A Multi-Vision-Based Intelligent Obstacle Avoidance Method and System for AMR Robots

Abstract

This application relates to the field of robot obstacle avoidance technology, specifically to an intelligent obstacle avoidance method and system for AMR robots based on multi-view vision. The method includes: acquiring feature points and disparity of a point cloud; filtering abnormal and normal feature points using motion vectors of the feature points, and obtaining the significance and fluctuation intensity of disparity changes based on disparity change trends; obtaining directional hazard indicators based on changes in motion vector direction, obtaining spatial position consistency based on fluctuations in motion vector magnitude, and obtaining position hazard indicators based on optical flow, the significance of disparity changes, and the directional hazard indicators; determining the difference density through position hazard indicators and spatial position consistency, and obtaining a corrected difference density through disparity fluctuations and frequency corrections for different feature points; processing the corrected difference density based on hazard indicator fluctuations to obtain a spatial interpolation density, thereby performing obstacle avoidance through density difference calculations. This application improves obstacle avoidance performance.

Description

Technical Field

[0001] This application relates to the field of robot obstacle avoidance technology, specifically to an intelligent obstacle avoidance method and system for AMR robot operation based on multi-view vision. Background Technology

[0002] The core capability of Autonomous Mobile Robots (AMRs) is "autonomous movement," and obstacle avoidance is a prerequisite for ensuring their safe movement. As manufacturing and warehousing logistics upgrade towards intelligence, the requirements for the autonomy, flexibility, and environmental adaptability of AMRs are increasing, driving the development of obstacle avoidance sensor technology. Multi-view vision obstacle avoidance, due to its cost and perception characteristics—such as the ability to acquire depth information of a scene through images from different perspectives—is widely used in warehousing logistics, manufacturing workshops, and service industry robot scenarios. Since its core is a standard binocular camera, its cost is far lower than that of LiDAR, significantly reducing the overall manufacturing cost of AMRs and facilitating large-scale commercial deployment.

[0003] Currently, obstacle avoidance analysis for AMR robots requires multiple data processing steps, including simultaneous image acquisition, distortion correction, stereo matching, and depth calculation using multi-view vision. These processes generate massive amounts of data, resulting in a heavy computational burden on the robot. Consequently, the obstacle avoidance decisions made based on the images suffer from high latency, which in turn affects the actual obstacle avoidance performance of the AMR robot, leading to poor obstacle avoidance results. Summary of the Invention

[0004] To address the technical problem of poor obstacle avoidance performance, this application provides an intelligent obstacle avoidance method and system for AMR robot operation based on multi-view vision. The specific technical solution adopted is as follows:

[0005] Firstly, this application proposes an intelligent obstacle avoidance method for AMR robot operation based on multi-view vision, which includes the following steps:

[0006] Motion images of point clouds, their feature points, and corresponding parallaxes are acquired using the binocular cameras of an AMR robot.

[0007] Motion vectors of feature points are obtained based on optical flow algorithm, and abnormal and normal feature points are filtered according to motion vectors. The disparity of a point cloud is sorted according to time sequence to obtain disparity time sequence curve. The significance of disparity change and fluctuation intensity are obtained based on the disparity change trend of the time sequence curve.

[0008] The directional hazard index is determined based on the changes in motion vectors of feature points in different directions during the time series; the optical flow magnitude curve is obtained based on the magnitude of the motion vectors in the time series, and the significance of optical flow changes and fluctuation intensity are obtained using the same method as the disparity time series curve; the spatial position consistency is determined based on the differences in the significance of disparity and optical flow changes and fluctuation intensity; and the position hazard index is determined based on the significance of optical flow and disparity changes and the directional hazard index.

[0009] The interpolation density of feature points is determined based on the location hazard index and spatial location consistency of feature points; feature consistency is determined based on the fluctuation intensity of feature point disparity and the frequency difference between normal and abnormal feature points; the corrected interpolation density is obtained by correcting the interpolation density through feature consistency.

[0010] The spatial interpolation density is determined based on the fluctuation of the hazard index and the correction interpolation density; the density is graded by spatial interpolation density to generate a dense point cloud to complete intelligent obstacle avoidance.

[0011] In the aforementioned scheme, this application determines the interpolation density corresponding to the generation of a dense point cloud from a sparse point cloud by assessing the obstacle risk of the AMR robot, as well as the accuracy of the obstacle assessment and the relative movement trend of the point cloud. Simultaneously, it corrects the current interpolation density by considering the consistency of obstacle risk assessment with point clouds in adjacent spatial locations, thus achieving adaptive control over the amount of interpolation data generated for the dense point cloud. This reduces computational load while improving the accuracy of obstacle analysis and providing an accurate obstacle avoidance strategy.

