Suction mop falling detection method for cleaning robot, electronic device, and medium

Abstract

The application relates to the technical field of robots, in particular to a cleaning robot suction scraper falling detection method, an electronic device and a medium. According to the cleaning robot suction scraper falling detection method, an image of a suction scraper part of a cleaning robot is collected in real time by using a depth camera to obtain a target suction scraper monitoring picture; coordinate mapping is performed on the target suction scraper monitoring picture to obtain depth point cloud data; wherein the depth point cloud data corresponds to a camera measurement coordinate system; plane transformation is performed on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system to obtain corresponding plane point cloud data; the plane point cloud data is segmented to obtain suction scraper position point cloud corresponding to the suction scraper part; and suction scraper falling detection is performed on the suction scraper position point cloud to obtain a suction scraper falling detection result. In this way, the falling detection of the suction scraper part can be realized, and potential risks can be reduced.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and in particular to a method, electronic device, and medium for detecting suction and peeling detachment in cleaning robots. Background Technology

[0002] With the widespread adoption of intelligent technologies across various industries, cleaning robots are gradually replacing manual labor in the sanitation and maintenance of various locations. In scenarios with high cleaning demands, such as office buildings, hotels, shopping malls, subway stations, airports, and factories, the use of cleaning robots can significantly reduce manpower consumption and improve operational efficiency.

[0003] Existing floor cleaning robots are typically equipped with a squeegee to collect wastewater generated after cleaning. Because the squeegee requires frequent disassembly and maintenance, its structural design cannot be completely locked in place for easy assembly and disassembly. This structural feature creates a potential risk of the squeegee detaching during robot movement. Furthermore, limitations in cost and installation space make it difficult to install dedicated sensors to detect detachment. Therefore, the industry urgently needs a more reliable squeegee detachment detection solution to mitigate this potential risk. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method, electronic device, and medium for detecting the detachment of suction cups in cleaning robots, which can realize the detection of detachment of suction cup components and reduce potential risks.

[0005] The suction cup detachment detection method for a cleaning robot according to the first aspect of this application includes:

[0006] A depth camera is used to capture images of the suction cup component of the cleaning robot in real time, resulting in a monitoring image of the target suction cup.

[0007] Coordinate mapping is performed on the target suction and scraping monitoring screen to obtain depth point cloud data; wherein, the depth point cloud data corresponds to the camera measurement coordinate system;

[0008] A planar transformation is performed on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system, thereby obtaining the corresponding planar point cloud data;

[0009] The suction cup position point cloud corresponding to the suction cup component is obtained by segmenting the planar point cloud data.

[0010] The suction detachment detection was performed on the point cloud at the suction detachment location to obtain the suction detachment detection results.

[0011] According to some embodiments of this application, the step of performing suction detachment detection on the point cloud at the suction detachment location to obtain suction detachment detection results includes:

[0012] Curvature feature analysis is performed on the point cloud of each suction position to determine the curvature distribution feature corresponding to each suction position point cloud.

[0013] A monitoring target point cloud sequence is generated based on the point cloud of the suction position corresponding to each of the target suction monitoring screens.

[0014] By integrating the curvature distribution features of each suction position point cloud in the monitoring target point cloud sequence, a curvature feature sequence is obtained;

[0015] The curvature feature sequence is input into a pre-trained suction cup state detection model for detachment detection to obtain the suction cup detachment detection result.

[0016] According to some embodiments of this application, the step of performing curvature feature analysis on the point cloud at each suction cup location to determine the curvature distribution features corresponding to the point cloud at each suction cup location includes:

[0017] For each of the suction and squeegee locations, point-by-point curvature calculation is performed on the point cloud to obtain the local curvature features corresponding to each detection point.

[0018] Dynamic feature analysis is performed based on the local curvature features corresponding to each detection point in the multiple suction and grabbing location point clouds to determine the corresponding dynamic adaptive threshold for each detection point.

[0019] Based on each of the aforementioned dynamic adaptive thresholds, dynamic difference analysis is performed on the corresponding detection points to divide multiple detection points into salient feature points and ordinary feature points.

[0020] Based on the positional distribution of the salient feature points and the ordinary feature points in the suction location point cloud, the corresponding curvature distribution feature is generated.

[0021] According to some embodiments of this application, the step of performing point-by-point curvature calculation on the point cloud of each suction position to obtain the corresponding local curvature features includes:

[0022] Select a target detection point from the suction location point cloud, and determine the corresponding neighboring detection points of the target detection point;

[0023] Calculate the target point feature value of the target detection point, and the neighbor point feature value of each of the neighbor detection points;

[0024] Curvature features are calculated based on the target point feature value and the neighbor point feature values ​​of each of the neighboring detection points to obtain the local curvature features corresponding to the target detection point.

[0025] The target detection point is reselected from the unselected detection points in the suction position point cloud, and the corresponding neighboring detection points are re-determined. The calculation of the target point feature value and the neighboring point feature value of each of the neighboring detection points is performed until the local curvature feature of each detection point in the suction position point cloud is obtained.

[0026] According to some embodiments of this application, the step of performing suction detachment detection on the point cloud at the suction detachment location to obtain suction detachment detection results includes:

[0027] Obtain the point cloud height of each detection point in the point cloud of the suction cup location;

[0028] Based on the point cloud height, each detection point is divided into hit points and offset points;

[0029] Based on the location point cloud of the suction, the region is rasterized to obtain the corresponding probability raster map; wherein, the probability raster map includes multiple raster units;

[0030] Based on the hit points and blank points in each grid cell, determine the corresponding suction grid cell occupied by the suction component in the probability grid map from multiple grid cells;

[0031] Based on the suction grid unit, suction peeling detachment detection is performed to obtain the suction peeling detachment detection result.

[0032] According to some embodiments of this application, determining the corresponding suction grid cell occupied by the suction component in the probability grid map from a plurality of grid cells based on the hit points and blank points in each of the grid cells includes:

[0033] Based on the number of hit points and the number of blank points in the grid cell, calculate the grid occupancy probability corresponding to the suction grid cell;

[0034] When the grid occupancy probability reaches a preset occupancy probability threshold, the grid cell is determined as the suction grid cell.

[0035] According to some embodiments of this application, the step of performing a planar transformation on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system to obtain corresponding planar point cloud data includes:

[0036] Based on the camera measurement coordinate system and the reference coordinate system, position mapping analysis is performed to obtain the rotation mapping matrix and translation mapping matrix;

[0037] Based on the rotation mapping matrix and the translation mapping matrix, a rigid transformation matrix is ​​constructed;

[0038] For each initial calibration point in the depth point cloud data, the rigid transformation matrix is ​​applied to perform a planar transformation to obtain a detection point that corresponds one-to-one with the initial calibration point; wherein, the initial calibration point is based on the camera measurement coordinate system, and the detection point is based on the reference coordinate system;

[0039] The planar point cloud data is generated based on the detection points in the reference coordinate system that correspond one-to-one with the initial calibration points.

[0040] According to some embodiments of this application, obtaining the suction cup position point cloud corresponding to the suction cup component based on the segmentation of the planar point cloud data includes:

[0041] Obtain the suction cup installation orientation information of the suction cup component in the cleaning robot; wherein, the suction cup installation orientation information is based on the reference coordinate system;

[0042] Based on the suction nozzle installation orientation information, the suction nozzle discrimination area corresponding to the suction nozzle component is delineated in the reference coordinate system;

[0043] The planar point cloud data is segmented based on the suction-grabbing discrimination region to obtain the suction-grabbing location point cloud.

[0044] Secondly, embodiments of this application provide an electronic device, including: a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the suction and peeling detection method for a cleaning robot as described in any one of the embodiments of the first aspect of this application.

[0045] Thirdly, embodiments of this application provide a computer-readable storage medium storing a program that is executed by a processor to implement the suction and peeling detection method for a cleaning robot as described in any one of the embodiments of the first aspect of this application.

[0046] The suction cup detachment detection method, electronic device, and medium for cleaning robots according to the embodiments of this application have at least the following beneficial effects:

[0047] According to the suction cup detachment detection method for cleaning robots in this application embodiment, it is necessary to first use a depth camera to acquire images of the suction cup component of the cleaning robot in real time to obtain a target suction cup monitoring image; then, coordinate mapping is performed on the target suction cup monitoring image to obtain depth point cloud data; wherein, the depth point cloud data corresponds to the camera measurement coordinate system; a planar transformation is performed on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system to obtain corresponding planar point cloud data; based on the planar point cloud data, the suction cup position point cloud corresponding to the suction cup component is obtained; suction cup detachment detection is performed on the suction cup position point cloud to obtain the suction cup detachment detection result. In this way, the detachment detection of the suction cup component can be realized, reducing potential risks.

[0048] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0049] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0050] Figure 1 This application provides a schematic flowchart of a suction-capsule detachment detection method for cleaning robots.

