An automatic measurement system for elevator guide rails based on binocular vision

Abstract

The invention relates to the technical field of elevator guide rail measurement, and discloses an automatic elevator guide rail measurement system based on binocular vision, comprising: a video data processing module, which is used for processing video data of the elevator guide rail to obtain M key frame images; a vibration data processing module, which is used for processing vibration data to obtain M time domain data; a feature sequence construction module, which is used for respectively constructing a first feature sequence and a second feature sequence; a guide rail parameter measurement module, which is used for inputting the first feature sequence and the second feature sequence into a prediction model, and outputting the verticality and straightness of the elevator guide rail; the invention can establish a nonlinear mapping relationship between image data and vibration data and the verticality and straightness of the elevator guide rail in real time through the prediction model, and can capture the dynamic changes of the guide rail during the operation of the elevator through the prediction model, so as to realize real-time automatic measurement of the verticality and straightness of the elevator guide rail in a dynamic operation state.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator guide rail measurement, and more specifically, to an elevator guide rail automatic measurement system based on binocular vision. Background Art

[0002] As the basic supporting structure for elevator operation, elevator guide rails directly affect the operation quality of the elevator and the safety of passengers. The verticality and straightness of elevator guide rails are important parameters for measuring their installation quality and maintenance status.

[0003] The traditional method of measuring elevator guide rails is usually to fix a laser plumb line at the bottom or top of the elevator shaft and calibrate it so that the laser beam emitted by the laser plumb line is perpendicular to the horizontal plane to form a reference line. Then, receiving plates are placed at different heights of the elevator guide rails. By comparing the position difference between the laser landing points at different positions on the guide rail and the reference line, the verticality of the guide rail is obtained. Similarly, the laser plumb line is calibrated so that the laser beam it emits forms a reference line along the length of the guide rail. By comparing the position difference between the laser landing points at different positions on the guide rail and the reference line, the straightness of the guide rail is obtained.

[0004] However, the above scheme requires manual setting of the receiving plate, which is cumbersome and time-consuming. The position of the laser landing point is manually recorded, which is easily affected by human errors. In addition, the above scheme can only statically measure the verticality and straightness of the elevator guide rails, and cannot capture the dynamic changes of the guide rails during the operation of the elevator. If the guide rails have large deviations or deformations under dynamic conditions, it may shorten the service life of the elevator and increase the safety hazards of elevator use. Summary of the invention

[0005] The present invention provides an automatic measurement system for elevator guide rails based on binocular vision, which solves the technical problems in the above-mentioned background technology.

[0006] The present invention provides an elevator guide rail automatic measurement system based on binocular vision, comprising:

[0007] A data acquisition module, which is used to collect video data and vibration data of the elevator guide rail when the elevator rises from the starting position to the ending position and descends from the ending position to the starting position;

[0008] The vibration data is represented by the acceleration values ​​at N time points;

[0009] A video data processing module, which is used to process the video data of the elevator guide rail when the elevator is ascending and descending to obtain M key frame images, and extract image features of the M key frame images;

[0010] Image features are represented by a vector of size 1×A;

[0011] A vibration data processing module, which is used to process the vibration data of the elevator guide rail when the elevator is ascending and descending to obtain M time domain data, and extract vibration characteristics of the M time domain data;

[0012] The time domain data is represented by the acceleration values ​​at K time points, K = N / M;

[0013] The vibration characteristics are represented by a vector of size 1×B;

[0014] Where B=C+9, C=K, C corresponds to the acceleration values ​​of C time points after the time domain data is normalized, and 9 corresponds to the maximum value, minimum value, average value, standard deviation, peak-to-valley difference, skewness, kurtosis, kurtosis index and margin index of the acceleration values ​​of the time domain data;

[0015] A feature sequence construction module, which is used to process the image features of M key frame images and the vibration features of M time domain data of the elevator guide rail when the elevator is ascending and descending to construct a first feature sequence and a second feature sequence;

[0016] The first feature sequence includes M sequence units, each of which is represented by a first combination vector, and the first combination vector is obtained by splicing the image features of the key frame image when the elevator is ascending and the vibration features of the time domain data;

[0017] The second feature sequence includes M sequence units, each of which is represented by a second combination vector, and the second combination vector is obtained by splicing the image features of the key frame image when the elevator descends and the vibration features of the time domain data;

[0018] The guide rail parameter measurement module is used to input the first feature sequence and the second feature sequence into the prediction model, and output the verticality and straightness of the elevator guide rail from the starting position to the ending position.