[0012] In one embodiment, the motion vector of the normal feature point is a zero vector, while the motion vector of the abnormal feature point is not a zero vector.

[0013] In one embodiment, the method for obtaining the significance of disparity changes and the intensity of disparity fluctuations based on the disparity change trend of the time difference time series curve is as follows:

[0014] The derivative of each time series is obtained by taking the derivative of the disparity time series curve and denoted as the disparity change rate. The ReLU function is used to map all the disparity change rates to obtain the change rate curve.

[0015] Calculate the mean and standard deviation of all elements in the rate of change curve, and denote them as the significance of the disparity change and the intensity of the disparity fluctuation, respectively.

[0016] In one embodiment, the method for determining the directional hazard index based on the change in the motion vector of feature points in different directions during time is as follows:

[0017] The motion vector has a component on each coordinate axis. The change value of each component is recorded as the change amount. The standard deviation of all changes of each component is taken as the optical flow fluctuation of that component. The sum of the optical flow fluctuations of different components is taken as the directional hazard indicator during the movement of the AMR robot.

[0018] In one embodiment, the method for determining spatial location consistency based on the significance of changes in parallax and optical flow and the difference in fluctuation intensity is as follows:

[0019] The significance of the change and the intensity of the fluctuation are combined into a vector, and two-dimensional vectors are obtained for time difference and optical flow respectively. Spatial position consistency is analyzed based on the difference between the two-dimensional vectors of parallax and optical flow. The spatial position consistency is positively correlated with the difference between the two-dimensional vectors.

[0020] In one embodiment, the location hazard index is positively correlated with the significance of changes in optical flow and parallax, and the orientation hazard index.

[0021] In one embodiment, the interpolation density is positively correlated with the location hazard index and negatively correlated with spatial location consistency.

[0022] In one embodiment, the feature consistency is positively correlated with the difference in the fluctuation intensity, abnormal frequency, and normal frequency of the feature point disparity, respectively; the abnormal frequency is the proportion of abnormal feature points to all feature points, and the normal frequency is the proportion of normal feature points to all feature points.

[0023] In one embodiment, the spatial interpolation density is positively correlated with the fluctuation of the hazard index and the corrected interpolation density, respectively.

[0024] On the other hand, this application also provides an intelligent obstacle avoidance system for AMR robot operation based on multi-view vision, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the intelligent obstacle avoidance method for AMR robot operation based on multi-view vision described above.

[0025] The beneficial effects of this application are as follows:

[0026] This application determines the interpolation density for generating a denser point cloud from a sparse point cloud by assessing the obstacle risk of the AMR robot, as well as the accuracy of the obstacle assessment and the relative movement trend of the point cloud. Simultaneously, it adjusts the current interpolation density based on the consistency of obstacle risk assessment with point clouds in adjacent spatial locations, achieving adaptive control over the amount of interpolation data generated for the denser point cloud. This reduces computational load while improving the accuracy of obstacle analysis, providing a precise obstacle avoidance strategy. Attached Figure Description

[0027] To more clearly illustrate the technical solutions and advantages 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of an intelligent obstacle avoidance method for AMR robot operation based on multi-view vision, provided as an embodiment of this application. Detailed Implementation

[0029] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the intelligent obstacle avoidance method and system for AMR robot operation based on multi-view vision proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0031] Implementation Examples of Intelligent Obstacle Avoidance Methods and Systems for AMR Robots Based on Multi-View Vision:

[0032] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent obstacle avoidance method and system for AMR robot operation based on multi-view vision provided in this application.

[0033] Please see Figure 1 The illustration shows a flowchart of an intelligent obstacle avoidance method and system for AMR robot operation based on multi-view vision, according to an embodiment of this application. The method includes the following steps:

[0034] Step S001: Obtain the feature points and disparity of the point cloud.

[0035] The AMR uses a binocular camera to acquire binocular images; depth calculation is performed based on the images, and stereo matching is an important part of this process. The features of the point cloud are obtained through stereo matching, including the position in the image, disparity, and spatial position.

[0036] Specifically, binocular vision is used to obtain motion images of the robot, which include a left image and a right image. For each frame of the left and right images, the ORB feature point matching algorithm is used to obtain the corresponding feature point sets, and the feature points in the left and right images have a one-to-one correspondence. The disparity of the point cloud is determined by the positional difference of the same feature point in the left and right images.

