A method for obstacle recognition based on depth point cloud and a path planning method

Abstract

The application provides a depth point cloud-based obstacle identification method and a path planning method, the depth point cloud-based obstacle identification method comprising: obtaining original depth point cloud data through a depth camera and converting three-dimensional information thereof into a two-dimensional matrix; comparing the pitch angle change of adjacent point clouds with a set threshold in the two-dimensional matrix to distinguish planar point clouds from preliminary obstacle point clouds; and extracting obstacle information based on the preliminary obstacle point clouds. The depth point cloud-based obstacle identification method and the path planning method improve the accuracy, adaptability and rationality of the obstacle identification method and the safety of path planning based on a cost map by a robot.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of robot navigation, and more particularly relates to a method for obstacle recognition based on depth point cloud and a method for path planning. BACKGROUND

[0002] The path planning and motion control capability is a key factor for a robot to safely and efficiently perform a task. However, to achieve efficient path planning and motion control, the robot must have a strong complex environment perception capability. Especially in a variable environment, the perception capability of the robot is more stringent, which not only guarantees the safety of the robot during task execution, but also improves the efficiency of the robot during task execution and makes the robot more intelligent. The basis of path planning is to establish a safe and reliable real-time map.

[0003] In the prior art, the robot adopts a single / multi-line radar or depth camera scheme for obstacle detection. This method relies on some existing ground segmentation algorithms (such as the RANSAC plane fitting method) to select the points of the reference plane, and the points with a height difference exceeding the preset range of the reference plane are all obstacle points. Therefore, the traditional height difference detection method faces significant challenges in complex terrain: the height difference between gentle slopes (such as 1 cm / 10 cm) and steep slopes (such as 5 cm / 10 cm) is obvious, and real obstacles (such as steps) usually present a sudden height change (> 5 cm). If a high threshold (such as 5 cm) is set, low obstacles (2 cm toys) will be missed, and a low threshold (such as 1 cm) will misjudge the slope as an obstacle. This inherent contradiction in threshold setting, combined with the dynamic changes of mixed terrain (such as slope transition zones and concave-convex road surfaces) in the actual environment, makes it difficult for the traditional method to guarantee the completeness of obstacle detection and avoid terrain misjudgment. It can be seen that the obstacle recognition method of the prior art has the technical problems of insufficient accuracy, adaptability, and rationality.

[0004] In addition, after detecting the obstacle, it is mapped into a cost map. That is, after the planar point cloud is mapped into the cost map, the initial cost value of the planar point cloud is zero, and the obstacle is assigned a grid cost value. However, a flat surface does not mean safety and low energy consumption. Lack of quantitative evaluation of the subtle undulations of the road surface will cause the robot to bump or increase energy consumption, which leads to the problem of insufficient safety of the robot when planning a path based on the cost map. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a method for obstacle recognition based on depth point cloud and a method for path planning, to solve the technical problems of insufficient accuracy, adaptability, and rationality of the obstacle recognition method of the prior art, and the problem of insufficient safety of the robot when planning a path based on the cost map.

[0006] To achieve the above object, the technical scheme adopted by the present application is: a kind of obstacle identification method based on depth point cloud, comprising the following steps:

[0007] The original depth point cloud data is obtained by depth camera, and the three-dimensional information is converted into two-dimensional matrix;

[0008] The pitch angle change of adjacent point cloud in two-dimensional matrix is compared with the set threshold value, and the plane point cloud and the preliminary obstacle point cloud are distinguished;

[0009] Obstacle information is extracted based on preliminary obstacle point cloud.

[0010] The beneficial effects of the above embodiment are that: after the depth point cloud based on depth camera is projected into two-dimensional matrix, the method of comparing the pitch angle change of adjacent point cloud with the set threshold value is used to distinguish the plane point cloud and the preliminary obstacle point cloud, which has the characteristics of reflecting local geometric characteristics, robustness to terrain changes, universality of threshold design, "discrimination" view of uphill and downhill, and improves the accuracy, adaptability and rationality of obstacle avoidance.