[0019] Furthermore, the starting position, ending position, N, M and A are all custom parameters.

[0020] Furthermore, processing the video data of the elevator guide rail to obtain M key frame images includes the following steps:

[0021] Step S201, performing frame processing on the video data of the elevator guide rail at the same time interval to obtain L frame images, where L>M;

[0022] The time interval is a custom parameter;

[0023] Step S202, calculating the similarity between two adjacent sub-frame images among the L sub-frame images by using the structural similarity index, and if it is determined that the similarity is less than a similarity threshold, the next sub-frame image is used as a key frame image;

[0024] The similarity threshold is a custom parameter with a value range between 0 and 1;

[0025] Step S203: if it is determined that the number of key frame images is greater than M, the similarity threshold is lowered; if it is determined that the number of key frame images is less than M, the similarity threshold is increased until the number of key frame images obtained is equal to M.

[0026] Furthermore, the image features of the key frame image are extracted through a convolutional neural network model. The convolutional neural network model is connected to a first classifier and a second classifier respectively. The classification label of the first classifier is the deformation degree of the elevator guide rail, including: no deformation, slight deformation and obvious deformation. The classification label of the second classifier is the number of cracks on the surface of the elevator guide rail. The classification labels of the first classifier and the second classifier are obtained through manual labeling.

[0027] Furthermore, taking the time point of each key frame image as the center point, move forward to obtain the acceleration values ​​of K / 2 time points, and move backward to obtain the acceleration values ​​of K / 2 time points. When the number of acceleration values ​​obtained by moving forward is less than K / 2, continue to move backward to make up for the shortfall. When the number of acceleration values ​​obtained by moving backward is less than K / 2, continue to move forward to make up for the shortfall.

[0028] Furthermore, the peak-to-valley difference is equal to the maximum value of the acceleration value minus the minimum value of the acceleration value, the kurtosis index is equal to the maximum value of the acceleration value divided by the root mean square value, and the margin index is equal to the maximum value of the acceleration value divided by the absolute average value;

[0029] The calculation formula for skewness is as follows:

[0030] ;

[0031] The calculation formula of Kurtosis is as follows:

[0032] ;

[0033] in represents the acceleration value at the kth time point, represents the average value, Represents standard deviation.

[0034] Further, the prediction model includes: a first hidden layer, a second hidden layer, a first extraction layer, a second extraction layer, a concatenation layer, a fourth classifier, and a fifth classifier;

[0035] The first hidden layer includes M first units, the mth first unit inputs the mth sequence unit of the first feature sequence, and outputs a first update vector, 1≤m≤M;

[0036] The second hidden layer includes M second units, the mth second unit inputs the mth sequence unit of the second feature sequence, and outputs a second update vector;

[0037] The first extraction layer is used to extract the first update vector output by the Mth first unit of the first hidden layer;

[0038] The second extraction layer is used to extract the second update vector output by the Mth second unit of the second hidden layer;

[0039] The concatenation layer is used to concatenate the first update vector output by the Mth first unit of the first hidden layer and the second update vector output by the Mth second unit of the second hidden layer to obtain a third combination vector;

[0040] inputting the third combined vector into the fourth classifier and the fifth classifier respectively;

[0041] The classification label of the fourth classifier represents the verticality of the elevator guide rail from the starting position to the ending position;

[0042] The classification label of the fifth classifier represents the straightness of the elevator guide rail from the starting position to the ending position.