[0051] Figure 2 This application provides another schematic flowchart of a suction-capsule detachment detection method for cleaning robots;

[0052] Figure 3 This application provides another schematic flowchart of a suction-capsule detachment detection method for cleaning robots;

[0053] Figure 4 This application provides another schematic flowchart of a suction-capsule detachment detection method for cleaning robots;

[0054] Figure 5 This application provides another schematic flowchart of a suction-capsule detachment detection method for cleaning robots;

[0055] Figure 6 This application provides another schematic flowchart of a suction-capsule detachment detection method for cleaning robots;

[0056] Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0057] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0058] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0059] In the description of this application, it should be understood that the orientation descriptions, such as up, down, left, right, front, and back, are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0060] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0061] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setting," "installation," and "connection" should be interpreted broadly. Those skilled in the art can reasonably determine the specific meaning of the above terms in this application based on the specific content of the technical solution. Furthermore, the identification of specific steps in the following text does not imply a limitation on the order of steps or execution logic. The execution order and logic between each step should be understood and inferred from the content described in the embodiments.

[0062] With the widespread adoption of intelligent technologies across various industries, cleaning robots are gradually replacing manual labor in the sanitation and maintenance of various locations. In scenarios with high cleaning demands, such as office buildings, hotels, shopping malls, subway stations, airports, and factories, the use of cleaning robots can significantly reduce manpower consumption and improve operational efficiency.

[0063] The "squeegee" in a floor cleaning robot is a shorthand for a functional component, referring to the rubber or silicone scraper assembly installed at the rear of the robot close to the ground. Its core function is to collect residual wastewater after the robot has cleaned the floor and suck it into the wastewater tank, thus achieving the drying process after the floor is cleaned.

[0064] Structurally, a vacuum squeegee typically consists of two parts: a flexible scraper and a suction nozzle. The scraper slides close to the ground, collecting the dispersed wastewater into one place; the suction nozzle is connected to the negative pressure generated by a blower, which draws the collected wastewater away from the ground. This design allows the robot to complete continuous scrubbing and water suction operations in one go.

[0065] Existing floor cleaning robots are typically equipped with a suction squeegee to collect wastewater generated after cleaning the floor. Because the suction squeegee needs frequent disassembly and maintenance, its structural design cannot be completely locked in place for easy assembly and disassembly. This structural feature leads to a potential risk of the suction squeegee detaching during robot movement. Furthermore, due to cost and installation space constraints, it is difficult to install dedicated sensors to detect whether the suction squeegee has detached.

[0066] In practical applications, this technical defect can cause a variety of problems.

[0067] First, when the robot operates on rough, uneven ground, the suction squeegee experiences increased impact, raising the probability of it detaching. If the squeegee detaches and the robot continues to operate, it will be unable to properly collect wastewater, causing the cleaning operation to fail.

[0068] Secondly, detached suction cups may be dragged along as the robot moves, colliding with surrounding objects and causing equipment damage or safety hazards.

[0069] Therefore, the industry urgently needs a more usable suction peel detection solution that allows robots to stop operating promptly and notify maintenance personnel when an anomaly is detected, thus avoiding the aforementioned problems.

[0070] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method, electronic device, and medium for detecting the detachment of suction cups in cleaning robots, which can realize the detection of detachment of suction cup components and reduce potential risks.

[0071] Reference Figure 1 The suction cup detachment detection method for cleaning robots according to the embodiments of this application may include:

[0072] Step S101: Use a depth camera to capture images of the suction cup component of the cleaning robot in real time to obtain a monitoring image of the target suction cup.

[0073] Step S102: Perform coordinate mapping on the target suction and scraping monitoring screen to obtain depth point cloud data; wherein, the depth point cloud data corresponds to the camera measurement coordinate system;

[0074] Step S103: Perform a planar transformation on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-built reference coordinate system to obtain the corresponding planar point cloud data.

[0075] Step S104: Obtain the suction cup position point cloud corresponding to the suction cup component based on planar point cloud data segmentation;

[0076] Step S105: Perform suction detachment detection on the point cloud at the suction detachment location to obtain the suction detachment detection result.

[0077] According to the suction cup detachment detection method for cleaning robots in this application embodiment, it is necessary to first use a depth camera to acquire images of the suction cup component of the cleaning robot in real time to obtain a target suction cup monitoring image; then, coordinate mapping is performed on the target suction cup monitoring image to obtain depth point cloud data; wherein, the depth point cloud data corresponds to the camera measurement coordinate system; a planar transformation is performed on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system to obtain corresponding planar point cloud data; based on the planar point cloud data, the suction cup position point cloud corresponding to the suction cup component is obtained; suction cup detachment detection is performed on the suction cup position point cloud to obtain the suction cup detachment detection result. In this way, the detachment detection of the suction cup component can be realized, reducing potential risks.

[0078] In step S101 of some embodiments, a depth camera is used to capture images of the suction device of the cleaning robot in real time to obtain a monitoring image of the target suction device.

[0079] It should be noted that the technical essence of this application's embodiments lies in using a depth camera to perform active optical observation of the suction cup component, transforming the detachment detection problem into a real-time 3D geometric analysis task. In practice, the depth camera often needs to be installed facing the suction cup. Compared to adding dedicated devices for detecting suction cup detachment, such as microswitches or pull-cord sensors, the depth camera, as a general-purpose sensing module, can simultaneously handle suction cup monitoring and other auxiliary functions. This installation method introduces the depth camera module itself, but avoids complex mechanical structure modifications; only the angle of the fixing bracket needs to be adjusted, keeping the overall cost and power consumption within a controllable range.

[0080] The unique aspect of using depth cameras lies in introducing optical 3D measurement technology into the field of component loosening detection. Depth cameras output not only 2D images, but also the true 3D coordinates of each pixel. This upgrades the detection logic from a discrete event judgment of "whether a physical switch is triggered" to a continuous geometric evaluation of "whether the spatial occupancy state matches expectations." This shift requires cameras with stable optical performance, capable of resisting interference from ground spills and changes in ambient lighting (such as reflections from shopping mall glass curtain walls). In actual deployment, waterproof covers and filters are often added, and noise reduction preprocessing is performed at the algorithm level. More importantly, this step lays the foundation for the data accuracy of all subsequent processing stages: if the camera resolution is too low, the point cloud at the edge of the suction device will be sparse, and subsequent probability grid statistics may miss detections due to insufficient effective grids; if the camera angle is off, the height characteristics of the suction device in the reference coordinate system will not be obvious, affecting the accuracy of salient / common feature point (i.e., hit / miss points) classification. Therefore, the embodiments of this application are the starting point for the performance of the algorithm system. All subsequent coordinate transformation thresholds, probability update parameters, and grid counting logic need to be calibrated based on the camera's installation position and imaging characteristics, forming a strongly coupled design closed loop.

[0081] It is worth noting that "real-time data acquisition" is the core performance constraint of this application's embodiments, determining the detection response time and safety value. The detachment of the suction cap is a sudden mechanical failure; if there is a significant delay in data acquisition and processing, the robot may continue to operate for several meters after the anomaly occurs, leading to sewage spread, component damage, or collision accidents.

[0082] Some embodiments provide a parameterized method for calculating the camera mounting angle, transforming empirical, subjective adjustments into precise solutions based on geometric measurements. The formula for calculating the camera mounting angle is: θ = 90 - arctan(w / h). The physical meaning of this formula is to ensure that the optical axis of the depth camera precisely covers the maximum lateral range of the suction cup component. Here, θ represents the camera mounting angle, w represents the width of the suction cup protruding from the side of the robot, h is the camera mounting height, and arctan(w / h) calculates the angle between the line of sight from the camera position to the outermost edge of the suction cup and the vertical direction. Subtracting this value from 90 degrees yields the optimal downward tilt angle of the camera's optical axis relative to the horizontal plane. This design avoids the risk of introducing irrelevant background interference due to an excessively large camera field of view, or missing the suction cup edge due to an excessively small field of view, achieving precise matching between the detection area and the camera's observation range. The special feature of this calculation method lies in its reusability and adaptability. Traditional installation and adjustments rely on repeated manual trial and error, requiring time-consuming calibration for different models. This solution formula abstracts the installation process into two measurable engineering parameters, w and h. Once the robot's structural dimensions are determined, the theoretically optimal angle can be calculated in one go. It should be understood that when the suction cup width changes due to application scenarios (such as replacing different specifications of scraper strips), only the value of w needs to be remeasured and substituted into the formula, eliminating the need for tedious on-site adjustments. This parameter-driven methodology upgrades camera installation from a craftsman's operation to a standardized process, reducing deployment costs and technical barriers, making it particularly suitable for mass-produced models or maintenance scenarios requiring frequent parts replacement.

