A method for automatically extending the middle part of a belt conveyor based on displacement feedback

Abstract

The present application relates to the field of machine vision, in particular to a kind of automatic extension method of middle part of belt conveyor based on displacement feedback.The method comprises: obtaining the three-dimensional point cloud data of the nose of middle part of belt conveyor in real time;Extract stable point cloud in the three-dimensional point cloud data;Using improved weighted iterative closest point algorithm is registered to all stable point cloud and preset target point cloud set, obtain the optimal rotation matrix and translation vector, and analyze multiple deviation signals;Based on multiple deviation signals control the hydraulic actuator of middle part of belt conveyor carries out automatic adjustment until all deviation signals are less than preset threshold, complete the automatic extension and butt joint of middle part of belt conveyor.The scheme of the present application can accurately carry out the automatic extension and butt joint of middle part of belt conveyor.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine vision. More particularly, the present application relates to a method for automatically extending the middle part of a belt conveyor based on displacement feedback. BACKGROUND

[0002] In a modernized fully mechanized coal mining face, in order to ensure the continuity of coal mining, the crossheading belt conveyor must be synchronized with the continuously moving transfer equipment (such as hydraulic supports, etc.) at the front end to avoid material transport interruption or material accumulation.

[0003] At present, the current hydraulic-driven telescopic conveyor has realized basic automation, but the core problem is the lack of precise real-time sensing capability. Especially in the harsh environment of underground dust or water mist, the traditional sensor cannot accurately obtain the three-dimensional absolute pose of the middle part of the belt conveyor.

[0004] Therefore, how to accurately control the automatic docking of the belt conveyor to avoid misalignment, equipment jamming and material spilling is particularly important. SUMMARY

[0005] The purpose of the present application is to provide a method for automatically extending the middle part of a belt conveyor based on displacement feedback to solve the problem that the automatic docking of the belt conveyor cannot be accurately controlled in the prior art, resulting in misalignment, equipment jamming and material spilling. To this end, the present application provides a solution in the following aspect.

[0006] The present application provides a method for automatically extending the middle part of a belt conveyor based on displacement feedback, comprising:

[0007] real-time acquisition of three-dimensional point cloud data of the middle part of the belt conveyor;

[0008] extracting stable point clouds in the three-dimensional point cloud data, wherein the stable point clouds are point clouds with stability greater than a first threshold in the three-dimensional point cloud data;

[0009] using an improved weighted iterative closest point algorithm to register all stable point clouds and a preset target point cloud set to obtain an optimal rotation matrix and translation vector;

[0010] A plurality of deviation signals of the intermediate section head of the belt conveyor are parsed from the optimal rotation matrix and translation vector; the hydraulic actuators of the intermediate section of the belt conveyor are controlled based on the plurality of deviation signals to automatically adjust until all the deviation signals are less than a preset threshold, and the automatic extension and docking of the intermediate section of the belt conveyor are completed;

[0011] The improved weighted iterative closest point algorithm comprises an adaptive weight; the adaptive weight is positively correlated with a geometric feature value of any stable point cloud and negatively correlated with a minimum Euclidean distance of any stable point cloud to a key docking area; the geometric feature value represents a geometric space condition in a neighborhood range of each stable point cloud; and the key docking area is a clustering cluster with the largest number after clustering of all stable point clouds.

[0012] The above scheme effectively eliminates the interference of dynamic noise such as dust and water mist by acquiring three-dimensional point cloud data in real time and extracting stable point clouds; meanwhile, by comprehensively considering the geometric feature saliency (such as corners and edges) of the stable point clouds and the distance of the stable point clouds to the key docking area, and by using the weighted iterative closest point algorithm to register the stable point clouds and the target point clouds, the registration process can focus on the point cloud data corresponding to the core structural features in the docking task, rather than treating all point clouds equally, thereby improving the accuracy of pose solving, and finally realizing high-precision automatic extension and docking of the intermediate section of the belt conveyor, and improving the operation efficiency and safety.