[0043] Furthermore, the calculation formula of the mth first unit of the first hidden layer includes:

[0044] ;

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] ;

[0050] in represents the first update vector output by the mth first unit, , , , and represent the intermediate vector, the first gated output value, the second gated output value, the third gated output value and the fourth gated output value of the mth first unit respectively, and They represent the second gated output value and the third gated output value of the i-th first unit respectively, represents the mth sequence unit of the first feature sequence of the mth first unit input, represents the m-1th sequence unit of the first feature sequence input by the m-1th first unit, , , , and represent the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter and the fifth weight parameter of the mth first unit respectively, represents the bias parameter of the mth first unit, concat represents the concatenation operation, sigmoid represents the sigmoid activation function, and exp represents the power operation of the natural constant. Represents point-wise multiplication.

[0051] Furthermore, the M second units of the second hidden layer are constructed based on long short-term memory neural network units.

[0052] Furthermore, the verticality and straightness of the elevator guide rail from the starting position to the ending position are obtained by a laser plumb line and used as sample labels of training samples for training the prediction model.

[0053] The beneficial effects of the present invention are as follows: the present invention can establish a nonlinear mapping relationship between image data and vibration data and the verticality and straightness of elevator guide rails in real time through a prediction model, and can capture the dynamic changes of the guide rails during the operation of the elevator through the prediction model, thereby realizing real-time automatic measurement of the verticality and straightness of the elevator guide rails in a dynamic operating state. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a schematic diagram of an automatic measurement system for elevator guide rails based on binocular vision of the present invention;

[0055] Figure 2 It is a flow chart of the present invention for processing the video data of the elevator guide rail to obtain M key frame images;

[0056] Figure 3 It is a schematic diagram of the prediction model of the present invention.

[0057] In the figure: data acquisition module 101, video data processing module 102, vibration data processing module 103, feature sequence construction module 104, guide rail parameter measurement module 105, first hidden layer 301, second hidden layer 302, first extraction layer 303, second extraction layer 304, splicing layer 305, fourth classifier 306 and fifth classifier 307. DETAILED DESCRIPTION

[0058] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0059] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0060] like Figure 1 to Figure 3 As shown, an automatic measurement system for elevator guide rails based on binocular vision includes:

[0061] A data acquisition module 101, which is used to collect video data and vibration data of the elevator guide rail when the elevator rises from the starting position to the ending position and descends from the ending position to the starting position;

[0062] The vibration data is represented by the acceleration values ​​at N time points;

[0063] A video data processing module 102, which is used to process the video data of the elevator guide rail when the elevator is ascending and descending to obtain M key frame images, and extract image features of the M key frame images;

[0064] Image features are represented by a vector of size 1×A;

[0065] A vibration data processing module 103, which is used to process the vibration data of the elevator guide rail when the elevator is ascending and descending to obtain M time domain data, and extract vibration characteristics of the M time domain data;

[0066] The time domain data is represented by the acceleration values ​​at K time points, K = N / M;

[0067] The vibration characteristics are represented by a vector of size 1×B;

[0068] Where B=C+9, C=K, C corresponds to the acceleration values ​​of C time points after the time domain data is normalized, and 9 corresponds to the maximum value, minimum value, average value, standard deviation, peak-to-valley difference, skewness, kurtosis, kurtosis index and margin index of the acceleration values ​​of the time domain data;

[0069] A feature sequence construction module 104 is used to process the image features of the M key frame images and the vibration features of the M time domain data of the elevator guide rail when the elevator is ascending and descending to construct a first feature sequence and a second feature sequence;

[0070] The first feature sequence includes M sequence units, each of which is represented by a first combination vector, and the first combination vector is obtained by splicing the image features of the key frame image when the elevator is ascending and the vibration features of the time domain data;

[0071] The second feature sequence includes M sequence units, each of which is represented by a second combination vector, and the second combination vector is obtained by splicing the image features of the key frame image when the elevator descends and the vibration features of the time domain data;

[0072] The guide rail parameter measurement module 105 is used to input the first feature sequence and the second feature sequence into the prediction model, and output the verticality and straightness of the elevator guide rail from the starting position to the ending position.