[0083] In some specific embodiments, depth cameras are used to capture images of the suction cup component of the cleaning robot in real time. Specifically, two depth cameras can be installed on either side of the cleaning robot. The rationale for this technique stems from the complexity of the physical shape of the suction cup component and the operating environment. As a long, narrow component spanning the rear of the robot, the suction cup's length is typically close to or exceeds the robot's width. A single depth camera, limited by its field of view and installation position, cannot fully cover the entire span of the suction cup while maintaining sufficient image resolution. If a single camera is installed at the top center of the robot, it can barely cover the entire width, but the edges of the suction cup will be at the edge of the field of view, resulting in severe image distortion and reduced depth measurement accuracy, making it impossible to reliably detect edge detachment or loosening. Installing two depth cameras on either side of the robot, with their optical axes obliquely aligned with the middle section of the suction cup, allows for the capture of high-resolution local depth images from a direct viewing angle within their respective fields of view, ensuring that structural details along the entire length of the suction cup are clearly presented. This distributed sensing architecture not only resolves the contradiction between field of view coverage and measurement accuracy, but also enhances the fault tolerance of detection through spatial redundancy design. When one camera is splashed by sewage or temporarily blocked by debris on the ground, the other camera can still maintain normal monitoring, avoiding the paralysis of the entire detection system due to a single point of failure.

[0084] In step S102 of some embodiments, coordinate mapping is performed on the target suction and grabbing monitoring screen to obtain depth point cloud data; wherein, the depth point cloud data corresponds to the camera measurement coordinate system;

[0085] It should be noted that the core function of this application's embodiments is to transform two-dimensional depth images into three-dimensional point cloud data with a real geometric scale, completing the dimensional leap from image space to physical space. The raw monitoring images captured by the depth camera are essentially a combination of pixel matrices and depth values, with each pixel recording the distance information of the corresponding scene point. Coordinate mapping establishes the correspondence between pixel coordinates (u,v) and three-dimensional coordinates (x,y,z) through the camera's intrinsic parameter matrix, using a pinhole camera model to transform discrete distance measurements into a set of geometric points in continuous space. This process is not a simple data format conversion, but a crucial step in introducing real-world scale information, giving physical meaning to all subsequent geometric analyses (such as height segmentation and plane fitting). Without coordinate mapping, the depth image can only be treated as a grayscale image, making it impossible to extract the spatial height difference between the suction cup and the ground, and the entire detection scheme will lose its three-dimensional perception capability.

[0086] It is important to clarify that the camera measurement coordinate system is a local coordinate system with the camera's optical center as the origin and the optical axis as the z-axis, and its location is fixed depending on the camera's installation position. This statement provides a clear starting point for the planar transformation in step S103. If the coordinate system is not clearly defined in this step, subsequent transformations from the camera measurement coordinate system to the reference coordinate system will lack a mathematically valid transformation object, making it impossible to establish a mapping relationship for the homogeneous transformation matrix. The introduction of the camera measurement coordinate system unifies the independent observation data from each camera, which may change with the robot's movement, as "the original data to be transformed," laying the groundwork for eliminating coordinate heterogeneity. At the same time, this clear distinction also implies the limitations of single-camera observations: point cloud data in the camera measurement coordinate system lacks cross-view consistency and must rely on subsequent transformations to be used for cross-temporal or multi-camera fusion analysis.

[0087] The special feature of this application embodiment is that it realizes semantic dimensionality reduction from dense pixels to sparse point clouds, and at the same time completes the paradigm transformation of data from "sensor native representation" to "geometric universal representation", with each pixel of the depth image carrying depth information.

[0088] In some specific embodiments, depth image data is large and contains a significant amount of invalid background. Based on this coordinate mapping, invalid points can be quickly eliminated through depth thresholding and spatial cropping, retaining only the set of valid geometric points near the target area. This drastically reduces the data volume but increases the information density. This dimensionality reduction is not information loss, but rather a precise focus on the detection target, allowing subsequent algorithmic processing resources to concentrate on key areas.

[0089] Furthermore, the embodiments of this application inherently rely heavily on the accuracy of camera calibration. Coordinate mapping depends on pre-calibrated focal length and principal point coordinates, and the accuracy of these intrinsic parameters directly determines the geometric accuracy of the point cloud. If the calibration error is large, the generated point cloud will exhibit systematic distortion, causing subsequent height segmentation thresholds to fail. Therefore, the embodiments of this application are not only algorithmic processing but also an interface layer between hardware characteristics and software algorithms, encapsulating the camera's inherent optical imaging model into a standardized geometric output.

[0090] In some more specific embodiments, if pixel coordinates, i.e., the position of each point in the image, are represented as (u, v); and the depth value, i.e., the distance from the pixel to the camera, is represented as z; then the three-dimensional points in the depth point cloud data obtained by coordinate mapping are represented as (x, y, z). For target-grabbing monitoring images, coordinate mapping is performed to obtain depth point cloud data. The specific coordinate mapping formula can be found in the following formula:

[0091] ;

[0092] in, , These are the camera's focal lengths, corresponding to the x-axis and y-axis directions of the image, respectively; , These are the principal point coordinates, i.e., the position of the optical center on the image plane.

[0093] It should be noted that the coordinate mapping formula is based on a pinhole camera model, which assumes that the imaging process of a camera is similar to light passing through a small hole (i.e., the camera's lens) to form an inverted real image on an image sensor. In this model, every point in the image can be connected to the optical center of the camera by a straight line, which extends to a point in the scene, i.e., an object point in three-dimensional space. Coordinate mapping is a step in suction cup delamination detection; it not only converts image data into a three-dimensional point cloud that can be used for analysis but also provides a foundation for subsequent detection steps. The implementation of this step needs to consider factors such as accuracy, real-time performance, and robustness to ensure the effectiveness and reliability of the entire detection process.

[0094] In some embodiments, step S103 involves performing a planar transformation on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-built reference coordinate system, thereby obtaining the corresponding planar point cloud data.

[0095] It should be noted that the embodiments of this application address the problem of inconsistent coordinate references caused by robot movement and heterogeneous multi-camera setups, establishing the entire detection process on a stable and reliable spatial reference framework. The point cloud data acquired by the depth camera resides in the camera measurement coordinate system, which is fixed to the robot body with the camera's optical center as its origin. When the robot moves up and down slopes, over speed bumps, or makes sharp turns in scenarios such as office buildings and shopping malls, its posture continuously changes, causing the camera measurement coordinate system to drift in real-time relative to the actual position of the suction cup. If subsequent segmentation and judgment are directly based on this coordinate system, the position of the suction cup in the coordinate system will shift by centimeters due to changes in the robot's pitch angle, making it impossible to stably align the fixed threshold with the target area, and the detection logic will completely fail. Planar transformation can rotate and translate the point cloud data acquired by each camera at each moment to a pre-constructed reference coordinate system, which is typically based on the robot's body center or inertial coordinate system, maintaining absolute spatial stability. The transformed point cloud data eliminates motion interference, ensuring that the suction cup remains in a fixed position range in the coordinate system regardless of the operating conditions, thus creating a unified spatial premise for subsequent segmentation and probability statistics.

[0096] It is worth noting that the "planar point cloud data" in this embodiment does not refer to compressing a three-dimensional point cloud into a two-dimensional plane, but rather emphasizes that the transformed point cloud has been aligned to the horizontal reference plane defined in the reference coordinate system. After rotation and translation operations, the originally tilted point cloud is corrected to a standard posture parallel or perpendicular to the ground. At this time, the ground area no longer appears as a tilted plane, but is approximately horizontally distributed, and protruding obstacles such as suction cups appear as local height abrupt changes. This "planarization" process is not a dimensionality reduction, but a normalization of geometric posture, enabling the subsequent step S104 to reliably distinguish suction cups from the ground based on a fixed height threshold, avoiding misclassification caused by posture deviation. This embodiment retains complete three-dimensional geometric information, only unifying the spatial representation benchmark.

[0097] The unique feature of this embodiment is that it serves as a crucial spatial normalization hub in the entire logical chain. Without this transformation, the spatial reference between preceding and following steps is broken, and the detection process cannot be closed. Step S102 outputs the original point cloud in the camera measurement coordinate system, while step S104 requires segmentation based on a fixed geometric threshold; the two cannot be directly interfacing. A planar transformation is performed on the depth point cloud data, acting as a bridge to convert the observation data in the camera measurement coordinate system into standard data in the reference coordinate system, enabling the point cloud data segmentation to stably apply to the suction area.

[0098] Furthermore, for scenarios involving image acquisition using two depth cameras, this application embodiment achieves heterogeneous fusion of data from the two depth cameras on both sides. Each camera has its own independent camera measurement coordinate system; without unified transformation, it is impossible to use the same set of segmentation parameters or merge statistical results in the probabilistic raster map. Therefore, planar transformation is not only about eliminating motion interference but also a core operation for spatial alignment of multi-sensor data, embodying the inventive concept of integrating hardware with algorithms and managing distributed sensing resources with a unified coordinate system.