[0011] The present application also provides a path planning method based on the above-mentioned obstacle identification method based on depth point cloud, which comprises the following steps:

[0012] The plane point cloud in two-dimensional matrix is clustered according to the pitch angle change to generate several plane clusters;

[0013] The pitch angle change characteristics of each plane cluster in the same row or column in two-dimensional matrix are used to generate the first generation value;

[0014] After mapping the two-dimensional matrix to the cost map, the first generation value is assigned to the corresponding grid of the plane cluster as the initial generation value;

[0015] The path is planned based on the cost map.

[0016] The beneficial effects of the above embodiment are that: by projecting the point cloud to two-dimensional matrix and clustering analysis based on pitch angle change characteristics, efficient modeling and evaluation of complex terrain are realized. That is, first, the plane cluster is generated by clustering based on pitch angle change, which effectively distinguishes continuous terrain from real obstacles; first, the plane cluster is generated based on pitch angle change clustering, and then the first generation value is generated based on the pitch angle change rate and amplitude in row / column direction, which quantifies the passable degree of terrain; finally, by mapping to cost map and assigning initial generation value, fine terrain evaluation is provided for path planning, which not only retains the three-dimensional geometric characteristics of point cloud, but also realizes the balance of calculation efficiency and navigation accuracy, significantly improves the terrain perception and decision-making ability of robot in complex environment, that is, the safety of robot is comprehensively improved.

[0017] The application also provides another path planning method based on the above-mentioned obstacle identification method based on depth point cloud, which comprises the steps of:

[0018] The planar point cloud in the two-dimensional matrix is clustered according to the change of the pitch angle to generate a plurality of planar clusters;

[0019] The target type is obtained by classifying the corresponding RGB image area of each planar cluster, and the second generation value is generated according to the target type;

[0020] After the two-dimensional matrix is mapped into the cost map, the initial cost value of the grid corresponding to the planar cluster is given based on the second generation value;

[0021] The path is planned based on the cost map.

[0022] The above-mentioned embodiment has the beneficial effects that the above-mentioned method realizes the fine classification and evaluation of the planar area by combining the geometric characteristics of the three-dimensional point cloud and the semantic information of the RGB image: first, the planar clusters are generated by clustering based on the change of the pitch angle, then the target type with different physical characteristics is identified through the RGB image, and the second generation value reflecting the actual passing difficulty is generated; finally, the differentiated cost value is given to different planar types in the cost map, so that the robot path planning can not only avoid obstacles, but also intelligently select the optimal road surface type, significantly improving the safety, smoothness and energy efficiency ratio of navigation, and solving the planning defects of traditional methods caused by ignoring semantic information, i.e. "geometrically feasible but actually infeasible".

[0023] The application also provides another path planning method based on the above-mentioned obstacle identification method based on depth point cloud, which comprises the steps of

[0024] The planar point cloud in the two-dimensional matrix is clustered according to the change of the pitch angle to generate a plurality of planar clusters;

[0025] The first generation value is generated according to the pitch angle change characteristics of each planar cluster in the same row or column in the two-dimensional matrix;

[0026] The target type is obtained by classifying the corresponding RGB image area of each planar cluster, and the second generation value is generated according to the target type;

[0027] After the two-dimensional matrix is mapped into the cost map, the initial cost value of the grid corresponding to the planar cluster is given based on the first generation value and the second generation value;

[0028] The path is planned based on the cost map.