[0013] Optionally, the stability is a ratio of a frame number of neighboring points of any point cloud in the three-dimensional point cloud data within a set historical frame number to the historical frame number;

[0014] The historical frame number is a frame number of a plurality of time points before a time point at which the three-dimensional point cloud data of the intermediate section head of the belt conveyor is acquired in real time;

[0015] The determination condition of the frame number of the neighboring points is that there is a historical point cloud in the point cloud data of any historical frame number, and a minimum Euclidean distance of the historical point cloud to the any point cloud is less than a set neighboring space radius threshold.

[0016] The above scheme defines the stability of the point cloud as the frequency of the existence of neighboring points in the continuous historical frames, and can more reliably distinguish the permanent structure of the belt conveyor head and the transient noise in the environment.

[0017] Optionally, the calculation method of the geometric feature value comprises:

[0018] For each stable point cloud, a three-dimensional covariance matrix of a point cloud set in a neighborhood range thereof is constructed; the point cloud set in the neighborhood range is composed of all neighboring points selected within a neighboring space radius threshold and centered on the stable point cloud;

[0019] and eigenvalue decomposition is performed on the three-dimensional covariance matrix to obtain three eigenvalues; the geometric eigenvalue is equal to the ratio of the minimum eigenvalue to the sum of the three eigenvalues.

[0020] The above scheme can accurately quantify the geometric complexity in the neighborhood range of each stable point cloud from a mathematical point of view, and effectively identify the corner and edge high information area which is crucial for pose registration.

[0021] Optionally, the adaptive weight is:

[0022]

[0023] In the formula, is the adaptive weight of the stable point cloud ; is the geometric eigenvalue of the stable point cloud ; is the average value of all geometric eigenvalues in the key docking area; is the minimum value of the Euclidean distance from the stable point cloud to all point clouds in the key docking area, is the key docking area, i is the serial number of the stable point cloud, is the average value of the Euclidean distance from the stable point cloud to all point clouds in the key docking area. is the adaptive weight of the stable point cloud

[0024] The adaptive weight is obtained by fusing the geometric eigenvalue of the stable point cloud and the distance from the stable point cloud to the key docking area, so that the algorithm can preferentially align the point cloud with rich geometric features and close to the docking core part, and effectively suppress the interference of non-key point clouds or flat area point clouds far away from the key docking area on the registration result.

[0025] Optionally, the clustering cluster is obtained by using a DBSCAN spatial clustering algorithm.

[0026] Optionally, the plurality of deviation signals include longitudinal position deviation, transverse position deviation and angle deviation; the longitudinal position deviation and the transverse position deviation are obtained by orthogonal decomposition of the translation vector; and the angle deviation is obtained by Euler angle decomposition of the rotation matrix.

[0027] Optionally, the hydraulic actuator of the middle part of the belt conveyor is automatically adjusted based on the plurality of deviation signals, including:

[0028] A multi-channel PID controller is used, the longitudinal position deviation, the transverse position deviation and the angle deviation are taken as inputs, and a control signal is output to drive the telescopic oil cylinder and the biasing oil cylinder of the middle part of the belt conveyor to act.

[0029] ​​The scheme can effectively avoid overshoot and oscillation in the adjustment process, and ensure that the head of the belt conveyor can be quickly, smoothly and accurately aligned with the target position.

[0030] Optionally, before extracting the stable point cloud, the three-dimensional point cloud data is further subjected to statistical outlier removal preprocessing to obtain denoised three-dimensional point cloud data.

[0031] Optionally, the target point cloud set is three-dimensional point cloud data collected and stored by the belt conveyor docking device in a stable state.

[0032] Optionally, the weighted registration error function when the weighted iterative closest point algorithm is used to register all stable point clouds and the preset target point cloud set is :

[0033] ;

[0034] In the formula, is a rotation matrix to be solved; is a translation vector to be solved; I is the total number of stable point clouds; is the i th target point cloud in the target point cloud set; is the adaptive weight of the i th stable point cloud ; is the square of the Euclidean distance.