[0099] Reference Figure 2 According to some embodiments of this application, step S103 performs a planar transformation on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-built reference coordinate system to obtain the corresponding planar point cloud data, which may include:

[0100] Step S201: Perform position mapping analysis based on the camera measurement coordinate system and the reference coordinate system to obtain the rotation mapping matrix and translation mapping matrix;

[0101] Step S202: Construct a rigid transformation matrix based on the rotation mapping matrix and the translation mapping matrix;

[0102] Step S203: Apply a rigid transformation matrix to each initial calibration point in the depth point cloud data to perform a planar transformation, and obtain detection points that correspond one-to-one with the initial calibration points; wherein, the initial calibration points are based on the camera measurement coordinate system, and the detection points are based on the reference coordinate system.

[0103] Step S204: Generate planar point cloud data based on the detection points in the reference coordinate system that correspond one-to-one with the initial calibration points.

[0104] In some embodiments, step S201 involves performing position mapping analysis based on the camera measurement coordinate system and the reference coordinate system to obtain a rotation mapping matrix and a translation mapping matrix.

[0105] It should be noted that position mapping resolution does not read pre-stored static parameters, but rather dynamically calculates the rotation matrix R and translation vector t of the camera measurement coordinate system relative to the reference coordinate system at the current moment based on the robot's real-time pose sensors (such as inertial measurement units or wheeled odometry).

[0106] It should be understood that during actual operation, the cleaning robot's posture continuously changes due to road bumps, turns, or inclines and declines, causing the camera coordinate system to shift with each frame. Therefore, the analysis process must be performed frame by frame to ensure that the transformation parameters are strictly synchronized with the physical state. The rotation matrix R and translation vector t obtained from the analysis are the basis for subsequent calculations.

[0107] In some embodiments, step S202 involves constructing a rigid transformation matrix based on the rotation mapping matrix and the translation mapping matrix;

[0108] It should be noted that the rigid transformation matrix constructed based on the rotation matrix and the translation vector can be expressed as:

[0109] ;

[0110] The homogeneous transformation matrix T is a 4×4 homogeneous matrix that combines rotation and translation operations. The 3×3 matrix R represents the rotation matrix, and the 3×1 column vector t represents the translation vector. The last row is [0 0 0 1], which is the standard form of a homogeneous coordinate system. The rotation matrix R describes how many angles the point cloud data needs to be rotated around the x, y, and z axes to align with the reference coordinate system, while the translation vector t describes how much distance needs to be moved along the x, y, and z axes after the rotation.

[0111] In this way, the mathematical encapsulation and computational optimization of rotation and translation operations are achieved. The special feature of the rigid transformation matrix is ​​its completeness and efficiency. It not only includes all six degrees of freedom of rigid body motion (three-axis rotation + three-axis translation), but also allows the transformation to be completed in a single step matrix multiplication through homogeneous coordinates, avoiding the step-by-step computational overhead of rotating first and then translating.

[0112] In step S203 of some embodiments, a rigid transformation matrix is ​​applied to each initial calibration point in the depth point cloud data to perform a planar transformation, thereby obtaining a detection point that corresponds one-to-one with the initial calibration point; wherein, the initial calibration point is based on the camera measurement coordinate system, and the detection point is based on the reference coordinate system.

[0113] In some embodiments, step S204 generates planar point cloud data based on the detection points in the reference coordinate system that correspond one-to-one with the initial calibration points.

[0114] It should be noted that the initial calibration point is represented as follows: For each initial calibration point P in the depth point cloud, a rigid transformation matrix T is applied to transform the point to obtain the detection point P′ in the reference coordinate system, which is represented as:

[0115] ;

[0116] The coordinates of the output point P′ are the result of rotation and translation.

[0117] It should be understood that applying a rigid transformation matrix to perform planar transformation on each initial calibration point in the depth point cloud data is a computationally intensive step. A depth point cloud can contain thousands to tens of thousands of points, and performing matrix multiplication point-by-point can take several milliseconds on an embedded platform. The unique feature of this embodiment is its ability to transform the point cloud data from the camera coordinate system to a stable reference coordinate system, which is crucial for accurate component monitoring by robots in dynamic environments. This transformation eliminates coordinate system changes caused by robot movement, making the state monitoring of the suction components more accurate and reliable.

[0118] Furthermore, planar point cloud data can be generated based on the detection points in the reference coordinate system that correspond one-to-one with the initial calibration points.

[0119] In some embodiments, step S104 involves obtaining the suction cup position point cloud corresponding to the suction cup component based on planar point cloud data segmentation.

[0120] It should be noted that although the point cloud data after planar transformation has been unified to the reference coordinate system, it still contains multiple geometric information such as the ground, debris, and walls. Directly judging its state would involve a huge amount of computation, and background noise would seriously interfere with the reliability of the results. Step S104 uses spatial geometric segmentation to retain only the point cloud subset corresponding to the suction cup component, thereby focusing the analysis scope.

[0121] It's important to note that this segmentation operation doesn't rely on complex deep learning models or time-consuming clustering algorithms. Instead, it's based on the relatively fixed installation position of the suction cup in the reference coordinate system, using preset coordinate thresholds for rapid filtering. For example, a threshold can be set in the x-direction to limit the suction cup's length, in the y-direction to limit its lateral span, and in the z-direction to distinguish its height above the ground, thus "cropping" the target from the massive point cloud. This method has extremely low computational overhead, yet it can achieve orders-of-magnitude compression of data in sub-millisecond time, reducing the computational burden on subsequent processing modules to only a few thousand points instead of hundreds of thousands, creating a fundamental efficiency prerequisite for real-time detection.

[0122] Reference Figure 3 According to some embodiments of this application, step S104, which obtains the suction cup position point cloud corresponding to the suction cup component based on planar point cloud data segmentation, may include:

[0123] Step S301: Obtain the suction cup installation orientation information of the suction cup component in the cleaning robot; wherein, the suction cup installation orientation information is based on the reference coordinate system;

[0124] Step S302: Based on the suction nozzle installation orientation information, delineate the suction nozzle discrimination area corresponding to the suction nozzle component in the reference coordinate system;

[0125] Step S303: Segment the planar point cloud data based on the suction and pick-up discrimination region to obtain the suction and pick-up position point cloud.

[0126] In step S301 of some embodiments, the suction cup component is used to obtain the suction cup installation orientation information in the cleaning robot; wherein, the suction cup installation orientation information is based on a reference coordinate system.

[0127] It is important to note that obtaining the installation orientation information of the suction cup component within the cleaning robot is crucial, as it provides the necessary spatial positioning reference for subsequent point cloud segmentation. The suction cup installation orientation information is based on a reference coordinate system, meaning it already considers the robot's global position and orientation, thus ensuring consistency and accuracy. In practice, this information may originate from the robot's design drawings, manufacturer-provided technical specifications, or be obtained through on-site measurements. Regardless of the source, this information needs to be accurate enough to precisely identify the suction cup component's position in the point cloud data.

[0128] In step S302 of some embodiments, the suction pick discrimination area corresponding to the suction pick component is delineated in the reference coordinate system according to the suction pick installation orientation information;

[0129] It's important to note that, based on the suction cup's installation orientation information, a discrimination region corresponding to the suction cup component is delineated in the reference coordinate system. This step essentially defines a region in three-dimensional space, which is expected to contain all the point cloud data of the suction cup component. This region is typically defined based on the suction cup component's geometry and dimensions, as well as its installation position on the robot. For example, if the suction cup component is a long, narrow structure, the discrimination region might be a cuboid or a cylinder. The precise definition of this region is crucial for subsequent point cloud segmentation, as it directly affects the accuracy of the segmentation results.

[0130] In some embodiments, step S303 involves segmenting the planar point cloud data based on the suction-squeeze discrimination region to obtain the suction-squeeze position point cloud.

[0131] It should be noted that the planar point cloud data is segmented based on the suction cup discrimination region to obtain the suction cup location point cloud. This step is the core of the entire process. It uses the information provided in the first two steps to filter out the point clouds belonging to the suction cup component from the entire point cloud dataset. This process may involve spatial geometric calculations, such as calculating the distance from a point to a plane and the shortest distance from a point to a line segment, to ensure that only point clouds located within the discrimination region are selected. The segmented point cloud data will be used for subsequent suction cup detachment detection, so its quality directly determines the reliability of the detection results.

[0132] The unique feature of this application's embodiments lies in transforming the spatial segmentation problem into a filtering problem based on predefined regions. This simplifies the algorithm's complexity and improves processing speed. Through predefined discrimination regions, this application's embodiments can quickly and accurately extract the desired portion from large amounts of point cloud data without requiring complex analysis of each point. This method not only improves efficiency but also reduces computational resource consumption, making the entire detection process more suitable for real-time operation in resource-constrained embedded systems.

