strong light and the "second fusion mode" in environments with weak light.

[0074] The system processes the point cloud data according to the currently active fusion mode. In the "first fusion mode," the system increases the filtering intensity for excessively bright points to reduce interference from strong light. In the "second fusion mode," the system reduces the degree of neglect of low-brightness points to avoid missing information in dark areas.

[0075] Through this adaptive processing, the system can effectively suppress the impact of ambient light on the quality of point cloud data, thereby outputting more accurate and reliable optimized point cloud data.

[0076] 102. Perform semantic analysis on the optimized point cloud data to generate a 3D obstacle map with semantic labels of various categories.

[0077] Optionally, step 102 may specifically include the following steps:

[0078] 1021. Construct a three-dimensional mesh structure. The three-dimensional mesh structure contains multiple units, and the multiple units can cover the spatial range described by the optimized point cloud data.

[0079] Among them, the three-dimensional mesh structure is a method of discretizing three-dimensional space. It divides a continuous three-dimensional space into a series of regular or irregular small regions, each region being called a cell. This structure helps in the organization and management of spatial data.

[0080] In this embodiment, the system first constructs a three-dimensional mesh structure covering the spatial range of the optimized point cloud data. This mesh structure divides the entire three-dimensional space into many small, regular cubic units (e.g., voxels). Each unit represents a specific region in the space and can accommodate a portion of the point cloud data. This partitioning method makes the processing of large-scale point cloud data more efficient and orderly.

[0081] 1022. Extract the geometric features and surface texture features of the point cloud data within each cell. The geometric features include the spatial density distribution and curvature variation of the point cloud data within the cell, and the surface texture features include the color pattern distribution of the point cloud data within the cell.

[0082] Among them, geometric features describe the spatial shape and distribution characteristics of point cloud data, such as the density of points (spatial density distribution) and the degree of surface curvature (curvature variation). Surface texture features describe the variation patterns of point cloud data in color or brightness, such as the uniformity or diversity of color distribution.

[0083] In this embodiment, for each cell in the constructed 3D mesh structure, the system analyzes the point cloud data it contains, extracting its unique geometric and surface texture features. For example, regarding geometric features, the system calculates the spatial density of the point cloud within the cell, i.e., the number of points per unit volume, and the curvature variation of the point cloud surface, which reflects the flatness or curvature of the surface. Regarding surface texture features, the system analyzes the color information of the point cloud within the cell, such as calculating a color histogram or average color value to describe its color pattern distribution. These features are key information for distinguishing different object categories.

[0084] 1023. Input geometric features and surface texture features into a pre-trained semantic recognition model, and output the object category identifier corresponding to each unit through the semantic recognition model.

[0085] The object category identifiers include ground, walls, furniture, medical equipment, and movable objects.

[0086] Among them, the semantic recognition model is a trained artificial intelligence model that can learn and recognize the feature patterns of different objects in point cloud data, thereby classifying the point cloud data into predefined object categories.

[0087] Specifically, the training process of semantic recognition models typically involves a large amount of labeled data. First, researchers collect point cloud data containing various real-world scenes and manually label each point or local region in these point cloud data, explicitly indicating its object category (e.g., ground, wall, furniture, medical equipment, etc.). This precisely labeled dataset is called the training set. Subsequently, these training sets are fed into a deep learning model. Through repeated learning and optimization of its internal parameters, the model gradually establishes a complex mapping relationship between input geometric features (such as spatial density and curvature variations) and surface texture features (such as color pattern distribution) and output object category identifiers. This process can be understood as the model "learning" the unique shapes, structures, and appearance patterns of different objects. Once the model is sufficiently trained and reaches the expected recognition accuracy, it can take the geometric and surface texture features of new, unseen point cloud units as input and accurately predict the object category identifier corresponding to that unit based on the patterns learned during training.

[0088] Object category labels are the classification results of the semantic recognition model for the object represented by each unit, such as "ground", "wall", "furniture", etc.

[0089] In this embodiment, the extracted geometric features and surface texture features are used as inputs to a pre-trained semantic recognition model. This model is typically a deep learning model, such as a convolutional neural network (CNN) or a graph neural network (GNN), which is trained on a large amount of real-world point cloud data with annotations to learn how to map specific feature patterns to specific object categories. During training, the model learns the unique geometric shapes, surface textures, and color patterns of different objects (such as the ground, walls, chairs, beds, wheelchairs, etc.). When new features are input, the model outputs the most likely object category identifier for each unit. For example, if a unit's point cloud data exhibits flat and broad geometric features and a uniform color distribution, the model may identify it as "ground"; if the point cloud data exhibits vertical and regular geometric features, it may be identified as "wall".