[0035] The beneficial effects of the present application are:

[0036] The scheme of the present application combines the geometric features of the stable point cloud and the distance of the stable point cloud to the key docking area to calculate the adaptive weight, and applies it to the weighted ICP registration algorithm, realizes high-precision and high-robustness calculation of the head pose of the belt conveyor, and further completes the precise automatic docking of the belt conveyor under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 A step flowchart of a displacement feedback-based automatic extension method of the middle part of the belt conveyor in the embodiment is schematically shown;

[0038] Figure 2 A schematic view of the three-dimensional point cloud data of the head of the middle part of the belt conveyor collected in real time in the embodiment is schematically shown;

[0039] Figure 3 A schematic view of the preprocessed three-dimensional point cloud data in the embodiment is schematically shown;

[0040] Figure 4 A schematic view of the extracted stable point cloud in the embodiment is schematically shown;

[0041] Figure 5 A schematic diagram of a key docking area acquired in the embodiment is shown. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the application will be clearly and completely described in conjunction with the drawings in the embodiments of the application.

[0043] The embodiment of the application provides a method for automatically extending a middle part of a belt conveyor based on displacement feedback, aiming at solving the problem that in the prior art, due to the harsh environment such as underground dust and water mist, the sensor sensing ability is limited, the real-time three-dimensional pose of the head of the belt conveyor cannot be accurately acquired, cumulative errors are generated in the automatic extension process, and the problems of docking misplacement, equipment jamming and material spilling are caused.

[0044] As shown in Figure 1 , the method for automatically extending the middle part of the belt conveyor based on displacement feedback in the embodiment comprises the following steps:

[0045] Step S1, acquiring three-dimensional point cloud data of the head of the middle part of the belt conveyor in real time, and preprocessing the three-dimensional point cloud data.

[0046] Specifically, a three-dimensional sensing device is installed on the head structure of the middle part of the belt conveyor, which is used to acquire three-dimensional point cloud data of the head and its surrounding environment in real time; the acquired three-dimensional point cloud data is a set of a series of data points with three-dimensional coordinates .

[0047] The three-dimensional sensing device is a three-dimensional laser radar or an industrial-grade structured light camera.

[0048] Exemplarily, a three-dimensional laser radar is installed at a fixed position of the roof or side of the crossheading tunnel, the scanning field of view of the three-dimensional laser radar covers the telescopic stroke range of the middle part of the belt conveyor, and the docking device area in front of the head of the middle part of the belt conveyor is mainly contained, the control system starts the three-dimensional laser radar at a preset frequency (for example, 10 Hz), and continuously acquires three-dimensional point cloud data of the docking device area at the current time . The three-dimensional point cloud data of the head of the middle part of the belt conveyor scanned at the current time is denoted as , and specific reference can be made to Figure 2 .

[0049] It can be understood that due to the complex underground environment, isolated noise points caused by dust particle reflection or equipment vibration will inevitably be contained in the acquired three-dimensional point cloud data, and considering that the isolated noise points are usually far away from the main structure and have few adjacent points, while the point cloud of the real structure has dense neighborhood distribution, therefore, the acquired three-dimensional point cloud data also needs to be denoised.

[0050] As a preferred solution, statistical outlier removal (SOM) algorithm is adopted to filter the three-dimensional point cloud data. , )algorithm is adopted to filter the three-dimensional point cloud data.

[0051] Specifically, by calculating the average distance of each point cloud to other point clouds in its neighborhood, and setting a standard deviation multiple threshold, the point cloud whose average distance exceeds the standard deviation multiple threshold is identified as an outlier and is removed, thereby effectively filtering out isolated noise points, and obtaining the denoised three-dimensional point cloud data (see Figure 3 ).

[0052] Step S2, extracting stable point clouds from the three-dimensional point cloud data.

[0053] Since there is dynamic interference in the environment where the middle part of the belt conveyor is located, the preprocessed three-dimensional point cloud data may still contain point clouds of non-fixed structure, therefore, in order to distinguish between real structure and temporary noise, stable point clouds with high stability in time sequence are screened out from the three-dimensional point cloud data in this embodiment.

[0054] Specifically, the process of obtaining stable point clouds is as follows:

[0055] Firstly, the point cloud data of multiple historical frames before the time of the three-dimensional point cloud data is obtained.