[0133] Furthermore, the design of this application embodiment reflects a deep understanding of real-world application scenarios. In the actual operating environment of a cleaning robot, the suction cup component may be affected by various factors, such as uneven ground and robot vibration. By precisely defining the suction cup discrimination area in the reference coordinate system, this application embodiment can adapt to these changes, ensuring accurate detection of the suction cup's state even in complex environments. This design not only improves detection accuracy but also enhances the system's robustness, enabling stable operation under various working conditions.

[0134] The unique feature of this application's embodiment lies in achieving a semantic bridge from general geometric data to specific target data. The input planar point cloud data is an objective mapping of physical space, without distinguishing object categories; while the output suction cup location point cloud has been given a clear semantic label, i.e., the part to be detected. More importantly, this segmentation process is integrated into the overall method: it relies on the stability of the reference coordinate system provided in step S103. If the point cloud has not undergone planar transformation, the segmentation threshold will fail due to changes in robot posture; furthermore, it determines the input quality of step S105. If inaccurate segmentation introduces ground points, the subsequent probability grid will misclassify the ground as suction cups, leading to false alarms. Therefore, the suction cup location point cloud is both the output of step S104 and the sole processing object of step S105, and its data purity and geometric accuracy become the bottleneck of the entire detection system's accuracy.

[0135] In step S105 of some embodiments, suction detachment detection is performed on the suction detachment point cloud to obtain suction detachment detection results.

[0136] It should be noted that, after the aforementioned data acquisition, coordinate mapping, planar transformation, and spatial segmentation, the embodiments of this application have refined the original depth image into a subset of point clouds containing only the suction-clip component. Step S105 performs detachment detection based on this, the core task of which is to transform the point cloud distribution in geometric space into a state judgment result.

[0137] It is worth noting that methods for detecting suction cup detachment from point clouds at suction cup locations can include at least curvature feature detection and probabilistic grid map detection. Curvature feature detection relies on prior knowledge of the suction cup component's geometry. It identifies edge and corner features by calculating the local curvature values ​​of each point in the point cloud, dynamically constructing a curvature distribution model. When the real-time curvature features deviate from a preset normal pattern exceeding a threshold, a detachment alarm is triggered. Probabilistic grid map detection projects the point cloud onto a two-dimensional grid, iteratively updating the occupancy probability using a Bayesian framework, and determining existence by counting the number of high-probability grid cells.

[0138] In some embodiments, curvature feature detection method can be applied alone for suction cup detachment detection, probabilistic grid map method can be applied alone for suction cup detachment detection, or curvature feature detection method and probabilistic grid map method can be applied together.

[0139] It is worth noting that in the embodiment that uses both curvature feature detection and probabilistic grid mapping methods, although the two methods are independent in processing logic, they can be integrated at the decision-making level. For example, the confidence level is highest when both methods simultaneously determine that the suction cup has detached; when the results conflict, secondary verification or a suspected state can be reported, reducing the risk of false alarms caused by sudden changes in ambient lighting or ground material due to a single algorithm. This dual-track parallel design can cope with the complex and variable operating environment of the cleaning robot. A single visual feature may fail due to sewage reflection, uneven ground, or interference from suction cup residue, and the complementarity of multiple methods can effectively improve the robustness of the system.

[0140] Reference Figure 4 According to some embodiments of this application, step S105, which performs suction cap detachment detection on the suction cap location point cloud to obtain suction cap detachment detection results, may include:

[0141] Step S401: Perform curvature feature analysis on the point cloud at each suction cup position to determine the curvature distribution features corresponding to the point cloud at each suction cup position.

[0142] Step S402: Generate a monitoring target point cloud sequence based on the point cloud of the suction position corresponding to each target suction monitoring screen;

[0143] Step S403: Integrate the curvature distribution features of the point cloud corresponding to each suction position in the monitoring target point cloud sequence to obtain the curvature feature sequence;

[0144] Step S404: Input the curvature feature sequence into the pre-trained suction cup state detection model for detachment detection to obtain suction cup detachment detection results.

[0145] It should be noted that step S105 plays a decision-making role in the suction cup detachment detection method. It determines whether the suction cup has detached based on the curvature characteristics of the point cloud at the suction cup location. The special feature of this step is that it combines geometric feature analysis with a machine learning model to achieve intelligent recognition and detection of the suction cup's state.

[0146] In some embodiments, step S401 involves performing curvature feature analysis on the point cloud at each suction cup position to determine the curvature distribution features corresponding to the point cloud at each suction cup position.

[0147] It should be noted that, as the front end of feature extraction, curvature feature analysis is performed on the point cloud of the suction cup position in each frame. Curvature, as a geometric invariant describing the degree of local bending of a surface, can effectively capture structurally abrupt regions such as the edges and corners of the suction cup scraper. The calculation process typically involves constructing a covariance matrix based on the point neighborhood, obtaining the principal curvature through eigenvalue decomposition, and then statistically analyzing the distribution characteristics of curvature such as mean, variance, and histogram. The advantage of this feature representation lies in its robustness; curvature is invariant to translation and rotation. Even if the robot experiences slight bumps that cause the overall point cloud to shift, the geometric structural characteristics of the suction cup remain stable, providing a reliable input foundation for subsequent models. For slender, rigid components like suction cups, the curvature distribution under normal conditions exhibits a regular bimodal pattern (high curvature at the scraper edge and low curvature in the middle plane). However, detachment or loosening leads to discretization of the curvature distribution. This difference provides a learnable pattern for the model to distinguish states.

[0148] Reference Figure 5 According to some embodiments of this application, step S401 performs curvature feature analysis on the point cloud at each suction cup location to determine the curvature distribution features corresponding to the point cloud at each suction cup location, which may include:

[0149] Step S501: Perform point-by-point curvature calculation on the point cloud of each suction point to obtain the local curvature features corresponding to each detection point.

[0150] Step S502: Dynamic feature analysis is performed based on the local curvature features corresponding to each detection point in the point cloud of multiple suction and squeegee locations to determine the corresponding dynamic adaptive threshold for each detection point.

[0151] Step S503: Based on each dynamic adaptive threshold, perform dynamic difference analysis on the corresponding detection points to divide multiple detection points into salient feature points and ordinary feature points.

[0152] Step S504: Generate the corresponding curvature distribution features based on the positional distribution of significant and ordinary feature points in the suction location point cloud.

[0153] It should be noted that curvature feature extraction is used to extract local shape features from geometric data. This method can help identify and distinguish different geometric structures in point clouds, such as edges, corners, and planar regions. The following is a detailed analysis of the curvature feature extraction method described in the figure:

[0154] In step S501 of some embodiments, the curvature of the point cloud at each suction point is calculated point by point to obtain the local curvature features corresponding to each detection point.

[0155] It's important to note that the point-by-point curvature calculation, which constructs the microscopic representation foundation of the point cloud's geometric features, presents unique accuracy requirements and computational challenges in suction cup detection scenarios. Curvature calculation typically involves constructing a local covariance matrix based on each detection point and its K nearest neighbors, obtaining the principal curvature components through eigenvalue decomposition. Mathematically, this quantifies the degree of curvature of the tiny surface at that point. For suction cup components, the curvature values ​​at the scraper edges and support corners are significantly higher than in the central planar region; these local geometric abrupt changes constitute the structural fingerprint of the suction cup. The unique aspect of point-by-point calculation is that it endows each point in the point cloud with independent feature description capabilities, rather than treating the point cloud as a homogeneous whole. This means that even if the suction cup only exhibits localized cracks or edge wear, the corresponding curvature anomalies can be accurately captured, providing a fine-grained data foundation for early fault warning.

[0156] In some embodiments, due to the computational load introduced by this calculation, searching the neighborhood and decomposing the matrix point cloud containing thousands of points may take tens of milliseconds on an embedded platform. Therefore, in actual implementations, downsampling or spatial hashing is often used to accelerate neighborhood queries, and GPU parallel computing is used to compress the processing time of a single frame to less than 5 milliseconds to meet real-time requirements.

[0157] According to some embodiments of this application, step S501, which performs point-by-point curvature calculation on the point cloud of each suction cup location to obtain the corresponding local curvature features, may include:

[0158] Select the target detection point from the suction location point cloud, and determine the corresponding neighboring detection points of the target detection point;

[0159] Calculate the target location feature value of the target detection point, and the neighbor location feature value of each neighbor detection point;

[0160] Curvature features are calculated based on the feature values ​​of the target point and the feature values ​​of the neighboring points of each neighboring detection point to obtain the local curvature features corresponding to the target detection point.

[0161] The process involves reselecting target detection points from the unselected detection points in the suction and shovel position point cloud, re-determining the corresponding neighboring detection points of the target detection point, and returning the target point feature value and the neighboring point feature values ​​of each neighboring detection point to calculate the target point feature value, until the local curvature features of each detection point in the suction and shovel position point cloud are obtained.