[0073] In one embodiment of the present invention, a binocular camera is installed outside the elevator car to collect video data of the elevator guide rail, and a vibration sensor is installed outside the elevator car to collect vibration data of the elevator guide rail.

[0074] It should be noted that installing a binocular camera and a vibration sensor on the outside of the elevator car facilitates customizing the starting and ending positions for collecting video data and vibration data. In addition, installing a vibration sensor on the outside of the elevator car is more convenient than installing it directly on the elevator guide rail, facilitates daily maintenance, and will not affect the normal operation of the elevator.

[0075] In one embodiment of the present invention, the starting position, the ending position, N, M and A are all custom parameters. Preferably, the starting position and the ending position set the position where the elevator arrives at each floor. For example, if the starting position is set on the 1st floor, then the ending position is set on the 2nd floor; N is set according to the time it takes for the elevator to rise from the starting position to the ending position. For example, if the time it takes for the elevator to rise from the starting position to the ending position is 3 seconds, and according to the sampling frequency of 100 Hz, N is set to 300; M is set to 6, then K=N / M=300 / 6=50, then B=50+9=59, that is, the vibration characteristics are represented by a vector of size 1×59; A is set to 16, then the size of the first combined vector and the second combined vector is 1×(59+16)=1×75.

[0076] It should be noted that the interval between the starting position and the ending position can also be divided into fine granularity to improve the accuracy of measuring the verticality and straightness of the elevator guide rail. In addition, the sampling frequency of the vibration sensor can be appropriately increased to increase the dimensions of the first combined vector and the second combined vector, which can improve the prediction accuracy of the prediction model to a certain extent.

[0077] In one embodiment of the present invention, Figure 2 As shown, processing the video data of the elevator guide rail to obtain M key frame images includes the following steps:

[0078] Step S201, performing frame processing on the video data of the elevator guide rail at the same time interval to obtain L frame images, where L>M;

[0079] The time interval is a custom parameter. Preferably, the time interval is set to 0.1 seconds. For example, if the duration of the video data of the elevator guide rail is 3 seconds, then 3 seconds divided by 0.1 seconds can obtain 30 frame images, that is, L=30;

[0080] Step S202, calculating the similarity between two adjacent sub-frame images in the L sub-frame images by using SSIM (Structural Similarity Index), and if it is determined that the similarity is less than a similarity threshold, the latter sub-frame image is used as a key frame image;

[0081] The similarity threshold is a custom parameter with a value range of 0 to 1. Preferably, the similarity threshold is set to 0.5;

[0082] Step S203: if it is determined that the number of key frame images is greater than M, the similarity threshold is lowered; if it is determined that the number of key frame images is less than M, the similarity threshold is increased until the number of key frame images obtained is equal to M.

[0083] In one embodiment of the present invention, image features of key frame images are extracted by a convolutional neural network model, and the convolutional neural network model is respectively connected to a first classifier and a second classifier, the classification label of the first classifier is the deformation degree of the elevator guide rail, including: no deformation, slight deformation and obvious deformation, and the classification label of the second classifier is the number of cracks on the surface of the elevator guide rail. The classification labels of the first classifier and the second classifier are obtained by manual labeling.

[0084] It should be noted that the convolutional neural network model can also be connected to a third classifier. The classification label of the third classifier is the degree of corrosion on the surface of the elevator guide rail, including: no corrosion, slight corrosion and severe corrosion. The size and step size of the convolution kernel of the convolutional neural network model are custom parameters. For example, the size of the unified key frame image is 128×128×3, and the convolutional neural network model sets 3 convolutional layers. The size of the convolution kernel of each convolutional layer is 3×3, and the output channels are 32, 64 and 128 respectively. Each convolutional layer is connected to a maximum pooling layer. The pooling window size of the maximum pooling layer is 2×2, and the step size is 2. Then, the feature map size after three pooling is 16×16×128, and then it passes through the global average pooling layer to obtain a vector of size 1×128, which is finally input into the fully connected layer. The weight parameter of the fully connected layer is designed to be a matrix of size 128×A. The image feature obtained by multiplying the two is a vector of size 1×A.