[0090] 1024. Merge areas with the same object category identifier in adjacent units to form continuous object blocks, and label each continuous object block with a unified semantic label.

[0091] A contiguous object block is a region representing a complete object, composed of adjacent units with the same object category identifier. Semantic tags are descriptive names given to these contiguous object blocks to identify their category and function.

[0092] In this embodiment, after each unit is assigned an object category identifier, the system checks adjacent units. If these adjacent units have the same object category identifier, they are merged together to form a larger, continuous object block. For example, if multiple adjacent "ground" units are identified, they will be merged into a large "ground" block. Subsequently, the system assigns a unified semantic label to each formed continuous object block. This label not only includes the object category but can also include more specific instance information, such as "living room floor" or "bedroom wall," to facilitate subsequent identification and management.

[0093] 1025. In a 3D mesh structure, mark object blocks whose height is higher than the preset passage height and object blocks of the impassable category as obstacles.

[0094] In this embodiment, the system further analyzes these semantically tagged object blocks. First, it determines which object blocks might obstruct the robot's passage based on a preset robot access height threshold (e.g., the robot's height plus a certain safety margin). Any object block whose height exceeds this threshold, regardless of its semantic category, will be marked as an obstacle. Second, for certain specific categories of impassable objects (e.g., walls, large fixed furniture), even if their height is below the access height threshold, they will be directly marked as obstacles due to their inherent impassability. For example, an object block identified as a "wall" will be marked as an obstacle as a whole, even if part of its area is below the robot's access height.

[0095] 1026. Integrate all units with semantic labels and obstacle markers to generate a 3D obstacle map that includes spatial location, object category identifiers, and obstacle markers.

[0096] A 3D obstacle map is a map in 3D space that uses a grid structure to mark the location, shape, and semantic category information of all obstacles.

[0097] In this embodiment, after the above processing, the system integrates all continuous object blocks with semantic tags and unit information marked as obstacles. Finally, a comprehensive 3D obstacle map is generated. This map not only contains the precise 3D location of each spatial unit, but also clearly indicates the object category to which the unit belongs (such as ground, furniture, etc.), and whether it is marked as an obstacle.

[0098] This application enables the generation of a 3D obstacle map containing rich semantic information by performing refined semantic analysis on optimized point cloud data, which significantly improves the robot's ability to understand the environment.

[0099] 103. Based on the spatial distribution of objects in the 3D obstacle map, construct an object relationship topology map and identify passable gaps between related objects in the object relationship topology map.

[0100] The object relationship topology graph is an abstract graph structure that treats each object block in the environment as a "node," and the spatial relationships between object blocks (such as proximity) are represented as "connecting lines" or "edges" that connect these nodes. The connecting lines in the graph also record information such as the distance between objects. A passable gap refers to the space between two objects in the environment that is wide enough for the elderly care robot to pass safely.

[0101] Optionally, step 103 may specifically include the following steps:

[0102] 1031. Extract the spatial location and boundary range of all continuous object blocks with semantic labels in the 3D obstacle map.

[0103] In this embodiment, the system first traverses the previously generated 3D obstacle map, identifying all continuous object blocks with semantic labels. For each identified object block, the system accurately extracts its spatial location in 3D space (e.g., its geometric center coordinates) and its boundary extent (e.g., a minimum bounding cube, i.e., the minimum and maximum coordinates of the bounding box). This information forms the basis for subsequent calculations of distances between objects.

[0104] 1032. Calculate the minimum distance between any two object blocks based on spatial location and boundary range. When the minimum distance is less than the preset proximity judgment distance, it is determined that the two object blocks have a spatial relationship.

[0105] In this embodiment, the system pairs all extracted object blocks and calculates the minimum distance between each pair. This distance can be obtained by calculating the shortest Euclidean distance between the bounding boxes of two object blocks or between their point cloud data. For example, for two object blocks A and B, the minimum distance d(A,B) can be calculated as follows (taking bounding boxes as an example): d(A,B)=min(dist(p_A,p_B)) where p_A is any point on the bounding box of object block A, and p_B is any point on the bounding box of object block B.