[0037] Therefore, for each point cloud in the AMR robot, the feature point positions in the left and right images, the time difference of the point cloud, and the spatial position of the point cloud are obtained, which constitute the feature information of the point cloud.

[0038] At this point, the feature information of the point cloud has been obtained.

[0039] Step S002: Filter out abnormal and normal feature points by the motion vectors of feature points, and obtain the significance of disparity changes and the intensity of fluctuations based on the disparity change trend.

[0040] For different image frames, a sparse optical flow algorithm (such as the LK algorithm) is used to obtain the motion direction and magnitude of each feature point in adjacent frame images, which are then represented by motion vectors. This involves obtaining the motion vector for each feature point. The above processing is performed on the feature points in both the left and right images. The motion is then divided into two categories: one where the robot is stationary relative to itself, and the other where it moves relative to the robot. If the AMR robot needs to perform intelligent obstacle avoidance, it's because the robot's motion state is inconsistent with the motion states of other objects, leading to a collision risk. Therefore, intelligent obstacle avoidance is necessary in this situation.

[0041] Therefore, it is necessary to perform anomaly detection on the motion vectors of the motion image. Feature points that are relatively stationary relative to the robot's motion state are recorded as normal feature points and marked with 1; feature points that are relatively moving relative to the robot's motion state are recorded as abnormal feature points and marked with 0. Relative stationary means that the motion vector is a zero vector, and relative moving means that the motion vector is not a zero vector.

[0042] It should be noted that, since the optical flow method is used to analyze the direction of motion, the temporal images of the same camera are matched. Therefore, there is also a temporal matching relationship for the feature point data of the AMR robot; that is, for each point cloud, the feature information of the point cloud from all the collected images is used to form a temporal sequence.

[0043] Since parallax is essentially the projection of three-dimensional environmental information onto a two-dimensional image, obstacle analysis reflects the relative position and motion state of obstacles with respect to the AMR robot. It is based on this feature that the risk of collision with obstacles is quantified.

[0044] For each point cloud, the disparities at all times are sorted sequentially and fitted to a temporal disparity curve. The trend of disparity change within the temporal neighborhood is analyzed. If the disparity increases rapidly, it indicates that the distance is rapidly shortening during the analysis of the temporal frame image, causing the obstacle to approach rapidly in the AMR robot's field of vision. When the disparity is relatively stable, it indicates that the disparity of the left and right temporal images remains basically unchanged during the motion, and the motion of the feature point is basically consistent with the motion of the AMR robot. When the disparity decreases rapidly, it indicates that the robot is moving away from the feature point.

[0045] Therefore, based on the above analysis, the derivative of the temporal disparity curve is calculated, and the derivative of each point cloud is recorded as the disparity change rate. If the derivative remains near 0, it indicates that the motion is stable. If it is greater than 0, it is closer to the three-dimensional spatial position of the point cloud; if it is less than 0, it is farther away from the three-dimensional spatial position of the point cloud.

[0046] Because the closer an AMR robot is to the point cloud in three-dimensional space, the greater the risk of collision with the point cloud, while moving away from the point cloud or maintaining stability will not increase the risk of collision.

[0047] Therefore, the ReLU function is used to map all disparity change rates to obtain the change rate curve; the ReLU function means that disparity change rates less than or equal to 0 are represented as 0, and disparity change rates greater than 0 remain unchanged.

[0048] The mean and standard deviation of all elements in the rate of change curve are calculated and denoted as the significance of disparity change and the intensity of disparity fluctuation, respectively. The significance of disparity change reflects how quickly the AMR changes as it approaches obstacles in the point cloud over time; the intensity of disparity fluctuation is used to represent the stability characteristics of the disparity change rate.

[0049] Thus, the intensity of parallax fluctuations and the significance of parallax changes were obtained.

[0050] Step S003: Obtain directional hazard index based on the change in motion vector direction in the time series, obtain the change significance and fluctuation intensity of optical flow based on the fluctuation of motion vector magnitude, combine the two to obtain spatial position consistency, and obtain position hazard index based on the change significance of optical flow parallax and directional hazard index.

[0051] The optical flow of feature points also reflects the motion information of the AMR robot relative to obstacles. Unlike temporal disparity curves, which analyze the motion of an object by comparing left and right views, the optical flow direction of feature points is determined by the temporal motion of the feature points, and disparity does not consider changes in motion direction. Therefore, the optical flow sequence is obtained by sorting the motion vectors of feature points in time.