[0162] It should be noted that step S501 is a fundamental step in the curvature feature extraction method. Its core task is to perform point-by-point curvature calculation on the point cloud at each suction cup location to obtain the local curvature features corresponding to each detection point. The special feature of this step is that it treats each point in the point cloud as a potential feature point and quantifies the local curvature by calculating the geometric properties of each point and its neighborhood.

[0163] First, target detection points are selected from the point cloud of the suction cup location, and the corresponding neighborhood detection points are determined. This process essentially involves selecting regions of interest within the point cloud, which could be the edges, corners, or other critical parts of the suction cup. The strategy for selecting target detection points may be based on prior knowledge, such as the geometry of the suction cup and the expected shedding pattern, or through automated methods, such as cluster analysis, to identify significant geometric feature regions in the point cloud. Determining neighborhood detection points involves defining a neighborhood range, which determines the number and distribution of points considered in the curvature calculation. The choice of neighborhood significantly impacts the accuracy and robustness of the curvature calculation.

[0164] Next, the target point feature values ​​and the neighboring point feature values ​​of each neighboring target point are calculated. This step involves performing geometric analysis on the selected point and its neighboring points to extract feature values ​​that describe the local surface shape. These feature values ​​may include the point's normal vector, principal curvature, Gaussian curvature, etc. Calculating these feature values ​​typically requires constructing the point's local covariance matrix and performing eigenvalue decomposition to obtain curvature values ​​that describe the degree of local surface curvature.

[0165] Then, curvature features are calculated based on the feature values ​​of the target point and the feature values ​​of its neighboring points to obtain the local curvature features corresponding to the target detection point. This step is the core of curvature feature extraction; it combines the feature values ​​of the target point and its neighboring points to calculate a curvature value that reflects the local geometric changes of that point. Local curvature features can reveal subtle geometric changes in the point cloud, which is crucial for identifying key features such as the edges and corners of the hook.

[0166] In some more specific embodiments, based on the calculation of the normal vector and curvature of the point cloud, a covariance matrix is ​​constructed through neighborhood points, and the eigenvalue ratio reflects the curvature. Local curvature features are represented as... , can be represented as:

[0167] ;

[0168] Wherein, if the feature value of the target detection point is represented as The feature values ​​of each neighboring detection point corresponding to the target detection point are represented as follows: , ... It should be understood that relatively high local curvature characteristics correspond to areas such as edges or corners.

[0169] Finally, target detection points are reselected from the previously unselected detection points in the suction location point cloud, and the corresponding neighboring detection points are redefined. The process then returns the target point feature value and the neighboring point feature values ​​of each neighboring detection point, continuing until the local curvature features of each detection point in the suction location point cloud are obtained. This iterative process ensures that every point in the point cloud is analyzed, enabling the comprehensive extraction of the curvature features of the entire suction location point cloud. The unique feature of this method is its ability to provide local curvature information for each point in the point cloud, providing rich geometric data for subsequent feature analysis and state detection.

[0170] In summary, step S501 provides a geometric feature-based method for suction cup detachment detection through point-by-point curvature calculation. The unique feature of this method is its ability to capture the local geometric changes of each point in the point cloud, thus providing a foundation for identifying key features of the suction cup. This method is not only applicable to suction cup detachment detection but can also be extended to other application scenarios requiring detailed geometric analysis.

[0171] In step S502 of some embodiments, dynamic feature analysis is performed based on the local curvature features corresponding to each detection point in the multiple suction and pickling position point clouds to determine the corresponding dynamic adaptive threshold for each detection point.

[0172] It should be noted that dynamic feature analysis and adaptive threshold determination are the core innovations of this application's embodiments, distinguishing them from traditional fixed threshold methods. Their uniqueness lies in shifting the threshold setting from being driven by human experience to being driven by data distribution. Traditional methods typically set a single curvature threshold, but in real-world scenarios, factors such as the suction cup material, ground reflection, and wastewater adhesion can cause the curvature benchmark value to drift. For example, in a humid environment, a water film on the suction cup surface can alter local reflection characteristics, resulting in an overall lower calculated curvature.

[0173] In some more specific embodiments, step S502 dynamically calculates the adaptive threshold for each point or each local region by analyzing the curvature statistics (such as mean and standard deviation) of multiple historical frames or multiple point clouds in the same frame, as expressed as:

[0174] Ω = μ + k·σ;

[0175] Where μ is the mean curvature of the neighborhood, σ is the standard deviation, and k is the sensitivity adjustment coefficient. This adaptive thresholding method can automatically adjust according to environmental changes, thereby improving the robustness of detection.

[0176] This design allows the threshold to automatically adjust to changes in ambient light and ground material, preventing fixed parameters from failing in strong or low light. For dual-camera configurations, the curvature distribution baseline differs between the two cameras due to variations in installation height and angle. Dynamic analysis can generate independent thresholds for each side, enabling personalized feature selection. This process incorporates temporal or spatial contextual information. It should be understood that the dynamic adaptive threshold depends not only on the current point but also incorporates statistical characteristics from neighboring points or historical frames, enabling the differentiation between transient noise and persistent structural deformation.

[0177] In some embodiments, step S503 involves performing dynamic difference analysis on the corresponding detection points based on each dynamic adaptive threshold, so as to divide the multiple detection points into salient feature points and ordinary feature points.

[0178] It should be noted that the unique feature of using dynamic adaptive thresholds for difference analysis and point classification is that it achieves a refined transformation from continuous numerical features to discrete semantic labels.

[0179] In some embodiments, for each detection point, the embodiments of this application calculate its local curvature. The difference between the value and the dynamically adaptive threshold Ω, if If a point is found to be salient, it is considered a salient feature point; otherwise, it is considered an ordinary feature point. This judgment logic is not a simple binarization but incorporates consideration of the degree of relative deviation. It should be understood that salient feature points not only require a high absolute value of curvature but also require a sufficiently significant deviation from the local reference, thereby effectively suppressing misjudgments of isolated high-curvature points caused by measurement noise. In the suction cup detection scenario, salient feature points can correspond to the physical edges of the suction cup, the corners of the mounting bracket, or the breaks caused by detachment, while ordinary feature points correspond to the flat surface of the suction cup or the background ground.

[0180] In some more specific embodiments, the classification process elevates the point cloud from a "set of geometric coordinates" to a "set of semantic labels," eliminating the need for subsequent steps to process the original floating-point coordinates. Instead, it simplifies the computation by statistically analyzing the spatial distribution of the labels. The classification results can be used as intermediate output to generate a visual diagnostic image: salient feature points are marked in red, and ordinary points are marked in blue. Maintenance personnel can visually observe the integrity of the edges, facilitating manual review.

[0181] In some embodiments, step S504 generates a corresponding curvature distribution feature based on the positional distribution of significant and ordinary feature points in the suction location point cloud.

[0182] It should be noted that by aggregating the spatial distribution of salient and ordinary feature points to generate curvature distribution features, the feature abstraction from micro-level point location classification to macro-level component description is completed. The unique aspect of this step lies in its multi-dimensional statistical strategy, which not only calculates the ratio of the two types of points but also focuses on their spatial clustering characteristics, geometric arrangement patterns, and topological connections. For example, the salient feature points of a normal suction cup should form two parallel lines along its long axis (corresponding to the two edges of the suction cup), while a detached suction cup may only have a linear distribution on one side or appear as scattered clusters of points. By performing Euclidean clustering on the salient feature points and calculating the spacing between cluster centers, the parallelism of the edges can be quantified; by analyzing the consistency of the normal vectors of ordinary feature points, the smoothness of the suction cup surface can be evaluated.

[0183] In some specific embodiments, these macroscopic features are encoded into a low-dimensional vector (e.g., [edge parallelism, surface smoothness, edge integrity]) as input to a machine learning model. This point-to-feature vector compression achieves data dimensionality reduction and information purification, preserving key discriminative information while avoiding direct input of thousands of raw point clouds into the model, thus meeting the computational resource constraints of embedded systems. Furthermore, the generation process of this distribution feature is interpretable: each component can be traced back to a specific physical meaning, facilitating debugging and fault diagnosis, and meeting the transparency requirements of industrial applications.

[0184] In some embodiments, step S402 involves generating a monitoring target point cloud sequence based on the point cloud of the suction position corresponding to each target suction monitoring screen.

[0185] It should be noted that the point cloud sequence is generated based on continuous monitoring footage. This means that the embodiments of this application not only focus on the geometric shape of a single frame, but also on the changing trend of the suction cup state over time. For example, the suction cup may experience a gradual loosening over several seconds before detaching, which is manifested as a slow drift of the curvature feature sequence; while a sudden detachment corresponds to a drastic jump in the curvature feature.