[0056] Among them, the three-dimensional point cloud data of multiple historical frames can be the point cloud data of a set number of historical frames.

[0057] The set number of historical frames can be historical frames; preferably, the value range of to frames.

[0058] Secondly, the ratio of the number of frames in which any point cloud in the three-dimensional point cloud data has a neighboring point in the set number of historical frames to the set number of historical frames is calculated, and the ratio is taken as the stability of the point cloud.

[0059] Among them, the judgment condition of the neighboring point is that there is a historical point cloud in the point cloud data of any historical frame number, and the minimum Euclidean distance from the point cloud to the point cloud is less than a set neighboring space radius threshold.

[0060] Among them, the neighboring space radius threshold can be determined according to the ranging noise characteristics of the laser radar and the device size; in this embodiment, it is preferably .

[0061] Then, after obtaining the stability of all point clouds, the stability is compared with a first threshold, if the stability of any point cloud is greater than the first threshold, the point cloud is determined as a stable point cloud.

[0062] wherein, Figure 4 is a stable point cloud diagram.

[0063] To realize adaptive screening of stable point clouds, the first threshold is the difference between the mean value and the standard deviation of the stability of all points in the three-dimensional point cloud data.

[0064] Step S3, using an improved weighted iterative closest point algorithm to register all stable point clouds and a preset target point cloud set to obtain an optimal rotation matrix and translation vector.

[0065] wherein, the weighted iterative closest point algorithm introduces weights in the weighted registration error function; in each iteration, the algorithm finds a target point cloud matched with the stable point cloud, and calculates a transformation parameter that minimizes the weighted registration error. However, the existing weights only consider the distance in the calculation, and do not consider the geometric characteristics, which makes the accuracy not high when calculating the weighted registration error function.

[0066] Therefore, after obtaining all stable point clouds, an adaptive weight needs to be assigned to each stable structure point for registration alignment in the subsequent registration process.

[0067] wherein, the adaptive weight acquisition process is:

[0068] First, the geometric characteristic value of each stable point cloud is calculated.

[0069] The calculation method of the geometric characteristic value is as follows:

[0070] For each stable point cloud , a point cloud set in the neighborhood range is constructed; based on the point cloud set in the neighborhood range, a three-dimensional covariance matrix is constructed, and the three-dimensional covariance matrix is subjected to eigenvalue decomposition to obtain three eigenvalues, and the ratio of the minimum eigenvalue to the sum of all eigenvalues is taken as the geometric characteristic value of the corresponding stable point cloud.

[0071] The neighborhood range is a point cloud set composed of all neighboring points selected in the neighborhood space radius threshold with the stable point cloud as the center.

[0072] The eigenvalue decomposition of the three-dimensional covariance matrix can be performed by the principal component analysis method.

[0073] In this embodiment, the three-dimensional covariance matrix can be a matrix.

[0074] Exemplarily, if the eigenvalues of a stable point cloud are , , and , then the geometric characteristic value is For: .

[0075] It can be understood that the above geometric feature values actually reflect the geometric complexity or corner point nature of the neighborhood range of each stable point cloud. If the stable point cloud is located on a flat surface, the point cloud in the neighborhood range mainly expands along two main directions, and there is almost no expansion in the normal direction. At this time and are large, and is very close to 0, resulting in being extremely small. Conversely, if the stable point cloud is located at a corner or a region with complex geometric structure, the neighborhood point cloud significantly expands in three main directions, and the sizes of the three feature values are relatively close and are all not 0. At this time occupies a large proportion of the total, resulting in being large.

[0076] The reason for obtaining the above geometric feature values is that the stable point cloud contains various geometric shapes such as planes and edges. Considering that the three feature values obtained by principal component analysis can reflect the distribution shape of the local point cloud, and the greater the proportion of the minimum feature value, the more uniform the distribution of the local point cloud in the three-dimensional space, at this time, the stable point cloud is more likely to be located at the device edge or the connecting surface.

[0077] Secondly, obtain the key docking area.