[0106] After calculating the minimum interval distance, the system compares it with the preset proximity judgment distance. If d(A,B) < proximity judgment distance, it is determined that the two object blocks have a spatial relationship, meaning that they are close enough to form a gap or influence each other.

[0107] The proximity judgment distance is a preset threshold used to determine whether two object blocks are close enough to be considered spatially related.

[0108] 1033. Construct an object relationship topology graph, treat each object block as a node, establish a connection line between two nodes that have a spatial relationship, and record the corresponding minimum interval distance value between the corresponding object blocks on the connection line.

[0109] In this embodiment, the system constructs an object relationship topology graph based on the results of 1032 steps. In this graph, each independent object block is abstracted as a node. If two object blocks are determined to have a spatial relationship, a connection line (or "edge") is established between their corresponding nodes. This connection line also has an attribute, which records the minimum distance value between the two object blocks calculated in 1032 steps. In this way, the objects in the entire environment and their mutual proximity relationships are clearly represented in the form of a graph.

[0110] 1034. Identify the passable gaps between related objects in the topology graph. Traverse all node pairs with connecting lines and perform passability judgment based on the minimum interval distance value recorded by the connecting lines.

[0111] In this embodiment, to identify potential passable gaps, the system traverses all node pairs with connecting lines in the object relationship topology graph. For each connecting line, the system reads the minimum distance value between two related object blocks recorded on it. This distance value serves as a key basis for determining whether the robot can pass through the gap, thereby performing a preliminary passability assessment.

[0112] 1035. When the minimum interval distance value is greater than the passage width threshold of the elderly care robot body, it is marked as a passable gap. When the minimum interval distance value is less than or equal to the passage width threshold and the associated object block contains the category of movable object, dynamic passage feasibility marking is performed, and the object relationship topology map with passable gap marking is output.

[0113] The passage width threshold is the minimum width required for the elderly care robot to safely pass through a gap. It is usually slightly larger than the robot's physical width to allow for a safety margin. A dynamic passage feasibility marker is a special identifier used to indicate that although the current width of a gap is insufficient for passage, a passable path can theoretically be created by moving movable objects on either side.

[0114] The design aims to further explore the spatial relationships between objects in the environment from a static 3D obstacle map, especially to identify the paths that the robot may take and to assess the feasibility of these paths, thereby providing more intelligent and flexible information for subsequent path planning.

[0115] In this embodiment of the application, the system will ultimately determine the type of gap based on the minimum interval distance value obtained in step 1034 and the preset passage width threshold.

[0116] Directly passable gap: If the minimum interval distance on a certain connecting line is greater than the passable width threshold of the elderly care robot body, then this gap is directly marked as a passable gap, indicating that the robot can pass through safely and without obstruction.

[0117] Dynamically passable gaps: If the minimum gap distance on a connecting line is less than or equal to the passability width threshold, but at the same time, at least one of the two associated object blocks constituting this gap is identified as a movable object category (e.g., a lightweight chair, a trolley, etc.), then this gap will be assigned a dynamic passability tag. This means that although the current gap is insufficient for passage, it may become passable by moving the relevant objects through robots or external intervention.

[0118] After the above judgment and marking, the system will output an updated object relationship topology map with passable gap markings, which contains the passability status information of all gaps.

[0119] This application constructs an object relationship topology map and identifies passable gaps within it, enabling the transformation of discrete object information in the environment into a structured relationship network. It also allows for a refined evaluation of potential passage paths, including identifying areas that require dynamic interaction for passage. This provides robots with a higher level of environmental understanding and more flexible path planning capabilities.

[0120] 104. In the object relationship topology graph, determine the spatial relationship of items. When it is determined that there is a situation where the spatial relationship of items blocks the passage of high-priority items, it is determined that there is an abnormal spatial relationship of items.

[0121] Optionally, step 104 may specifically include the following steps:

[0122] 1041. Obtain the semantic label priority weights in the topological map of object relationships marked with passable gaps and the 3D obstacle map. The priority weights are pre-associated with the semantic labels.

[0123] The semantic label priority weight is a pre-assigned importance value for different object categories (i.e., semantic labels) in the environment. For example, "medical equipment" might be given a higher priority weight because it is crucial to the health of the elderly, while "ordinary chair" might have a lower priority weight.