[0029] The above-mentioned embodiment has the beneficial effect that the method innovatively combines the point cloud geometric features and the visual semantic information reflecting the physical characteristics, realizing the complementary advantages of multi-dimensional environment evaluation: the first generation value generated by the pitch angle change characteristics accurately quantifies the terrain geometric features, solving the problem of pure visual method being insensitive to geometric features; meanwhile, the second generation value generated based on the RGB classification supplements the semantic level physical properties, making up for the lack of semantic understanding of pure point cloud method. The intelligent fusion of the two ensures the accurate perception of the robot to the complex terrain geometric features, and realizes the functional judgment of the road surface type, so that the path planning system can take into account both the "geometric physical passability" and the "actual applicability", achieving the optimal balance between safety and efficiency in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0031] Figure 1 A flowchart of the obstacle recognition method based on depth point cloud provided by the first embodiment of the present application;

[0032] Figure 2 A schematic diagram of the angle between the point cloud in the depth camera coordinate system and the X axis;

[0033] Figure 3 A schematic diagram of converting three-dimensional information of depth point cloud data into a two-dimensional matrix;

[0034] Figure 4 A flowchart of the obstacle recognition method based on depth point cloud provided by the second embodiment of the present application;

[0035] Figure 5 A flowchart of the path planning method provided by the fifth embodiment of the present application;

[0036] Figure 6 An example diagram of the depth point cloud visualization scene effect provided by the fifth embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the technical problems, technical solutions and beneficial effects of the present application more clear, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0038] In addition, the terms "first", "second", etc. are used only for descriptive purposes and should not be construed as indicating or implying relative importance or an indicated number of technical features. Therefore, the features defined as "first", "second", etc. can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0039] Please refer to Figure 1 The depth point cloud based obstacle recognition method provided by the first embodiment of the present application will be described. The depth point cloud based obstacle recognition method comprises:

[0040] Step S1, obtaining original depth point cloud data through a depth camera and converting three-dimensional information thereof into a two-dimensional matrix;

[0041] Step S2, comparing the change of the pitch angle of adjacent point clouds with a set threshold in the two-dimensional matrix to distinguish between planar point clouds and preliminary obstacle point clouds;

[0042] Step S3, extracting obstacle information based on the preliminary obstacle point clouds.

[0043] It can be understood that in step S1, the original depth point cloud data can be directly obtained through the SDK interface of the depth camera, and then the three-dimensional information of the depth point cloud data is converted into a two-dimensional matrix based on the existing calibration method. Thus, the spatial relative relationship of the depth camera and any point cloud in the two-dimensional matrix can be obtained.

[0044] In this way, in step S2, after projecting the 3D point cloud into the two-dimensional matrix, the method of comparing the change of the pitch angle of adjacent point clouds with a set threshold is used to distinguish between planar point clouds and preliminary obstacle point clouds. Compared with the prior art, the beneficial effects are:

[0045] (1) The pitch angle reflects local geometric features

[0046] The pitch angle in the two-dimensional matrix is the ratio of the height difference and the horizontal distance, which is essentially a measure of the local slope. Therefore, in the gentle slope scene, the pitch angle is small (for example, 5°), even if the height difference accumulates is large, but the angle change is gentle, which avoids false detection of obstacles. On the contrary, in the obstacle pitch angle mutation (for example, more than 30°), it forms an obvious discontinuous geometric shape with the surrounding ground, which prevents the obstacle from being missed.

[0047] (2) Robustness to terrain changes

[0048] When facing a gentle slope, the pitch angles of the adjacent point clouds are close, even if the height difference is accumulated, the angle change is still below the threshold. When facing a steep slope or uneven ground, if the slope is consistent (such as a fixed inclination angle), the pitch angle change is still small. When facing an obstacle, the pitch angle changes abruptly (such as from 0° to 45°), directly triggering the threshold judgment.

[0049] (3) Universality of threshold design

[0050] Only by setting an angle threshold (for example, 15°) can gentle slopes, steep slopes and ground fluctuations be filtered at the same time, and only real obstacles are retained. There is no need to adjust parameters for different terrains.