[0078] In this embodiment, the key docking area is: performing spatial clustering on all stable point clouds to obtain a plurality of clustering clusters, and taking the clustering cluster containing the largest number of point clouds as the key docking area; as shown in Figure 5 , the plurality of stable point clouds in the circle constitute the key docking area.

[0079] As a preferred scheme, a spatial clustering algorithm is adopted, two parameters of a neighborhood radius and a minimum number of core objects are set, all stable point clouds are divided into a plurality of clustering clusters, and the clustering cluster containing the largest number of point clouds is taken as the key docking area . The key docking area usually corresponds to the main structure of the machine head or the part with significant geometric features such as the docking flange.

[0080] Secondly, according to the geometric feature values of each stable point cloud and the distance from the key docking area, the adaptive weight of the corresponding stable point cloud is calculated.

[0081] Since uniform weights weaken the influence of stable point clouds, resulting in the decline of docking accuracy, the adaptive weights are calculated by fusing the geometric saliency and the distance to the key docking area, so that the subsequent registration process of stable point clouds and target point clouds focuses on the core area.

[0082] Specifically, the calculation formula of the adaptive weight is:

[0083] ;

[0084] In the formula, is the adaptive weight of the stable point cloud ; is the geometric feature value of the stable point cloud ; is the average value of all geometric feature values in the key docking area; is the minimum value of the Euclidean distance of the stable point cloud to all point clouds in the key docking area, is the key docking area, and i is the serial number of the stable point cloud, is the average value of the Euclidean distance of the stable point cloud to all point clouds in the key docking area.

[0085] reflects the closeness of the stable point cloud to the key docking area. The greater the value, the closer the stable point cloud is to the key docking area, and the higher weight should be given in the registration; otherwise, the smaller the value, the farther the point is from the key area, and the smaller the adaptive weight. Among them, makes the decay rate of the distance penalty adaptive. When is large, it means that the key docking area itself has very rich features, at this time, only becomes large to have an impact in the denominator; and if is small, it proves that the stable point cloud is close to or located in the key area, at this time, the weight is relatively large, and more attention is paid to the point clouds in or near the key docking area; otherwise, when is small, it means that the key docking area is relatively flat, at this time, the weight will become steep with the decline of , and has a large proportion in the denominator, resulting in a rapid decline in the weight, so that more attention will not be paid to the stable point cloud.

[0086] The above formula dynamically and adaptively determines the "penalty" strength of other stable point clouds due to too far distance by using the average value of the key docking area itself. Among them, since the key docking area contains corner points and plane structures, and the geometric feature values of different point clouds are different, is not equal to 0.

[0087] wherein, is a reference distance, used for normalization.

[0088] It should be noted that the greater the geometric feature value, the higher the geometric saliency of the local structure where the corresponding stable point cloud is located, and the more suitable it is as a registration feature point, and the higher the adaptive weight; on the contrary, the smaller the geometric feature value, the stable point cloud corresponding to the stable point cloud is located in a large plane area or a noise interference area, and the smaller the adaptive weight.

[0089] In this embodiment, after obtaining the adaptive weight of each stable point cloud, all stable point clouds are registered and aligned with the pre-stored target point cloud set, so as to solve the optimal rotation matrix and translation vector.

[0090] Specifically, the process of obtaining the optimal rotation matrix and translation vector is as follows:

[0091] Firstly, a weighted registration error function is constructed.

[0092] wherein, the weighted registration error function is:

[0093] ;

[0094] In the formula, is the rotation matrix to be solved; is the translation vector to be solved; I is the total number of stable point clouds; is the i-th target point cloud in the target point cloud set; is the adaptive weight of the i-th stable point cloud ; is the square of the Euclidean distance.

[0095] wherein, is the new position of the stable point cloud after rotation and translation transformation, is the spatial deviation between the new position after transformation and the target point cloud , the greater the value, the farther the new position after transformation deviates from the target point cloud , the worse the local registration effect; on the contrary, the smaller the value, the stable point cloud is aligned with the target point cloud .

[0096] The greater the adaptive weight, the stable point cloud is given a higher weight in the weighted registration error calculation; on the contrary, the smaller the adaptive weight, the deviation can be tolerated during registration.