[0124] In this embodiment, the system first obtains the object relationship topology map constructed and labeled in step 103. This map contains accessibility information for all object blocks in the environment and their spatial relationships and gaps. Simultaneously, the system also obtains pre-set semantic label priority weights from the 3D obstacle map. These weights are numerical values ​​assigned to different object categories (such as "medical equipment," "furniture," and "movable objects") based on the actual application scenario and elderly care needs, to measure their importance. For example, the priority weight for "medical equipment" can be set to 5, "furniture" to 3, and "movable objects" to 1.

[0125] 1042. Locate the continuous object block with the highest priority weight in the object relationship topology graph, and extract all passable gaps pointing to the corresponding block as necessary passage routes.

[0126] High-priority item corridors refer to paths or areas of space leading to or surrounding items with high priority weights. These corridors are crucial for robots to perform specific tasks or for ensuring the safety of the elderly.

[0127] In this embodiment, the system identifies contiguous object blocks with the highest priority weights in the object relationship topology graph based on the semantic tag priority weights obtained in 1041 steps. For example, if "medical equipment" is set as the highest priority, the system will locate all object blocks labeled "medical equipment". Subsequently, the system extracts all traversable gaps (including directly traversable and dynamically traversable gaps) directly connected to these high-priority object blocks from the topology graph and defines the paths represented by these gaps as necessary passageways. These routes are the key channels for the robot to approach or serve these high-priority items.

[0128] 1043. Traverse all connecting lines on necessary passage routes. When the minimum interval distance recorded by the connecting line is less than the passage width threshold of the elderly care robot and no dynamic passage feasibility marker is set, determine that the corresponding gap is in a blocked state.

[0129] Necessary routes are a type of high-priority item path, specifically referring to the critical paths that robots must traverse to reach or serve high-priority items.

[0130] In this embodiment, the system checks all necessary routes extracted in 1042 steps one by one. For each route, the system checks its corresponding connecting line in the object relationship topology graph and reads the minimum interval distance value between two related object blocks recorded on the connecting line. Simultaneously, the system checks whether the connecting line has a dynamic accessibility flag set. If the following two conditions are met:

[0131] Condition 1: The minimum interval distance is less than the passage width threshold of the elderly care robot body (i.e., the gap is too narrow and the robot cannot pass through).

[0132] Condition 2: The gap does not have a dynamic passage feasibility marker (i.e., the obstruction is fixed and immovable, or although it is movable, it is not identified as a movable object, so the blockage cannot be resolved by moving it).

[0133] The system will then determine that the gap is blocked. This means that the critical path to high-priority items is blocked by an obstacle that cannot be easily removed.

[0134] 1044. When any necessary passageway is detected to be blocked, output the judgment result of abnormal item spatial relationship.

[0135] A blocked state describes a situation where a gap or passage is obstructed by an obstacle, preventing the robot from passing safely. An abnormal spatial relationship of objects refers to an undesirable state in which the arrangement of objects in the environment is unreasonable, causing the necessary passage routes of high-priority objects to be blocked, which may affect the normal functioning of the robot or the safety of the elderly.

[0136] In this embodiment, if the system finds any necessary route to be blocked during the 1043-step traversal, it will immediately output a judgment result indicating that there is an abnormal spatial relationship between items in the current environment. This means that at least one high-priority item's access path is blocked by an immovable obstacle, and the robot needs to pay special attention to this situation and handle it specially in subsequent navigation planning.

[0137] This application intelligently determines the spatial relationships of objects in an object relationship topology map, especially identifying situations where the passage of high-priority objects is blocked. This enables the timely detection of potential navigation problems or safety hazards in the environment, thereby providing robots with a higher level of situational awareness, allowing them to respond to complex environments more proactively and intelligently.

[0138] 105. Based on the 3D obstacle map, object relationship topology map, passable gaps, and priority weights corresponding to semantic tags, combined with the judgment results of the spatial relationship of abnormal items, generate the navigation path of the elderly care robot.

[0139] Optionally, step 105 may specifically include the following steps:

[0140] When an abnormal spatial relationship is identified, the obstacle risk weight factor in the safety factor calculation is dynamically adjusted based on the semantic tags and priority weights involved in the abnormal spatial relationship. In the object relationship topology map, alternative passable gaps near high-priority object channels blocked by the abnormal spatial relationship are identified. Based on the adjusted obstacle risk weight factor and the corresponding alternative passable gaps, the initial calculated path is corrected or replanned to generate a navigation path that avoids the abnormal blockage area and meets the preset safety factor threshold.