[0051] (4) "Discrimination" view between uphill and downhill

[0052] In the uphill scenario in the two-dimensional matrix, the pitch angle change of the adjacent point clouds is often greater than the actual slope, while in the downhill scenario, the pitch angle change of the adjacent point clouds is often less than the actual slope, thus forming a "discrimination" view when facing uphill and downhill scenarios with the same slope, that is, uphill is more likely to be judged as an obstacle, while downhill is relatively easier to be judged as a road surface. However, this "discrimination" view is exactly opposite to the difficulty of uphill and downhill, that is, under normal circumstances, uphill is relatively difficult than downhill, thus requiring a smaller slope (for example, less than 14°) to be judged as a flat surface, and downhill is relatively easy than uphill, thus requiring a larger slope (for example, greater than 16°) to be judged as an obstacle, which improves the rationality of the definition of obstacles.

[0053] Thus, in step S3, the obstacle information is extracted based on the preliminary obstacle point cloud, while the accuracy, adaptability and rationality of obstacle avoidance are improved.

[0054] It is worth supplementing that the adjacent point clouds can be two adjacent point clouds in the same column of the two-dimensional matrix, or two adjacent point clouds in the same row of the two-dimensional matrix. Preferably, the adjacent point clouds are two adjacent point clouds in the same column of the two-dimensional matrix.

[0055] It is worth supplementing that before step S1, an external parameter calibration step of the depth camera and the radar can also be included, by accurately calibrating the relative poses (rotation matrix and translation vector) of the two, the sensor data in different coordinate systems is unified to the radar coordinate system, solving the spatial alignment problem. This step can eliminate the spatial deviation caused by the difference in installation position, and the rotation matrix and translation vector of the radar between two time points can be solved by matching the point cloud data between two frames, and the sensor pose drift in the robot motion can also be compensated in real time through dynamic calibration, ensuring that the depth camera point cloud and the radar data are strictly aligned in space and time, providing high-precision geometric consistency guarantee for subsequent obstacle recognition, coordinate conversion and obstacle avoidance decision.

[0056] It is worth mentioning that after step S3, an obstacle point cloud data conversion step can also be included,

[0057] Based on the extracted obstacle information, the obstacle information converted into an obstacle avoidance module available is published. In this way, a depth camera can be used instead of a multi-line laser radar, not only improving the accuracy of obstacle recognition, but also breaking the software ecological barrier.

[0058] Further, please refer to Figure 2 In step S1, after obtaining the original depth point cloud data, and before converting its three-dimensional information into a two-dimensional matrix, a step is included:

[0059] Calculate the angle angle x of each depth point cloud with the X-axis in the depth camera coordinate system.

[0060] If angle x >a*f, a is a proportional coefficient, and f is the horizontal field of view angle of the depth camera, then the depth point cloud is filtered;

[0061] The formula for the angle angle x is:

[0062] ,

[0063] In the formula, x and y are the X and Y axis coordinates of the corresponding depth point cloud.

[0064] It can be understood that the camera has a limited field of view, and the data at the edge may not be accurate. By calculating the angle angle x of each point using the above formula, the points with too large an angle are removed, and only the valid point data in front is retained, which can improve the accuracy of the depth point cloud data.

[0065] Specifically, please refer to Figure 3 In step S1, converting the three-dimensional information of the original depth point cloud data into a two-dimensional matrix includes the following steps:

[0066] Calculate the row number row Id of each depth point cloud projected into the two-dimensional matrix, and the calculation formula of the row number row Id is:

[0067] ,

[0068] In the formula, x, y, and z are the X, Y, and Z axis coordinates of the corresponding depth point cloud, dis xy is the horizontal distance of the depth point cloud projected into the X-Y plane in the camera coordinate system, ang z is the angle of the depth point cloud with the X-Y plane, and ang vminang is the minimum pitch angle of the depth camera. v This represents the angular resolution of the depth camera in the vertical direction.

[0069] Calculate the column number col of the point cloud projection into the two-dimensional matrix at each depth. Id Column number col Id The calculation formula is:

[0070] ,

[0071] In the formula, ang y Let ang be the angle between the depth point cloud and the Y-axis. hmin ang is the minimum horizontal angle of the depth camera. h The horizontal angular resolution of the depth camera;

[0072] According to the row number Id With column number col Id Determine the position of each depth point cloud in a two-dimensional matrix.