[0097] Secondly, the weighted iterative nearest point (ICP) algorithm is performed on the stable point cloud set and the target point cloud set to obtain the rotation matrix and translation vector that minimize the value of the weighted registration error function between each stable point cloud and the target point cloud after transformation.

[0098] After several iterations until convergence, an optimal rotation matrix and translation vector can be obtained, which accurately describes the spatial transformation relationship from the current nose pose to the target pose.

[0099] Since the implementation steps of the improved weighted iterative nearest point (Weighted ICP) algorithm are existing technologies, they will not be described in detail here.

[0100] Among them, the target point cloud is the three-dimensional point cloud data collected and stored by high-precision scanning equipment under historical stable conditions of the belt conveyor docking equipment, which can be regarded as the ideal docking target pose.

[0101] Step S4: Extract multiple deviation signals from the optimal rotation matrix and translation vector to achieve automatic extension and docking control of the middle section of the belt conveyor.

[0102] After obtaining the optimal rotation matrix and translation vector, it is necessary to extract the deviation signal that can be used to control the actuator.

[0103] Specifically, the longitudinal position deviation can be obtained by orthogonally decomposing the translation vector along the longitudinal and transverse axes of the conveyor coordinate system. and lateral position deviation Perform Euler angle decomposition on the rotation matrix (e.g., by...). (Sequence), extract the rotation angle around the vertical axis, which is the angular deviation. .

[0104] The above three deviation signals ( Together, these constitute a complete description of the deviation between the current nose pose and the target pose.

[0105] In this embodiment, based on the above three deviation signals, the hydraulic actuator in the middle of the belt conveyor is controlled to make automatic adjustments.

[0106] In one embodiment, a multi-channel proportional-integral-differential (PI-DI) method is used. The controller receives longitudinal position deviation, lateral position deviation, and angular deviation as input error signals and processes them accordingly. The control law calculates the corresponding control quantities to output corresponding control signals, which are used to drive the telescopic and alignment cylinders in the middle section of the belt conveyor to perform precise actions. For example, longitudinal position deviation. The main control is to extend or retract the telescopic oil cylinder, while the lateral position deviation and the angle deviation The two sides of the adjustment oil cylinder are differentially controlled to realize the translation and steering of the machine head.

[0107] It should be noted that the automatic adjustment of the hydraulic actuator of the middle part of the belt conveyor in the embodiment is a closed-loop feedback system, that is, the three-dimensional point cloud data is continuously collected, the deviation signal is calculated, the hydraulic cylinder is driven to adjust, and until the absolute values of all deviation signals are less than the corresponding preset threshold, indicating that the machine head of the belt conveyor has been accurately aligned with the target position, and the automatic extension and docking process of the middle part of the belt conveyor is completed.

[0108] The preset threshold values of the lateral and longitudinal directions can be , and the angle deviation can be .

[0109] The scheme of the present application realizes high-precision and high-robustness perception of the belt conveyor head pose by introducing time-dimension stability analysis and space-dimension adaptive weight, and finally completes the automatic precise docking in a complex environment in combination with closed-loop feedback control, which significantly improves the operation efficiency and safety.

[0110] In the description of the present specification, the meaning of "a plurality of" is at least two, such as two, three or more, etc., unless otherwise explicitly specified.

[0111] Although the present specification has shown and described several embodiments of the present application, it is obvious to those skilled in the art that such embodiments are provided only in an exemplary manner. Those skilled in the art will think of many changes, changes and alternatives without departing from the idea and spirit of the present application.