[0141] The obstacle risk weighting factor is a parameter used in path planning algorithms to quantify the "danger level" of obstacles. A higher factor value indicates that the robot will tend to avoid the obstacle or area when planning its path. Alternative passable gaps refer to other gaps or passages in the environment that the robot can navigate when its expected main path is blocked. The preliminary calculated path is the basic path generated by the path planning algorithm before considering the spatial relationships of unusual objects.

[0142] Alternatively, when no abnormal spatial relationships are determined, a navigation path is generated based on the 3D obstacle map, the object relationship topology map, passable gaps, and the priority weights corresponding to semantic tags.

[0143] Specifically, when no abnormal spatial relationships are determined, the 3D coordinates of the current and target positions of the elderly care robot are determined in the 3D obstacle map, and all marked passable gaps in the object relationship topology map are extracted to form a gap set. For each gap in the gap set, the extended boundaries of the object surfaces on both sides of the gap are connected to generate a virtual channel space. Discrete sampling points are set in the virtual channel space, and the minimum distance from each sampling point to the nearest obstacle surface is calculated. Based on the minimum distance and the semantic label priority weights of the objects on both sides of the corresponding gap, the channel safety index is calculated. A path search space consisting of all virtual channel spaces is established between the current position and the target position. In the path search space, the connection path that optimizes the accumulated channel safety index and meets the preset minimum safety threshold is selected, and the coordinate point sequence of the connection path is transformed into the navigation path of the elderly care robot.

[0144] The virtual passageway space is a conceptual three-dimensional region constructed based on passable gaps by connecting the extended boundaries of the object surfaces on both sides of the gap. It defines the range within which the robot can safely traverse. Discrete sampling points are a series of points uniformly or randomly distributed within the virtual passageway space, used to evaluate the safety of different locations within the passageway space. The passageway safety index is a comprehensive indicator used to quantify the safety of the robot's passage within a specific passageway space, considering the distance between the robot and obstacles as well as the semantic priority of surrounding objects. The path search space is an abstract graph structure composed of all virtual passageway spaces and their connectivity relationships, in which the robot searches for the optimal path from its current position to its target position. A preset minimum safety threshold is a minimum acceptable value for the passageway safety index; paths below this threshold are considered unsafe.

[0145] The design aims to intelligently generate a navigation path that avoids obstacles while maintaining safety and efficiency, based on real-time environmental perception and semantic understanding, especially the judgment of high-priority item passage blockages, thus ensuring the safe and reliable operation of the elderly care robot in complex environments.

[0146] In this embodiment of the application, the process of generating the navigation path for the elderly care robot is divided into two cases depending on whether there are abnormal spatial relationships between items:

[0147] Scenario 1: When an abnormal spatial relationship between items is determined to exist: If the system determines in step 104 that an abnormal spatial relationship between items exists in the environment (i.e., the high-priority item passage is blocked), the navigation path generation will adopt a strategy of correction or replanning. The specific process is as follows:

[0148] First, the system dynamically adjusts the obstacle risk weight factor in the safety factor calculation based on the semantic tags and their priority weights involved in the spatial relationships of abnormal items. For example, if a passage to "medical equipment" is blocked by an "immovable cabinet," the system will significantly increase the obstacle risk weight factor of the "cabinet" and its surrounding area, making it as difficult as possible to avoid in path planning.

[0149] Secondly, the system will identify alternative passable gaps near high-priority item channels that are blocked by abnormal item spatial relationships in the object relationship topology map. This means that the robot will find other paths to bypass the blocked areas in order to reach or approach the high-priority items.

[0150] Finally, based on the adjusted obstacle risk weighting factor and the identified alternative passable gaps, the system will revise or replan the initial calculated path. Revision may involve local adjustments to the existing path, while replanning involves completely recalculating a new path. The ultimate goal is to generate a navigation path that effectively avoids abnormal congestion areas and whose safety factor meets a preset threshold.