[0073] Understandably, the above method, by combining the installation parameters of the depth camera (such as horizontal / vertical field of view and angular resolution) with precise geometric calculation formulas, achieves high-fidelity projection of point cloud data into a two-dimensional matrix. This not only fully preserves the spatial distribution characteristics of the original 3D point cloud but also establishes quantifiable topological relationships through row and column mapping. This projection method overcomes the information loss problem caused by fixed resolution or simple dimensionality reduction in traditional methods. It is particularly suitable for the top-down installation of depth cameras in this application, providing a high-precision data structure foundation for subsequent pitch angle calculation and obstacle recognition based on row and column relationships, significantly improving the detection accuracy and real-time performance of the algorithm in complex terrain.

[0074] Specifically, the steps for calculating the pitch angle changes of adjacent point clouds include:

[0075] Let P1(x1,y1,z1) and P2(x2,y2,z2) be adjacent point clouds in a two-dimensional matrix;

[0076] Then the pitch angle change diff between adjacent point clouds P1 and P2 ang The calculation formula is:

[0077] ,

[0078] In the formula, diff x diff y diff z These are the coordinate differences between points P1 and P2 on the X-axis, Y-axis, and Z-axis, respectively.

[0079] It can be understood that the above pitch angle calculation formula significantly improves the accuracy of obstacle detection by normalizing the height change and the horizontal distance. The formula not only considers the height difference in the vertical direction, but also combines the horizontal displacement, so that the calculation result truly reflects the actual slope change between point clouds, effectively eliminating the error caused by uneven point cloud density or different measurement distances. Compared with the simple plane fitting method detection, this angle calculation method based on geometric relationship can more accurately distinguish between real obstacles and terrain undulations, especially suitable for low obstacle detection in complex environments, greatly reducing the misjudgment rate caused by fixed height threshold in traditional methods.

[0080] For reference Figure 4 The second embodiment of the present application provides a depth point cloud-based obstacle recognition method. The second embodiment is based on the first embodiment.

[0081] Specifically, the step of extracting obstacle information based on the preliminary obstacle point cloud includes:

[0082] The preliminary obstacle point cloud is clustered, and the number of clustered point clouds is set as a condition to remove noise points;

[0083] The height of each type of clustered point cloud is calculated, the height difference of this type of point cloud is calculated, and the calculated height difference is compared with the required obstacle height. If the calculated height difference is greater than the detected obstacle height, it is considered that this type of point cloud is a real obstacle point cloud;

[0084] Based on the real obstacle point cloud, an obstacle is generated in the map.

[0085] It can be understood that the combination of the above two steps of point cloud clustering and height calculation of each type of point cloud significantly improves the accuracy and robustness of obstacle detection: through clustering operation, the discrete preliminary obstacle point cloud is classified into independent objects with physical meaning, effectively filtering noise points and isolated misjudgment points; then the height difference of each type of point cloud is calculated to further verify the authenticity and size characteristics of the obstacle, ensuring that only significant obstacles meeting the height threshold will be finally identified, thereby achieving high-precision obstacle detection in complex environments while avoiding misjudgment of ground micro undulations or sensor noise.

[0086] The following is a description of the path planning method provided by the third to seventh embodiments of the present application.

[0087] In the cost map of the existing path planning, the initial cost value of the planar point cloud mapped into the cost map is zero, and only the obstacle is assigned a grid cost value, that is, the planar road surface in the cost map is not finely classified, that is, the geometric characteristics and physical characteristics (such as gentle slope, uneven or slippery road surface) of different planar terrains are not distinguished, and the robot may select a geometrically feasible but actually unsafe path (such as steep slope or wet slippery area); secondly, the lack of quantitative evaluation of the slight ups and downs of the road surface will cause the robot to bump or increase the energy consumption; more importantly, this coarse-grained processing will mask the transition area with potential risks (such as the junction of the slope and the plane), so that the robot cannot avoid the possible motion control risks or skidding hazards in advance, thus causing the problem of insufficient safety of the robot when planning the path based on the cost map.