[0085] In one embodiment of the present invention, the time point of each key frame image is taken as the center point, and the acceleration values ​​of K / 2 time points are obtained by moving forward, and the acceleration values ​​of K / 2 time points are obtained by moving backward. When the number of acceleration values ​​obtained by moving forward is less than K / 2, the method continues to move backward to make up for the shortfall. When the number of acceleration values ​​obtained by moving backward is less than K / 2, the method continues to move forward to make up for the shortfall.

[0086] For example, when N is set to 300 and M is set to 6, then K=N / M=300 / 6=50. When the number of acceleration values ​​obtained by moving forward is 20 and the number of acceleration values ​​obtained by moving backward is already 25, then continue to move backward to obtain the acceleration values ​​of 5 time points. This ensures that the size of each time domain is the same and coincides with the time point of the key frame image, which can improve the accuracy of the prediction model output to a certain extent.

[0087] In one embodiment of the present invention, the peak-to-valley difference is equal to the maximum value of the acceleration value minus the minimum value of the acceleration value, the kurtosis index is equal to the maximum value of the acceleration value divided by the root mean square value, and the margin index is equal to the maximum value of the acceleration value divided by the absolute average value;

[0088] The calculation formula for skewness is as follows:

[0089] ;

[0090] The calculation formula of Kurtosis is as follows:

[0091] ;

[0092] in represents the acceleration value at the kth time point, represents the average value, Represents standard deviation.

[0093] In one embodiment of the present invention, Figure 3 As shown, the prediction model includes: a first hidden layer 301, a second hidden layer 302, a first extraction layer 303, a second extraction layer 304, a concatenation layer 305, a fourth classifier 306 and a fifth classifier 307;

[0094] The first hidden layer includes M first units, the mth first unit inputs the mth sequence unit of the first feature sequence, and outputs a first update vector, 1≤m≤M;

[0095] The second hidden layer includes M second units, the mth second unit inputs the mth sequence unit of the second feature sequence, and outputs a second update vector;

[0096] The first extraction layer is used to extract the first update vector output by the Mth first unit of the first hidden layer;

[0097] The second extraction layer is used to extract the second update vector output by the Mth second unit of the second hidden layer;

[0098] The concatenation layer is used to concatenate the first update vector output by the Mth first unit of the first hidden layer and the second update vector output by the Mth second unit of the second hidden layer to obtain a third combination vector;

[0099] inputting the third combined vector into the fourth classifier and the fifth classifier respectively;

[0100] The classification label of the fourth classifier represents the verticality of the elevator guide rail from the starting position to the ending position;

[0101] The classification label of the fifth classifier represents the straightness of the elevator guide rail from the starting position to the ending position.

[0102] In one embodiment of the present invention, the calculation formula of the mth first unit of the first hidden layer includes:

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] in represents the first update vector output by the mth first unit, , , , and represent the intermediate vector, the first gated output value, the second gated output value, the third gated output value and the fourth gated output value of the mth first unit respectively, and They represent the second gated output value and the third gated output value of the i-th first unit respectively, represents the mth sequence unit of the first feature sequence of the mth first unit input, represents the m-1th sequence unit of the first feature sequence input by the m-1th first unit, , , , and represent the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter and the fifth weight parameter of the mth first unit respectively, represents the bias parameter of the mth first unit, concat represents the concatenation operation, sigmoid represents the sigmoid activation function, and exp represents the power operation of the natural constant. Represents point-wise multiplication.

[0110] It should be noted that, if the number of dimensions of the combination vector corresponding to the first feature sequence input to the first unit is 75, the fifth weight parameter can be designed as a vector of size 150×1, and the fourth gated output value is a real number between 0 and 1. The second weight parameter and the third weight parameter can be designed as a vector of size 75×1, and the first gated output value and the second gated output value are both real numbers. The third weight parameter can be designed as a matrix of size 75×32, and the third gated output value is a vector of size 1×32. According to the above calculation formula, the size of the intermediate vector is also 1×32. The first weight parameter can be designed as a matrix of size 32×16, and the first update vector output by the first unit is finally 1×16.