[0052] Because the direction of optical flow changes continuously during short-term normal motion, for two temporally adjacent optical flows, each has a component on a different coordinate axis. The change value of each component is recorded as the variation; that is, two temporally adjacent optical flows have a variation in each component. The standard deviation of all variations in each component in each time series is calculated as the optical flow fluctuation of that component. The sum of the optical flow fluctuations of different components is used as the directional hazard index during the AMR robot's motion. The directional hazard index is obtained through the stability of directional changes; the more frequent the directional changes, the greater the directional hazard index.

[0053] For each optical flow, the membrane length of its motion vector is obtained as the optical flow magnitude, and an optical flow magnitude curve is obtained based on time series. The optical flow magnitude curve has the same function as the disparity curve; an increasing optical flow indicates that the corresponding AMR robot is closer to that 3D spatial position, and vice versa. Therefore, the optical flow magnitude curve and the disparity curve have the same function; thus, the same calculation method as the time difference curve is used to obtain the significance of optical flow changes and the intensity of optical flow fluctuations for the optical flow magnitude curve.

[0054] Therefore, the same analytical method as for disparity curves is used to analyze optical flow curves. Since disparity curves are obtained by comparing left and right views, while optical flow curves are obtained by matching optical flow through temporal feature points, the two can mutually verify each other. By constructing a vector from the significance of changes and the intensity of fluctuations, a two-dimensional vector is obtained for both disparity and optical flow.

[0055] Spatial location consistency is analyzed based on the difference analysis of two-dimensional vectors.

[0056] The consistency of spatial location is positively correlated with the difference in two-dimensional vectors.

[0057] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large. The specific relationship is determined by the actual application, and this application does not impose any special restrictions.

[0058] Preferably, in this embodiment, the expression for spatial location consistency is:

[0059] , Indicates the significance of changes in parallax. This indicates the intensity of parallax fluctuations. Indicating the significance of changes in optical flow, Indicates the wave intensity of optical flow. Represents a linear normalization function. This indicates the spatial consistency of feature points. Spatial consistency reflects the accuracy of the location analysis; the smaller the value, the more consistent the locations.

[0060] The directional hazard index judges solely by changes in the direction of optical flow. However, this directional analysis alone presents the following problem: because it doesn't consider changes in optical flow intensity, even when the robot approaches an obstacle in a straight line, its orientation towards the obstacle may not change significantly, yet the hazard remains high. Therefore, a time-series analysis combining optical flow magnitude and parallax magnitude is necessary. Higher parallax / optical flow changes indicate a higher risk at that 3D spatial location.

[0061] Therefore, location hazard indicators are determined based on the significance of changes in optical flow and parallax, as well as directional hazard indicators.

[0062] The location hazard index is positively correlated with the significance of changes in optical flow and parallax, and with the orientation hazard index.

[0063] Preferably, in this embodiment, the expression for the location hazard index is:

[0064] , Indicates the significance of changes in parallax. Indicating the significance of changes in optical flow, Indicates directional danger indicators. Represents a linear normalization function. This indicates the locational hazard index of a feature point.

[0065] Thus, based on the above steps, the location hazard index and spatial location consistency of each feature point have been obtained.

[0066] Step S004: Determine the difference density through location hazard index and spatial location consistency, and obtain the corrected difference density through parallax fluctuation and frequency correction of different feature points.

[0067] If the location hazard index of the sparse point cloud is low and the spatial consistency is strong, it indicates that the current sparse point cloud enables the AMR robot to accurately determine the presence of anomalies at that location. In this case, the interpolation density of the sparse point cloud is relatively low. Conversely, if the location hazard index of the sparse point cloud is high, but the spatial consistency is poor (i.e., low accuracy), it means that the current hazard assessment may lead the AMR robot to make incorrect obstacle avoidance strategies due to insufficient data. In this case, a higher interpolation density should be applied to ensure the accuracy and reliability of the AMR robot's obstacle avoidance information.

[0068] Therefore, the interpolation density of feature points is determined based on the location hazard index and spatial location consistency of feature points.

[0069] The interpolation density is positively correlated with the location hazard index and negatively correlated with spatial location consistency.

[0070] It should be noted that negative correlation means that when one variable increases, the other variable decreases accordingly, and the two variables change in opposite directions. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small. The specific relationship is determined by practical application, and this application does not impose any special restrictions.

[0071] Preferably, in this embodiment, the expression for the interpolation density is:

[0072] , This indicates the consistency of the spatial location of feature points. Indicators representing the locational hazard of feature points. This represents the interpolation density of feature points.