[0088] The path planning method provided in the third embodiment of the present application will be described. The third embodiment is based on the first and / or second embodiment.

[0089] The path planning method comprises the steps of:

[0090] The planar point cloud in the two-dimensional matrix is clustered according to the change of the pitch angle to generate a plurality of planar clusters;

[0091] The first cost value is generated according to the pitch angle change characteristics of each planar cluster in the same row or column of the two-dimensional matrix;

[0092] After the two-dimensional matrix is mapped into the cost map, the initial cost value of the corresponding grid of the planar cluster is assigned based on the first cost value;

[0093] The path is planned based on the cost map.

[0094] However, the above method realizes efficient modeling and evaluation of complex terrain by projecting the point cloud into a two-dimensional matrix and clustering analysis based on the pitch angle change characteristics. That is, first, the planar clusters are generated by clustering according to the pitch angle change, effectively distinguishing between continuous terrain and real obstacles; then, the first cost value is generated based on the pitch angle change rate and amplitude in the row / column direction, quantifying the passable degree of the terrain; finally, by mapping into the cost map and assigning the initial cost value, a fine terrain evaluation is provided for path planning, which not only retains the three-dimensional geometric characteristics of the point cloud, but also realizes the balance between calculation efficiency and navigation accuracy, significantly improving the terrain perception and decision-making ability of the robot in complex environments, that is, the safety of the robot is comprehensively improved.

[0095] The path planning method provided in the fourth embodiment of the present application will be described. The fourth embodiment is based on the first and / or second embodiment.

[0096] The path planning method comprises the steps of:

[0097] The planar point cloud in the two-dimensional matrix is clustered according to the change of the pitch angle to generate a plurality of planar clusters;

[0098] According to the target type, a second generation value is generated;

[0099] After the two-dimensional matrix is mapped into the cost map, the planar cluster corresponding to the grid is assigned an initial generation value based on the second generation value;

[0100] The path is planned based on the cost map.

[0101] It can be understood that the above method realizes the fine classification and evaluation of the planar area by combining the geometric features of the three-dimensional point cloud and the semantic information of the RGB image: first, the planar clusters are generated based on the change of the pitch angle, then the target types (such as rubber road surface, cement road surface, grassland, water accumulation, sand ground, etc.) with different physical characteristics are identified through the RGB image, and the second generation value reflecting the actual passing difficulty is generated; finally, the different planar types are assigned with differentiated generation values in the cost map, so that the robot path planning can not only avoid obstacles, but also intelligently select the optimal road surface type (such as preferentially selecting hardened road surface instead of grassland), which significantly improves the safety, smoothness and energy efficiency ratio of navigation, and solves the planning defects of traditional methods caused by ignoring semantic information, i.e., "geometrically feasible but actually infeasible".

[0102] Please refer to Figure 5 and Figure 6 for the fifth embodiment of the path planning method provided by the present application. The fifth embodiment is based on the first and / or second embodiment.

[0103] The path planning method comprises the steps of:

[0104] The planar point cloud in the two-dimensional matrix is clustered according to the change of the pitch angle to generate a plurality of planar clusters;

[0105] According to the pitch angle change characteristics of each planar cluster in the same row or column of the two-dimensional matrix, a first generation value is generated;

[0106] According to the target type, a second generation value is generated;

[0107] After the two-dimensional matrix is mapped into the cost map, the planar cluster corresponding to the grid is assigned an initial generation value based on the first generation value and the second generation value;

[0108] The path is planned based on the cost map.

[0109] It can be understood that the method innovatively combines the point cloud geometric features and the visual semantic information reflecting the physical characteristics, realizes the complementary advantages of multi-dimensional environment evaluation: the first generation value generated by the pitch angle change characteristics accurately quantifies the terrain geometric features (such as slope, flatness), solves the problem that the pure visual method is not sensitive to geometric features; and the second generation value generated based on the RGB classification supplements the semantic physical properties (such as road surface material, softness, friction coefficient, safety), makes up for the lack of semantic understanding of the pure point cloud method. The intelligent fusion of the two ensures the accurate perception of the robot to the complex terrain geometric features, and realizes the functional judgment of the road surface type, so that the path planning system can take into account both the "geometric and physical passability" and the "actual applicability", and achieve the optimal balance between safety and efficiency in complex scenarios.