[0151] Scenario 2: When no abnormal spatial relationships are identified: When the system determines that no abnormal spatial relationships exist in the environment, the navigation path will be generated based on the 3D obstacle map, object relationship topology map, passable gaps, and the priority weights corresponding to semantic tags, performing conventional path planning. The specific process is as follows:

[0152] First, the system determines the 3D coordinates of the current and target locations of the elderly care robot on a 3D obstacle map. Simultaneously, it extracts all marked passable gaps from the object relationship topology graph, forming a gap set.

[0153] Secondly, for each gap in the gap set, the system connects the extended boundaries of the object surfaces on both sides of the gap, thereby generating a virtual passage space. This virtual passage space represents the specific area within which the robot can safely pass.

[0154] Next, within each virtual channel space, the system sets up a series of discrete sampling points. For each sampling point, the system calculates its minimum distance to the nearest obstacle surface. Then, based on this minimum distance and the semantic label priority weights of the objects on both sides of the corresponding gap, the system calculates the channel safety index for that sampling point. For example, if a sampling point is close to a low-priority obstacle and is relatively far away, the safety index is high; conversely, if it is close to a high-priority obstacle or very close, the safety index is low.

[0155] Then, the system will establish a path search space between the current location and the target location, consisting of all connected virtual channel spaces. This search space can be viewed as a graph, where nodes are sampling points within the virtual channel spaces, and edges are walkable connections between sampling points.

[0156] Finally, in the path search space, the system uses a path search algorithm (such as A* algorithm or Dijkstra's algorithm) to select a connection path that optimizes the accumulated channel safety index and meets a preset minimum safety threshold. The coordinate sequence of this connection path will be transformed into the navigation path for the elderly care robot.

[0157] This application, by combining the determination results of the spatial relationships of abnormal objects, can flexibly adjust the path planning strategy, generate the optimal safe path under normal circumstances, and intelligently avoid high-risk areas under abnormal circumstances, thereby significantly improving the navigation intelligence, safety and robustness of elderly care robots in complex home environments.

[0158] Figure 2 This application provides a schematic diagram of the structure of an autonomous navigation system for elderly care robots based on depth vision, as shown in the embodiments below. Figure 2 As shown, the system includes:

[0159] The acquisition module 21 is used to acquire point cloud data of the environment using a three-dimensional vision sensor, and based on the point cloud data, determine the intensity value and incident angle change of the ambient light, and switch the point cloud fusion mode according to the intensity value and incident angle change to suppress ambient light interference and obtain optimized point cloud data.

[0160] The generation module 22 is used to perform semantic analysis on the optimized point cloud data and generate a 3D obstacle map with semantic labels of various categories.

[0161] The identification module 23 is used to construct an object relationship topology map based on the spatial distribution of objects in the 3D obstacle map, and to identify passable gaps between related objects in the object relationship topology map.

[0162] The judgment module 24 is used to judge the spatial relationship of items in the object relationship topology graph. When it is judged that there is a situation where the spatial relationship of items blocks the channel of high-priority items, it is determined that there is an abnormal spatial relationship of items.

[0163] The generation module 22 is also used to generate a navigation path for the elderly care robot based on the three-dimensional obstacle map, the topology of object relationships, the passable gaps and the priority weights corresponding to semantic tags, combined with the judgment results of the spatial relationships of abnormal items.

[0164] Figure 2 A depth vision-based autonomous navigation system for elderly care robots can perform... Figure 1 The implementation principle and technical effects of the autonomous navigation method for elderly care robots based on depth vision in the illustrated embodiment will not be elaborated further. The specific methods by which each module and unit performs operations in the autonomous navigation system for elderly care robots based on depth vision in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0165] In one possible design, Figure 2 The autonomous navigation system for elderly care robots based on depth vision, as shown in the embodiment, can be implemented as a computing device, such as... Figure 3As shown, the computing device may include a storage component 31 and a processing component 32;

[0166] Storage component 31 stores one or more computer instructions, wherein one or more computer instructions are invoked and executed by processing component 32.

[0167] Processing component 32 is used for the above Figure 1 An example of an autonomous navigation method for elderly care robots based on depth vision.

[0168] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0169] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0170] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0171] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0172] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0173] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0174] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown illustrates an autonomous navigation method for elderly care robots based on depth vision.

[0175] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0176] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0177] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications 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 this application.