[0186] In step S403 of some embodiments, the curvature distribution features of the point cloud corresponding to each suction position in the monitoring target point cloud sequence are integrated to obtain a curvature feature sequence.

[0187] It's important to note that integrating the curvature distribution features of each frame into a feature sequence essentially provides a time-dependent input structure for the machine learning model. The unique aspect of this serialized representation lies in its capture of the dynamic evolution of the detachment state, enabling the model to learn normal behavior patterns and abnormal deviation patterns over time, thus possessing strong predictive capabilities. Compared to grid mapping methods that rely solely on single-frame probability statistics, temporal modeling can identify critical states where detachment is imminent but not yet fully complete, enabling preventative maintenance.

[0188] In step S404 of some embodiments, the curvature feature sequence is input into a pre-trained suction cup state detection model for detachment detection to obtain suction cup detachment detection results.

[0189] It's important to note that inputting curvature feature sequences into a pre-trained suction cup state detection model signifies a shift in detection logic from manual rule-driven to a combination of data-driven and knowledge-driven approaches. The suction cup state detection model can employ recurrent neural networks (such as LSTM or GRU) or spatiotemporal convolutional networks, capable of capturing long-range dependencies within the sequence. Training the suction cup state detection model requires labeled data, including curvature feature sequences under various operating conditions such as normal operation, different degrees of looseness, and complete detachment. Supervised learning enables the model to automatically acquire the ability to distinguish states. During the inference phase, the suction cup state detection model performs forward propagation on the input sequence, outputting the detachment probability or state category. The unique aspect of this design is that it transforms the complex physical judgment problem into a pattern recognition problem, avoiding the subjectivity and limitations of manually set thresholds. For example, when sewage splashing causes localized gaps in a single frame's point cloud, traditional thresholding methods might misjudge it as detachment, but the machine learning model can combine temporal context to determine that this is transient interference rather than genuine detachment, significantly reducing the false alarm rate. Furthermore, the scalability of the suction cup state detection model allows for continuous updates and optimizations as data accumulates, and it can even be migrated to different robot models or suction cup components of different shapes, demonstrating the advantage of the algorithm's versatility.

[0190] Reference Figure 6 According to some embodiments of this application, step S105, which performs suction cap detachment detection on the suction cap location point cloud to obtain suction cap detachment detection results, may include:

[0191] Step S601: Obtain the point cloud height of each detection point in the suction position point cloud;

[0192] Step S602: Based on the point cloud height, divide each detection point into hit points and offset points;

[0193] Step S603: Based on the captured location point cloud, perform region rasterization to obtain the corresponding probability raster map; wherein, the probability raster map includes multiple raster units;

[0194] Step S604: Based on the hit points and blank points in each grid cell, determine the corresponding suction grid cell occupied by the suction component in the probability grid map from multiple grid cells.

[0195] Step S605: Perform suction peeling detection based on suction peeling grid units to obtain suction peeling detection results.

[0196] It should be noted that the embodiments of this application construct a suction status evaluation system based on spatial occupancy statistics. Its technical approach takes point cloud height information as the starting point and realizes a rapid conversion from original three-dimensional measurement to state determination through geometric classification, probabilistic grid modeling and threshold statistics.

[0197] In some embodiments, step S601 involves obtaining the point cloud height of each detection point in the suction position point cloud;

[0198] It should be noted that extracting the point cloud height of each detection point in the suction cup location point cloud is crucial. Height information is the most intuitive physical quantity for distinguishing the suction cup from the ground, directly reflecting the spatial positional relationship of the components without complex feature calculations. In actual deployment, height acquisition is not simply a matter of reading the z-coordinate; it needs to be combined with the planar transformation results from step S103 to ensure that the height value is based on the horizontal plane of the reference coordinate system, avoiding systematic deviations introduced by robot tilting.

[0199] In step S602 of some embodiments, each detection point is divided into hit points and offset points based on the point cloud height;

[0200] It's important to note that, based on the point cloud height, each detection point is divided into hit points and miss points. This classification process transforms continuous point cloud height data into discrete state determinations (hit or miss), laying the foundation for subsequent probability statistics. A unique aspect is that this step allows for setting a suitable height threshold that can distinguish between suction cups and the ground, while remaining flexible enough to adapt to different ground materials and robot posture changes.

[0201] In some embodiments, step S603 involves rasterizing the region based on the captured location point cloud to obtain a corresponding probabilistic raster map; wherein the probabilistic raster map includes multiple raster units.

[0202] It should be noted that the process involves rasterizing the captured point cloud to obtain a corresponding probabilistic raster map. The unique aspect of this step is that it converts 3D point cloud data into a 2D raster map, which not only simplifies data processing but also enables probabilistic statistics. The construction of the raster map requires consideration of the raster size and resolution, as these parameters directly affect the detection accuracy and robustness.

[0203] In some embodiments, step S604 involves determining the corresponding suction grid cell occupied by the suction component in the probability grid map from multiple grid cells based on the hit points and blank points in each grid cell.

[0204] It should be noted that, based on the hit points and blank points in each grid cell, the corresponding suction-clip grid cell occupied by the suction-clip component in the probabilistic grid map is determined from multiple grid cells. The unique aspect of this step is that it utilizes probabilistic statistical methods to determine the actual area occupied by the suction-clip component. By analyzing the hit points and blank points in the grid cells, this embodiment can more accurately identify the location of the suction-clip component, maintaining high detection accuracy even under partial occlusion or changes in lighting conditions.

[0205] This discrimination mechanism enables the embodiments of this application to filter out anomalies caused by sewage splashes or dust obstruction, thereby improving the robustness of detection. In the embodiment with dual-side camera configuration, the grid maps generated by the two cameras can be independently updated and then fused. During fusion, a Bayesian product rule is used to increase the probability of regions observed consistently on both sides and suppress the probability of unilaterally abnormal regions, thus achieving preliminary anomaly self-checking at the fusion level.

[0206] According to some embodiments of this application, step S604, which determines the corresponding suction grid cell occupied by the suction component in the probability grid map from multiple grid cells based on the hit points and blank points in each grid cell, may include:

[0207] Based on the number of hit points and blank points in the grid cell, calculate the grid occupancy probability corresponding to the grabbing grid cell;

[0208] When the grid occupancy probability reaches a preset occupancy probability threshold, the grid cell is identified as a grab grid cell.

[0209] It's important to note that the grid occupancy probability of each suction grid cell is calculated based on the number of hit points and blank points within the grid cell. This calculation process essentially involves a statistical analysis of the occupancy status of each grid cell. Hit points refer to those points identified as belonging to the suction component, while blank points are those points that were not hit, typically corresponding to the ground or other non-suction areas. By counting the number of hit points and blank points in each grid cell, a preliminary occupancy status can be obtained. Then, using this statistical data, combined with Bayes' theorem or other probability models, the occupancy probability of each grid cell is calculated. This probability value reflects the likelihood that the grid cell is occupied by the suction component and is a crucial basis for subsequent judgments of the suction status.

[0210] When the grid occupancy probability reaches a preset occupancy probability threshold, the grid cell is identified as a suction grid cell. The unique aspect of this step is the introduction of a threshold judgment to determine the grid cell's state. The preset occupancy probability threshold is a key parameter that needs to be set according to the actual application scenario and detection requirements. The threshold setting needs to balance detection sensitivity and specificity to avoid excessive false alarms and false negatives. When the occupancy probability of a grid cell exceeds this threshold, this embodiment of the application classifies it as a suction grid cell, meaning that the grid cell is very likely to be occupied by the suction component. Compared to simple counting or threshold comparison, this probability-based determination method can better adapt to different environmental conditions and sensor noise, improving detection accuracy and robustness.

[0211] This application embodiment also involves the dynamic updating and maintenance of grid cells. As the robot moves and the environment changes, the state of the grid cells may change. Therefore, this application embodiment needs to continuously update the occupancy probability of the grid cells and re-evaluate the state of the grid cells based on the latest data. This dynamic updating mechanism allows the detection results to reflect the actual state of the suction components in real time, providing the robot with the latest information for decision-making.

[0212] For example, the occupancy probability update mechanism is a core component of probabilistic grid maps, used to update the occupancy probability of obstacles in the environment in real time. This mechanism is commonly used in robot navigation and environmental modeling, especially when processing data from sensors such as LiDAR, depth cameras, etc. The update mechanism includes processing hit points and miss points, which correspond to the cases where obstacles are detected and those where obstacles are not detected, respectively.

[0213] When the depth camera detects an obstacle, it increases the occupancy probability of the corresponding grid cell in the grid map. This process is achieved using Bayes' theorem, as follows:

[0214] ;

[0215] in, It is the odds value of the grid cell s after the data z is observed (i.e., the ratio of the probability of occupancy to the probability of non-occupancy). It is the probability of observing data z under the condition that the grid cells are not occupied (i.e., free space). It is the probability of observing data z under the condition that the grid cells are occupied. It is the prior odds value, that is, the odds value of the raster cells occupied before the observation data.