[0110] The path planning method provided by the sixth embodiment of the present application will be described. The sixth embodiment is based on the fifth embodiment.

[0111] The step of assigning the initial generation value of the grid corresponding to the planar cluster based on the first generation value and the second generation value includes:

[0112] The sum of the first generation value and the second generation value is taken as the initial generation value of the grid corresponding to the planar cluster.

[0113] The path planning method provided by the seventh embodiment of the present application will be described. The seventh embodiment is based on the fifth embodiment.

[0114] The step of assigning the initial generation value of the grid corresponding to the planar cluster based on the first generation value and the second generation value includes:

[0115] According to the influence degree of the geometric terrain features of different classification results on the robot's passability, a proportion coefficient of the first generation value and the second generation value is preset;

[0116] The initial generation value of the grid corresponding to the planar cluster is obtained by weighted summation through the proportion coefficient.

[0117] It can be understood that the method realizes the intelligentization and adaptive optimization of environment evaluation by dynamically adjusting the weight proportion of geometric features and semantic information: according to the actual influence degree of different types of terrain on the robot's passability (such as steep slope focusing on geometric features, wet and slippery road surface focusing on material properties), a differentiated value proportion coefficient is preset, and the final generation value is generated by weighted fusion. This flexible weight mechanism not only retains the advantage of accurate quantification of point cloud data on terrain geometric characteristics, but also fully utilizes the recognition ability of visual information on road surface material, so that the cost map can automatically adjust the evaluation focus according to different scene characteristics, providing more accurate and more practical passability demand-oriented environment evaluation results for path planning, thereby significantly improving the navigation adaptability and decision rationality of the robot in complex and variable environments.

[0118] It is worth supplementing that in the third to seventh embodiments of the present application, the pitch angle change characteristic can refer to the pitch angle change rate or the pitch angle change size. Preferably, the pitch angle change characteristic includes the pitch angle change rate and the pitch angle change size to more accurately evaluate the influence degree of the geometric terrain feature on the robot.

[0119] It is worth supplementing that in the fourth to seventh embodiments of the present application, the classification refers to classification detection using a deep learning classification model to distinguish target types, wherein the method for classification detection using a deep learning classification model is a mature prior art.

[0120] The above only describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for obstacle recognition based on depth point cloud, characterized in that, The method comprises the following steps: obtaining original depth point cloud data through a depth camera and converting three-dimensional information of the original depth point cloud data into a two-dimensional matrix; comparing pitch angle changes of adjacent point clouds in the two-dimensional matrix with a set threshold value to distinguish between planar point clouds and preliminary obstacle point clouds; extracting obstacle information based on the preliminary obstacle point clouds; after obtaining the original depth point cloud data and before converting three-dimensional information of the original depth point cloud data into a two-dimensional matrix, the method further comprises the following steps: Calculate the angle between each depth point cloud and the X axis in the depth camera coordinate system angle x ; if angle x > a * f, a is a proportional coefficient, and f is the horizontal field of view angle of the depth camera, then the depth point cloud is filtered; where angle x The formula is: ， wherein x and y are X and Y axis coordinates of the depth point cloud.