[0073] However, the above analysis faces the following problems in practice. In order to ensure the timeliness of information acquisition by AMR robots, sparse point clouds are often used for interpolation (to create a more detailed grid) to generate dense point clouds. However, if the sparse point cloud is insufficient to reflect the high danger index, and there is actually a real risk of robot collision at that location, the interpolation density will be insufficient, which will lead to an incorrect judgment of the safety of the surrounding environment and thus a collision.

[0074] At this point, risk assessment is aided by marking feature points, as abnormal feature points are the cause of collisions during AMR robot movement. It should be noted that label anomalies can be categorized into two types: optical flow vectors that are larger or smaller than those of normal labels. The distinction between these two types has already been analyzed using optical flow magnitude curves to avoid misidentifying collision risks due to actual distance. Therefore, the frequency of anomalies and corresponding disparity fluctuations in the time series are statistically analyzed to determine feature consistency.

[0075] If the difference between the abnormal frequency and the normal label frequency is more significant, and the parallax fluctuation is greater, it indicates that the current temporal feature judgment of the feature point has a high consistency, and thus the accuracy of the risk judgment of the feature point is higher.

[0076] Therefore, feature consistency is determined based on the fluctuation intensity of feature point disparity and the difference between abnormal and normal frequencies. The abnormal frequency is the proportion of abnormal feature points among all feature points, and the normal frequency is the proportion of normal feature points among all feature points.

[0077] The feature consistency is positively correlated with the difference in fluctuation intensity, abnormal frequency, and normal frequency of feature point disparity.

[0078] Preferably, in this embodiment, the expression for feature consistency is:

[0079] , Indicates normal frequency. Indicates abnormal frequency. This indicates the intensity of parallax fluctuations. Represents a linear normalization function. This indicates the consistency of features among feature points.

[0080] If the feature consistency is high, it means that the information obtained by the current AMR robot through the camera is sufficient to accurately determine the obstacle avoidance strategy, and a large interpolation density is not required.

[0081] Therefore, the interpolation density of feature points is corrected based on feature consistency to obtain the corrected interpolation density.

[0082] The corrected interpolation density is positively correlated with the interpolation density and negatively correlated with feature consistency.

[0083] Preferably, in this embodiment, the expression for correcting the interpolation density is:

[0084] , Represents the interpolation density of feature points. This indicates the feature consistency of feature points. This represents the corrected interpolation density of the feature points.

[0085] Thus, the corrected interpolation density of the feature points has been obtained.

[0086] Step S005: Based on the fluctuation of the hazard index, the spatial interpolation density is obtained by processing the correction difference density, and the density difference is used to complete obstacle avoidance.

[0087] The interpolation density information of feature points (in three-dimensional space) in the binocular images of the AMR robot was obtained using the above method. The three-dimensional spatial position of each feature point and its corresponding interpolation density were determined. To ensure the effectiveness of the feature point analysis, there are usually a large number of obstacle feature points that pose a collision risk to the robot. Therefore, the spatial neighborhood information of each feature point was obtained, and the fluctuation of the hazard index of the feature point within the neighborhood was calculated. In this embodiment, the KNN algorithm is used. If the fluctuations remain relatively consistent, it indicates that the risk assessment of obstacles within the spatial neighborhood remains consistent, resulting in a lower interpolation density. Conversely, if the fluctuations remain relatively consistent, the interpolation density needs to be increased.

[0088] Therefore, the spatial interpolation density is determined based on the fluctuation of the risk index and the corrected interpolation density.

[0089] The spatial interpolation density is positively correlated with the fluctuation of the hazard index and the corrected interpolation density.

[0090] Preferably, in this embodiment, the expression for the spatial interpolation density is:

[0091] , Represents the corrected interpolation density of feature points. This indicates fluctuations in risk indicators. Represents a linear normalization function. This represents the spatial interpolation density of feature points.

[0092] Density is graded based on spatial interpolation density, and spatial interpolation meshes of varying refinements are constructed based on this density, thereby generating a dense point cloud. This dense point cloud is then used to identify obstacles during the movement of the AMR robot (which may be stationary), and to determine the travel path and obstacle avoidance trajectory, thus achieving intelligent obstacle avoidance for the AMR robot.

[0093] This completes the intelligent obstacle avoidance system for the AMR robot.

[0094] Based on the same inventive concept as the above methods, embodiments of the present invention also provide an intelligent obstacle avoidance system for AMR robot operation based on multi-view vision, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described intelligent obstacle avoidance methods for AMR robot operation based on multi-view vision.