Claims

1. A method for autonomous navigation of elderly care robots based on depth vision, characterized in that, include: The system uses a 3D vision sensor to collect point cloud data of the environment, and based on the point cloud data, determines the intensity value of ambient light and the change in incident angle. According to the intensity value and the change in incident angle, the system switches the point cloud fusion mode to suppress ambient light interference and obtain optimized point cloud data. Semantic analysis is performed on the optimized point cloud data to generate a 3D obstacle map with semantic labels of various categories; Based on the spatial distribution of objects in the three-dimensional obstacle map, an object relationship topology map is constructed, and passable gaps between related objects are identified in the object relationship topology map. In the object relationship topology diagram, the spatial relationship of items is determined. When it is determined that there is a situation where the spatial relationship of items blocks the passage of high-priority items, it is determined that there is an abnormal spatial relationship of items. Based on the 3D obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, combined with the determination results of the spatial relationships of abnormal items, a navigation path for the elderly care robot is generated, including: When an abnormal spatial relationship is determined, the obstacle risk weight factor in the safety factor calculation is dynamically adjusted according to the semantic tags and priority weights involved in the abnormal spatial relationship; in the object relationship topology map, alternative passable gaps near the high-priority item channels blocked by the abnormal spatial relationship are identified; based on the adjusted obstacle risk weight factor and the corresponding alternative passable gaps, the preliminary calculated path is corrected or replanned to generate a navigation path that avoids the abnormal blocking area and meets the preset threshold safety factor; Alternatively, when the existence of the abnormal object spatial relationship is not determined, a navigation path is generated based on the three-dimensional obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags.

2. The method according to claim 1, characterized in that, When the spatial relationship of the abnormal items is not determined, a navigation path is generated based on the 3D obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, including: When the existence of the abnormal object spatial relationship is not determined, the three-dimensional coordinates of the current position and target position of the elderly care robot are determined in the three-dimensional obstacle map, and all marked passable gaps in the object relationship topology map are extracted to form a gap set. For each gap in the gap set, the extended boundaries of the object surfaces on both sides of the gap are connected to generate a virtual channel space; Discrete sampling points are set within the virtual channel space. The minimum distance from each sampling point to the nearest obstacle surface is calculated. Based on the minimum distance and the semantic label priority weights of the objects on both sides of the corresponding gap, the channel safety index is calculated. Establish a path search space between the current location and the target location, consisting of all connected virtual channel spaces; In the path search space, a connection path is selected that optimizes the accumulated channel safety index and satisfies a preset minimum safety threshold. The coordinate point sequence of the connection path is then transformed into a navigation path for the elderly care robot.

3. The method according to claim 1, characterized in that, Based on the point cloud data, the intensity value and incident angle change of the ambient light are determined, and the point cloud fusion mode is switched according to the intensity value and incident angle change to suppress ambient light interference, resulting in optimized point cloud data, including: The intensity value of ambient light is obtained by calculating the average brightness of the color components of all points in the point cloud data. Based on the geometric distribution and brightness variation of the point cloud data, the change in the incident angle of ambient light is determined by identifying the direction with the largest brightness gradient in the point cloud and calculating the change in this direction between consecutive frames. When the intensity value changes or the incident angle changes by more than a preset threshold, the point cloud fusion mode is switched. The point cloud fusion mode includes a first fusion mode and a second fusion mode. The first fusion mode is enabled when the intensity value of the ambient light is higher than the threshold, and the second fusion mode is enabled when the intensity value of the ambient light is lower than the threshold. The point cloud data is processed according to the switched point cloud fusion mode. In the first fusion mode, the filtering weight of high brightness points is increased during the point cloud fusion process. In the second fusion mode, the degree of ignoring low brightness points during the point cloud fusion process is reduced to suppress ambient light interference and output optimized point cloud data.

4. The method according to claim 1, characterized in that, Semantic analysis is performed on the optimized point cloud data to generate a 3D obstacle map with semantic labels of various categories, including: A three-dimensional mesh structure is constructed, which contains multiple units, and the multiple units can cover the spatial range described by the optimized point cloud data; Extract the geometric features and surface texture features of the point cloud data within each unit. The geometric features include the spatial density distribution and curvature variation of the point cloud data within the unit, and the surface texture features include the color pattern distribution of the point cloud data within the unit. The geometric features and surface texture features are input into a pre-trained semantic recognition model, and the semantic recognition model outputs the object category identifier corresponding to each unit. The object category identifier includes ground, wall, furniture, medical equipment, and movable objects. Regions with the same object category identifier in adjacent units are merged to form continuous object blocks, and each continuous object block is labeled with a unified semantic tag. In the three-dimensional mesh structure, object blocks with a height higher than a preset passage height and object blocks of the impassable category are marked as obstacles. Integrate all units with semantic labels and obstacle markers to generate a 3D obstacle map that includes spatial location, object category identifiers, and obstacle markers.