[0111] It should be noted that the size of the second update vector output by the second unit of the second hidden layer may be the same as or different from the size of the first update vector output by the first unit of the first hidden layer.

[0112] In one embodiment of the present invention, the M second units of the second hidden layer are constructed based on LSTM (Long Short-Term Memory Neural Network) units.

[0113] In one embodiment of the present invention, the M second units of the second hidden layer may also be constructed based on RNN (recurrent neural network) units or GRU (gated neural network) units.

[0114] In one embodiment of the present invention, the verticality and straightness of the elevator guide rail from the starting position to the ending position are obtained by a laser plumb meter and used as sample labels of training samples for training the prediction model.

[0115] It should be noted that in the training process of the prediction model, the difference between the value output by the prediction model at each iteration and the sample label of the training sample is specified as the loss function, and then the gradient information of the weight parameters and bias parameters in the prediction model is calculated by the chain rule. The weight parameters and bias parameters are updated in reverse based on the gradient information combined with the gradient descent algorithm.

[0116] Furthermore, thresholds can be set for the verticality and parallelism output by the prediction model respectively. When the set thresholds are exceeded, an alarm is issued to avoid dangerous situations in the elevator.

[0117] Furthermore, the prediction model can output the parallelism of the elevator guide rail from the starting position to the ending position, that is, a nonlinear mapping relationship between the image data and the vibration data and the parallelism is established through the prediction model, which will not be elaborated here.

[0118] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An automatic measurement system for elevator guide rails based on binocular vision, characterized in that: include: A data acquisition module, which is used to collect video data and vibration data of the elevator guide rail when the elevator rises from the starting position to the ending position and descends from the ending position to the starting position; The vibration data is represented by the acceleration values ​​at N time points; A video data processing module, which is used to process the video data of the elevator guide rail when the elevator is ascending and descending to obtain M key frame images, and extract image features of the M key frame images; Image features are represented by a vector of size 1×A; A vibration data processing module, which is used to process the vibration data of the elevator guide rail when the elevator is ascending and descending to obtain M time domain data, and extract the vibration characteristics of the M time domain data; The time domain data is represented by the acceleration values ​​at K time points, K = N / M; The vibration characteristics are represented by a vector of size 1×B; Where B=C+9, C=K, C corresponds to the acceleration value of C time points after the time domain data is normalized, and 9 corresponds to the maximum value, minimum value, average value, standard deviation, peak-to-valley difference, skewness, kurtosis, kurtosis index and margin index of the acceleration value of the time domain data; A feature sequence construction module, which is used to process the image features of M key frame images of the elevator guide rail when the elevator is ascending and descending and the vibration features of M time domain data to construct a first feature sequence and a second feature sequence; The first feature sequence includes M sequence units, each of which is represented by a first combination vector, and the first combination vector is obtained by splicing the image features of the key frame image when the elevator is ascending and the vibration features of the time domain data; The second feature sequence includes M sequence units, each of which is represented by a second combination vector, and the second combination vector is obtained by splicing the image features of the key frame image when the elevator descends and the vibration features of the time domain data; The guide rail parameter measurement module is used to input the first feature sequence and the second feature sequence into the prediction model, and output the verticality and straightness of the elevator guide rail from the starting position to the ending position.

2. The automatic elevator guide rail measurement system based on binocular vision according to claim 1 is characterized in that: The start position, end position, N, M, and A are all custom parameters.

3. The automatic elevator guide rail measurement system based on binocular vision according to claim 1 is characterized in that: Processing the video data of the elevator guide rail to obtain M key frame images includes the following steps: Step S201, performing frame processing on the video data of the elevator guide rail at the same time interval to obtain L frame images, where L>M; The time interval is a custom parameter; Step S202, calculating the similarity between two adjacent sub-frame images among the L sub-frame images by using the structural similarity index, and if it is determined that the similarity is less than a similarity threshold, the next sub-frame image is used as a key frame image; The similarity threshold is a custom parameter with a value range between 0 and 1; Step S203: if it is determined that the number of key frame images is greater than M, the similarity threshold is lowered; if it is determined that the number of key frame images is less than M, the similarity threshold is increased until the number of key frame images obtained is equal to M.