[0095] 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 scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

[0096] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A multi-view vision-based intelligent obstacle avoidance method for AMR robots, characterized in that, The method includes the following steps: Motion images of point clouds, their feature points, and corresponding parallaxes are acquired using the binocular cameras of an AMR robot. Motion vectors of feature points are obtained based on optical flow algorithm, and abnormal and normal feature points are filtered according to motion vectors. The disparity of a point cloud is sorted according to time sequence to obtain disparity time sequence curve. The significance of disparity change and fluctuation intensity are obtained based on the disparity change trend of disparity time sequence curve. The directional hazard index is determined based on the changes in motion vectors of feature points in different directions during the time series; the optical flow magnitude curve is obtained based on the magnitude of the motion vectors in the time series, and the significance of optical flow changes and fluctuation intensity are obtained using the same method as the disparity time series curve; the spatial position consistency is determined based on the differences in the significance of disparity and optical flow changes and fluctuation intensity; and the position hazard index is determined based on the significance of optical flow and disparity changes and the directional hazard index. The interpolation density of feature points is determined based on the location hazard index and spatial location consistency of feature points; feature consistency is determined based on the fluctuation intensity of feature point disparity and the frequency difference between normal and abnormal feature points; the corrected interpolation density is obtained by correcting the interpolation density through feature consistency. The spatial interpolation density is determined based on the fluctuation of the hazard index and the correction interpolation density; the density is graded by spatial interpolation density to generate a dense point cloud to complete intelligent obstacle avoidance.

2. The multi-vision based AMR robot running intelligent obstacle avoidance method according to claim 1, wherein, The motion vector of a normal feature point is a zero vector, while the motion vector of an abnormal feature point is not a zero vector. 3.The multi-vision based AMR robot running intelligent obstacle avoidance method of claim 1, wherein, The method for obtaining the significance of disparity changes and the intensity of disparity fluctuations based on the disparity change trend of the disparity time series curve is as follows: The derivative of each time series is obtained by taking the derivative of the disparity time series curve and denoted as the disparity change rate. The ReLU function is used to map all the disparity change rates to obtain the change rate curve. Calculate the mean and standard deviation of all elements in the rate of change curve, and denote them as the significance of the disparity change and the intensity of the disparity fluctuation, respectively. 4.The multi-vision-based AMR robot operation intelligent obstacle avoidance method of claim 1, wherein, The method for determining the directional hazard index based on the change in the motion vector of feature points in different directions during time series is as follows: The motion vector has a component on each coordinate axis. The change value of each component is recorded as the change amount. The standard deviation of all changes of each component is taken as the optical flow fluctuation of that component. The sum of the optical flow fluctuations of different components is taken as the directional hazard indicator during the movement of the AMR robot. 5.The multi-vision based AMR robot running intelligent obstacle avoidance method of claim 1, wherein, The method for determining spatial location consistency based on the significance of changes in parallax and optical flow and the difference in fluctuation intensity is as follows: The significance of the change and the intensity of the fluctuation are combined into a vector, and two-dimensional vectors are obtained for parallax and optical flow respectively. Spatial position consistency is analyzed based on the difference between the two-dimensional vectors of parallax and optical flow. The spatial position consistency is positively correlated with the difference between the two-dimensional vectors. 6.The multi-vision based AMR robot running intelligent obstacle avoidance method of claim 1, wherein, The location hazard index is positively correlated with the significance of changes in optical flow and parallax, and with the orientation hazard index.

7. The multi-vision based AMR robot intelligent obstacle avoidance method of claim 1, wherein, The interpolation density is positively correlated with the location hazard index and negatively correlated with spatial location consistency. 8.The multi-vision based AMR robot running intelligent obstacle avoidance method of claim 1, wherein, The feature consistency is positively correlated with the fluctuation intensity, abnormal frequency and normal frequency of the feature point disparity, respectively; the abnormal frequency is the proportion of abnormal feature points to all feature points, and the normal frequency is the proportion of normal feature points to all feature points. 9.The multi-vision based AMR robot running intelligent obstacle avoidance method of claim 1, wherein, The spatial interpolation density is positively correlated with the risk index fluctuation and the corrected interpolation density respectively.

10. A multi-vision-based AMR robot running intelligent obstacle avoidance system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor implements the steps of the AMR robot intelligent obstacle avoidance method based on multi-view vision according to any one of claims 1-9 when executing the computer program.

Images