[0216] It should be noted that Bayes' theorem allows embodiments of this application to update the occupancy state of grid cells by combining prior knowledge and new observation data. When a hit occurs, p(z|s=1) is much larger than p(z|s=0), therefore the odds value increases significantly, indicating that the probability of the grid cell being occupied increases.

[0217] Additionally, miss point updates reduce the occupancy probability of the corresponding grid cell. This can be achieved using a logarithmic update method, reducing the grid's odds value. A miss point represents an obstacle that the sensor expected to detect but failed to do so. This could be because the obstacle is absent, or because of sensor noise or occlusion. The update mechanism balances the response to miss points to avoid erroneously reducing the occupancy probability due to occasional misses.

[0218] In summary, the embodiments of this application achieve a quantitative assessment of the state of the suction-and-tap component by calculating the grid occupancy probability and setting a threshold. The unique aspect of this method is that it transforms the geometric analysis of point cloud data into probabilistic statistics, providing a flexible and accurate assessment method for detachment detection. Furthermore, it involves dynamic updates and clustering segmentation operations to adapt to constantly changing environments and improve detection accuracy.

[0219] In some embodiments, step S605 involves detecting suction peeling failure based on the suction peeling grid unit to obtain the suction peeling failure detection result.

[0220] It should be noted that suction stick detachment detection is performed based on the suction stick grid cells to obtain the suction stick detachment detection results. This step is the final goal of the entire process; it integrates all the information from the preceding steps and determines whether the suction stick has detached by analyzing the state of the suction stick grid cells. This application's embodiment combines point cloud processing, rasterization, and probabilistic statistics to achieve accurate monitoring of the suction stick's state. This method can not only handle complex 3D data but also adapt to different environmental changes, providing technical support for the automation and intelligence of cleaning robots.

[0221] It should be understood that there are many ways to detect suction detachment based on the point cloud at the suction detachment location, and these are not limited to the examples mentioned above.

[0222] Reference Figure 7 , Figure 7 This illustration shows the hardware structure of an electronic device according to another embodiment. The electronic device may include:

[0223] The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0224] The memory 702 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and is called and executed by the processor 701 to execute the suction-and-tack detachment detection method for cleaning robots according to the embodiments of this application.

[0225] The input / output interface 703 is used to implement information input and output;

[0226] The communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0227] Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704);

[0228] The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0229] This application also provides a computer program product, which includes a computer program. The processor of a computer device reads and executes the computer program, causing the computer device to perform the above-described method for detecting suction-and-tweezer detachment from a cleaning robot.

[0230] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “including,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatuses.

[0231] It should be understood that in this disclosure, "at least one item" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0232] It should be understood that in the description of the embodiments of this application, "multiple" means two or more, "greater than", "less than", "exceeding" etc. are understood to exclude the number itself, and "above", "below", "within" etc. are understood to include the number itself.

[0233] In the several embodiments provided in this disclosure, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0234] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0235] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0236] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium may include: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code.

[0237] It should also be understood that the various implementation methods provided in this application can be combined arbitrarily to achieve different technical effects.

[0238] The above is a detailed description of the embodiments of this disclosure. However, this disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this disclosure. All such equivalent modifications or substitutions are included within the scope defined by the claims of this disclosure.

Claims

1. A method for detecting suction cup detachment in cleaning robots, characterized in that, include: A depth camera is used to capture images of the suction cup component of the cleaning robot in real time, resulting in a monitoring image of the target suction cup. Coordinate mapping is performed on the target suction and scraping monitoring screen to obtain depth point cloud data; wherein, the depth point cloud data corresponds to the camera measurement coordinate system; A planar transformation is performed on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system, thereby obtaining the corresponding planar point cloud data; The suction cup position point cloud corresponding to the suction cup component is obtained by segmenting the planar point cloud data. The suction hook detachment detection is performed on the point cloud at the suction hook location to obtain the suction hook detachment detection results; The step of performing suction detachment detection on the point cloud at the suction detachment location to obtain suction detachment detection results includes: Curvature feature analysis is performed on the point cloud of each suction position to determine the curvature distribution feature corresponding to the point cloud of each suction position. A monitoring target point cloud sequence is generated based on the point cloud of the suction position corresponding to each of the target suction monitoring screens. By integrating the curvature distribution features of each suction position point cloud in the monitoring target point cloud sequence, a curvature feature sequence is obtained; The curvature feature sequence is input into a pre-trained suction cup state detection model for detachment detection to obtain the suction cup detachment detection result.

2. The method according to claim 1, characterized in that, The step of performing curvature feature analysis on the point cloud at each suction cup location to determine the curvature distribution features corresponding to each suction cup location point cloud includes: For each of the suction and squeegee locations, point-by-point curvature calculation is performed on the point cloud to obtain the local curvature features corresponding to each detection point. Dynamic feature analysis is performed based on the local curvature features corresponding to each detection point in the multiple suction and grabbing location point clouds to determine the corresponding dynamic adaptive threshold for each detection point. Based on each of the aforementioned dynamic adaptive thresholds, dynamic difference analysis is performed on the corresponding detection points to divide multiple detection points into salient feature points and ordinary feature points. Based on the positional distribution of the salient feature points and the ordinary feature points in the suction location point cloud, the corresponding curvature distribution feature is generated.

3. The method according to claim 2, characterized in that, The step of performing point-by-point curvature calculation on the point cloud at each suction position to obtain the corresponding local curvature features includes: Select a target detection point from the suction location point cloud, and determine the corresponding neighboring detection points of the target detection point; Calculate the target point feature value of the target detection point, and the neighbor point feature value of each of the neighbor detection points; Curvature features are calculated based on the target point feature value and the neighbor point feature values ​​of each of the neighboring detection points to obtain the local curvature features corresponding to the target detection point. The target detection point is reselected from the unselected detection points in the suction position point cloud, and the corresponding neighboring detection points are re-determined. The calculation of the target point feature value and the neighboring point feature value of each of the neighboring detection points is performed until the local curvature feature of each detection point in the suction position point cloud is obtained.

4. The method according to claim 1, characterized in that, The process of performing suction detachment detection on the point cloud at the suction detachment location to obtain suction detachment detection results includes: Obtain the point cloud height of each detection point in the point cloud of the suction cup location; Based on the point cloud height, each detection point is divided into hit points and blank points; Based on the location point cloud of the suction, the region is rasterized to obtain the corresponding probability raster map; wherein, the probability raster map includes multiple raster units; Based on the hit points and blank points in each grid cell, determine the corresponding suction grid cell occupied by the suction component in the probability grid map from multiple grid cells; Based on the suction grid unit, suction peeling detachment detection is performed to obtain the suction peeling detachment detection result.

5. The method according to claim 4, characterized in that, The step of determining the corresponding suction grid cell occupied by the suction component in the probability grid map from multiple grid cells based on the hit point and the blank point in each of the grid cells includes: Based on the number of hit points and the number of blank points in the grid cell, calculate the grid occupancy probability corresponding to the suction grid cell; When the grid occupancy probability reaches a preset occupancy probability threshold, the grid cell is determined as the suction grid cell.

6. The method according to claim 1, characterized in that, The step of performing a planar transformation on the depth point cloud data to transform the depth point cloud data from the camera measurement coordinate system to a pre-constructed reference coordinate system to obtain corresponding planar point cloud data includes: Based on the camera measurement coordinate system and the reference coordinate system, position mapping analysis is performed to obtain the rotation mapping matrix and translation mapping matrix; Based on the rotation mapping matrix and the translation mapping matrix, a rigid transformation matrix is ​​constructed; For each initial calibration point in the depth point cloud data, the rigid transformation matrix is ​​applied to perform a planar transformation to obtain a detection point that corresponds one-to-one with the initial calibration point; wherein, the initial calibration point is based on the camera measurement coordinate system, and the detection point is based on the reference coordinate system; The planar point cloud data is generated based on the detection points in the reference coordinate system that correspond one-to-one with the initial calibration points.

7. The method according to claim 1, characterized in that, The step of obtaining the suction cup position point cloud corresponding to the suction cup component based on the segmentation of the planar point cloud data includes: Obtain the suction cup installation orientation information of the suction cup component in the cleaning robot; wherein, the suction cup installation orientation information is based on the reference coordinate system; Based on the suction nozzle installation orientation information, the suction nozzle discrimination area corresponding to the suction nozzle component is delineated in the reference coordinate system; The planar point cloud data is segmented based on the suction-grabbing discrimination region to obtain the suction-grabbing location point cloud.

8. An electronic device, characterized in that, include: The system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the suction and peeling detection method for a cleaning robot as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the suction and peeling detection method for a cleaning robot as described in any one of claims 1 to 7.

Images