4. The automatic elevator guide rail measurement system based on binocular vision according to claim 1 is characterized in that: The image features of the key frame image are extracted through a convolutional neural network model. The convolutional neural network model connects the first classifier and the second classifier. The classification label of the first classifier is the deformation degree of the elevator guide rail, including: no deformation, slight deformation and obvious deformation. The classification label of the second classifier is the number of cracks on the surface of the elevator guide rail. The classification labels of the first classifier and the second classifier are obtained through manual labeling.

5. The automatic elevator guide rail measurement system based on binocular vision according to claim 1 is characterized in that: Taking the time point of each key frame image as the center point, move forward to obtain the acceleration values ​​of K / 2 time points, and move backward to obtain the acceleration values ​​of K / 2 time points. When the number of acceleration values ​​obtained by moving forward is less than K / 2, continue to move backward to make up for the shortfall. When the number of acceleration values ​​obtained by moving backward is less than K / 2, continue to move forward to make up for the shortfall.

6. The automatic elevator guide rail measurement system based on binocular vision according to claim 1, characterized in that: The peak-to-valley difference is equal to the maximum acceleration value minus the minimum acceleration value, the kurtosis index is equal to the maximum acceleration value divided by the root mean square value, and the margin index is equal to the maximum acceleration value divided by the absolute average value; The calculation formula for skewness is as follows: ； The calculation formula of Kurtosis is as follows: ； in represents the acceleration value at the kth time point, represents the average value, Represents standard deviation.

7. The automatic elevator guide rail measurement system based on binocular vision according to claim 1, characterized in that: The prediction model includes: a first hidden layer, a second hidden layer, a first extraction layer, a second extraction layer, a concatenation layer, a fourth classifier and a fifth classifier; The first hidden layer includes M first units, the mth first unit inputs the mth sequence unit of the first feature sequence, and outputs a first update vector, 1≤m≤M; The second hidden layer includes M second units, the mth second unit inputs the mth sequence unit of the second feature sequence, and outputs a second update vector; The first extraction layer is used to extract the first update vector output by the Mth first unit of the first hidden layer; The second extraction layer is used to extract the second update vector output by the Mth second unit of the second hidden layer; The concatenation layer is used to concatenate the first update vector output by the Mth first unit of the first hidden layer and the second update vector output by the Mth second unit of the second hidden layer to obtain a third combination vector; inputting the third combined vector into the fourth classifier and the fifth classifier; The classification label of the fourth classifier represents the verticality of the elevator guide rail from the starting position to the ending position; The classification label of the fifth classifier represents the straightness of the elevator guide rail from the starting position to the ending position.

8. The automatic elevator guide rail measurement system based on binocular vision according to claim 7, characterized in that: The calculation formula of the mth first unit of the first hidden layer is include: ； ； ； ； ； ； in represents the first update vector output by the mth first unit, , , , and represent the intermediate vector, the first gated output value, the second gated output value, the third gated output value and the fourth gated output value of the mth first unit respectively, and They represent the second gated output value and the third gated output value of the i-th first unit respectively, represents the mth sequence unit of the first feature sequence of the mth first unit input, represents the m-1th sequence unit of the first feature sequence input by the m-1th first unit, , , , and represent the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter and the fifth weight parameter of the mth first unit respectively, represents the bias parameter of the mth first unit, concat represents the concatenation operation, sigmoid represents the sigmoid activation function, and exp represents the power operation of the natural constant. Represents point-wise multiplication.

9. The automatic elevator guide rail measurement system based on binocular vision according to claim 7, characterized in that: The M second units of the second hidden layer are constructed based on long short-term memory neural network units.

10. The automatic elevator guide rail measurement system based on binocular vision according to claim 1, characterized in that: The verticality and straightness of the elevator guide rail from the starting position to the ending position obtained by a laser plumb meter are used as sample labels of training samples for training the prediction model.

Images