Claims

1. A method for automatically extending a middle section of a belt conveyor based on displacement feedback, characterized in that, The method comprises the following steps: real-time acquisition of three-dimensional point cloud data of the middle part head of the belt conveyor; extracting stable point clouds from the three-dimensional point cloud data, wherein the stable point cloud is a point cloud with a stability greater than a first threshold value in the three-dimensional point cloud data; using an improved weighted iterative closest point algorithm to register all stable point clouds and a preset target point cloud set to obtain an optimal rotation matrix and a translation vector; resolving a plurality of deviation signals of the middle part head of the belt conveyor from the optimal rotation matrix and the translation vector; and controlling a hydraulic actuator of the middle part of the belt conveyor based on the plurality of deviation signals to automatically adjust until all deviation signals are less than a preset threshold value, thereby completing automatic extension and docking of the middle part of the belt conveyor; wherein the improved weighted iterative closest point algorithm comprises an adaptive weight; the adaptive weight is positively correlated with a geometric feature value of any stable point cloud and negatively correlated with a minimum Euclidean distance of any stable point cloud to a key docking area; the geometric feature value represents a geometric space condition in a neighborhood range of each stable point cloud; and the key docking area is a clustering cluster with the largest number of clusters after clustering of all stable point clouds.

2. A method for automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 1, characterized in that, the stability is a ratio of a frame number of a neighboring point of any point cloud in the three-dimensional point cloud data to a set historical frame number; the historical frame number is a frame number at a plurality of time points before a time point at which the three-dimensional point cloud data of the middle part head of the belt conveyor is acquired in real time; wherein the determination condition of the frame number of the neighboring point is that there is a historical point cloud in the point cloud data of any historical frame number, and a minimum Euclidean distance of the historical point cloud to the any point cloud is less than a set neighboring space radius threshold value.

3. A method for automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 1, characterized in that, the calculation method of the geometric feature value comprises: for each stable point cloud, constructing a three-dimensional covariance matrix of a point cloud set in a neighborhood range; the point cloud set in the neighborhood range is composed of all neighboring points selected within a neighboring space radius threshold value and centered on each stable point cloud; and performing eigenvalue decomposition on the three-dimensional covariance matrix to obtain three eigenvalues; the geometric feature value is equal to a ratio of a minimum eigenvalue to a sum of the three eigenvalues.

4. A method for automatic extension of the middle part of a belt conveyor based on displacement feedback according to claim 1, characterized in that, the adaptive weight is: ； In the formula, is the adaptive weight of the stable point cloud ; is the geometric feature value of the stable point cloud ; refers to the average value of all geometric feature values in the key docking area; is the minimum value of the Euclidean distance from the stable point cloud to all point clouds in the key docking area, is the key docking area, i is the serial number of the stable point cloud, is the average value of the Euclidean distance from the stable point cloud to all point clouds in the key docking area.

5. A method for automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 1, characterized in that, the clustering cluster is obtained by clustering using a DBSCAN spatial clustering algorithm.

6. A method for automatic extension of the middle part of a belt conveyor based on displacement feedback according to claim 1, characterized in that, the plurality of deviation signals include a longitudinal position deviation, a transverse position deviation, and an angle deviation; the longitudinal position deviation and the transverse position deviation are obtained by orthogonal decomposition of the translation vector; and the angle deviation is obtained by Euler angle decomposition of the rotation matrix.

7. A method of automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 6, characterized in that, controlling the hydraulic actuator of the middle part of the belt conveyor based on the plurality of deviation signals to automatically adjust comprises: using a multi-channel PID controller to take the longitudinal position deviation, the transverse position deviation, and the angle deviation as inputs and output a control signal to drive the telescopic cylinder and the bias cylinder of the middle part of the belt conveyor to act.

8. A method for automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 1, characterized in that, Before extracting the stable point cloud, the method further comprises statistical outlier removal preprocessing of the three-dimensional point cloud data to obtain denoised three-dimensional point cloud data.

9. A method for automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 1, characterized in that, the target point cloud set is three-dimensional point cloud data collected and stored by a belt conveyor docking device in a historical stable state.

10. A method for automatically extending the intermediate section of a belt conveyor based on displacement feedback according to claim 1, characterized in that, A weighted registration error function when registering all stable point clouds and a preset target point cloud set using a weighted iterative closest point algorithm is: ； wherein is the rotation matrix to be solved for; is the translation vector to be solved for; I is the total number of stable point clouds; is the i-th target point cloud in the target point cloud set; is the adaptive weight of the i-th stable point cloud ; is the square of the Euclidean distance.

Images