2. The depth point cloud based obstacle identification method of claim 1, wherein, converting three-dimensional information of the original depth point cloud data into a two-dimensional matrix comprises the following steps: The row number row of each depth point cloud projected into a two-dimensional matrix is calculated Id The calculation formula of the row number row is as follows: Id The calculation formula of the row number row is as follows: ， ， ， In the formula, x, y, z are X, Y, Z axis coordinates of the corresponding depth point cloud, dis xy is the horizontal distance of the depth point cloud projected to the X-Y plane under the camera coordinate system, ang z is the angle between the depth point cloud and the X-Y plane, ang vmin is the minimum pitch angle of the depth camera, ang v is the angular resolution in the vertical direction of the depth camera; The column number col of each depth point cloud projected into a two-dimensional matrix is calculated Id The calculation formula of the column number col is as follows: Id The calculation formula of the column number col is as follows: ， ， where ang y is the angle between the depth point cloud and the Y-axis, ang hmin is the minimum horizontal angle of the depth camera, ang h is the horizontal angle resolution of the depth camera; According to the row number row Id With the column number col Id The position of each depth point cloud in the two-dimensional matrix is determined.

3. The depth point cloud based obstacle identification method of claim 2, wherein, the step of calculating the pitch angle changes of the adjacent point clouds comprises: taking P1(x1, y1, z1) and P2(x2, y2, z2) as adjacent point clouds in the two-dimensional matrix; diff = tan-1 ( (P2 - P1) / D) (1) ang The calculation formula is: ， ， ， ， In the formula, diff x , diff y , and diff z are coordinate difference values of the P1 and P2 points between the X axis, the Y axis, and the Z axis, respectively.

4. The depth point cloud based obstacle identification method of claim 1, wherein, the step of extracting obstacle information based on the preliminary obstacle point clouds comprises: performing clustering on the preliminary obstacle point clouds, and removing noise points by setting the number of clustered point clouds as a condition; calculating the height of each type of clustered point cloud, and comparing the calculated height difference with a required obstacle height, and regarding the type of point cloud as a real obstacle point cloud if the calculated height difference is greater than the required obstacle height; generating an obstacle in a map based on the real obstacle point cloud.

5. A path planning method characterized by, The method for identifying obstacles based on depth point clouds according to any one of claims 1-4 comprises the following steps: performing clustering on the planar point clouds in the two-dimensional matrix according to the pitch angle changes to generate a plurality of planar clusters; generating a first generation value according to the pitch angle change characteristics of each planar cluster in the same row or column in the two-dimensional matrix; after mapping the two-dimensional matrix into a cost map, assigning an initial cost value of a grid corresponding to the planar cluster based on the first generation value; planning a path based on the cost map.

6. A path planning method characterized by, The method for identifying obstacles based on depth point clouds according to any one of claims 1-4 comprises the following steps: performing clustering on the planar point clouds in the two-dimensional matrix according to the pitch angle changes to generate a plurality of planar clusters; generating a second generation value according to the target type obtained by classifying an image region corresponding to each planar cluster; after mapping the two-dimensional matrix into a cost map, assigning an initial cost value of a grid corresponding to the planar cluster based on the second generation value; planning a path based on the cost map.

7. A path planning method characterized by, The method for identifying obstacles based on depth point clouds according to any one of claims 1-4 comprises the following steps: performing clustering on the planar point clouds in the two-dimensional matrix according to the pitch angle changes to generate a plurality of planar clusters; generating a first generation value according to the pitch angle change characteristics of each planar cluster in the same row or column in the two-dimensional matrix; generating a second generation value according to the target type obtained by classifying an image region corresponding to each planar cluster; after mapping the two-dimensional matrix into a cost map, assigning an initial cost value of a grid corresponding to the planar cluster based on the first generation value and the second generation value; planning a path based on the cost map.

8. The path planning method of claim 7, wherein, The step of assigning an initial cost value of a grid corresponding to the planar cluster based on the first generation value and the second generation value comprises: taking the sum of the first generation value and the second generation value as the initial cost value of the grid corresponding to the planar cluster.

9. The path planning method of claim 7, wherein, The step of assigning the initial generation value of the corresponding grid to the planar cluster based on the first generation value and the second generation value comprises: According to the influence degree of the geometric terrain features of different classification results on the passing of the robot, a proportion coefficient of the first generation value and the second generation value is preset; The initial generation value of the corresponding grid of the planar cluster is obtained by weighted summation of the proportion coefficient.

Images