5. The method according to claim 1, characterized in that, Based on the spatial distribution of objects in the 3D obstacle map, an object relationship topology map is constructed, and passable gaps between related objects are identified in the object relationship topology map, including: Extract the spatial location and boundary extent of all semantically labeled continuous object blocks in a 3D obstacle map; The minimum distance between any two object blocks is calculated based on the spatial location and boundary range. When the minimum distance is less than the preset proximity judgment distance, it is determined that the two object blocks have a spatial relationship. Construct an object relationship topology graph, treat each object block as a node, establish a connection line between two nodes that have the spatial relationship, and record the corresponding minimum interval distance value between the corresponding object blocks on the connection line; Identify the passable gaps between related objects in the object relationship topology graph, traverse all node pairs with connecting lines, and perform a passability judgment based on the minimum interval distance value recorded by the connecting lines; When the minimum interval distance value is greater than the passage width threshold of the elderly care robot body, it is marked as a passable gap. When the minimum interval distance value is less than or equal to the passage width threshold and the associated object block contains a movable object category, dynamic passage feasibility marking is performed, and an object relationship topology map with passable gap markings is output.

6. The method according to claim 1, characterized in that, In the object relationship topology diagram, the spatial relationships of items are determined. When it is determined that there is a situation where the spatial relationships of items block the passage of high-priority items, it is determined that there are abnormal spatial relationships of items, including: Obtain the semantic label priority weights in the object relationship topology map with passable gap markers and the 3D obstacle map, wherein the priority weights are pre-associated with the semantic labels; Locate the continuous object block with the highest priority weight in the object relationship topology graph, and extract all passable gaps pointing to the corresponding block as necessary passage routes. Traverse all connecting lines on necessary passage routes. When the minimum interval distance recorded by the connecting line is less than the passage width threshold of the elderly care robot and no dynamic passage feasibility marker is set, the corresponding gap is determined to be blocked. When any necessary route is detected to be blocked, the system outputs a determination result indicating the presence of abnormal spatial relationships between items.

7. An autonomous navigation system for elderly care robots based on depth vision, characterized in that, include: The acquisition module is used to acquire point cloud data of the environment using a three-dimensional vision sensor, and based on the point cloud data, determine the intensity value of ambient light and the change in incident angle, and switch the point cloud fusion mode according to the intensity value and the change in incident angle to suppress ambient light interference and obtain optimized point cloud data. The generation module is used to perform semantic analysis on the optimized point cloud data and generate a 3D obstacle map with semantic labels of various categories. The identification module is used to construct an object relationship topology map based on the spatial distribution of objects in the three-dimensional obstacle map, and to identify passable gaps between related objects in the object relationship topology map; The judgment module is used to judge the spatial relationship of items in the object relationship topology diagram. When it is judged that there is a situation where the spatial relationship of items blocks the channel of high-priority items, it is determined that there is an abnormal spatial relationship of items. The generation module is further configured to generate a navigation path for the elderly care robot based on the 3D obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags, combined with the determination results of the spatial relationships of abnormal items, including: When an abnormal spatial relationship is determined, the obstacle risk weight factor in the safety factor calculation is dynamically adjusted according to the semantic tags and priority weights involved in the abnormal spatial relationship; in the object relationship topology map, alternative passable gaps near the high-priority item channels blocked by the abnormal spatial relationship are identified; based on the adjusted obstacle risk weight factor and the corresponding alternative passable gaps, the preliminary calculated path is corrected or replanned to generate a navigation path that avoids the abnormal blocking area and meets the preset threshold safety factor; Alternatively, when the existence of the abnormal object spatial relationship is not determined, a navigation path is generated based on the three-dimensional obstacle map, the object relationship topology map, the passable gaps, and the priority weights corresponding to the semantic tags.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the autonomous navigation method for elderly care robots based on depth vision as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an autonomous navigation method for an elderly care robot based on depth vision as described in any one of claims 1 to 6.

Images