Device and method for detecting slump of mixed building materials based on machine vision

Abstract

The present invention relates to the technical field of slump detection of mixed building materials, and specifically, to a device and method for detecting the slump of mixed building materials based on machine vision. It includes the following steps: marking the collected video data to form a model training set and a test set; converting the video image frame into a grayscale image, using the SURF algorithm in combination with the ORB algorithm to extract the feature value point set of the final video image frame, using the FlannBasedMatcher algorithm to match the feature values ​​of adjacent frames, screening the matching points and calculating the relative displacement of the matching points, and removing the abnormal values ​​in the relative displacement of the matching points; using the polynomial fitting technology in combination with the Elastic Net regularization to construct a slump detection model; based on the new video data and in combination with the slump detection model, the slump is calculated in real time. The present invention constructs a slump detection model by combining machine vision feature point extraction with the Elastic Net regularization model, which can capture the slump of mixed materials in real time and accurately.

Description

Technical Field

[0001] The invention relates to the technical field of slump detection of mixed building materials, and in particular to a device and method for detecting the slump of mixed building materials based on machine vision. Background Art

[0002] A device and method for detecting the slump of mixed building materials based on machine vision aims to achieve real-time monitoring of the state of mixed building materials and improve the accuracy and response speed of slump detection. By combining SURF and ORB feature point extraction algorithms and coordinating with the FlannBasedMatcher algorithm for feature point matching and displacement calculation, real-time and accurate mixed material slump monitoring can be achieved.

[0003] Existing mixed building material slump detection devices and methods are usually difficult to accurately capture the real-time status of materials in complex construction environments. In addition, due to the difficulty and inefficiency of manual detection, safety hazards and lack of experience of personnel, the slump estimation values ​​of mixed building materials at different stages are inaccurate and untimely. Therefore, a device and method for detecting the slump of mixed building materials based on machine vision are provided. Summary of the invention

[0004] The purpose of the present invention is to provide a device and method for detecting the slump of mixed building materials based on machine vision, so as to solve the problem raised in the above background technology that the slump estimation values ​​of mixed building materials at different stages are inaccurate and untimely due to the difficulty and inefficiency of manual detection, the existence of safety hazards and the lack of experience of personnel.

[0005] To achieve the above object, the present invention provides a method for detecting the slump of mixed building materials based on machine vision, comprising:

[0006] S1. Label the collected video data according to the naming method of cement strength-area-slump to form a model training set and a test set;

[0007] S2, skip frame reading model training set video image frames, convert the video image frames into grayscale images, use image enhancement algorithm to enhance the contrast of grayscale images, use SURF algorithm combined with ORB algorithm to extract feature value point set of final video image frames, use FlannBasedMatcher algorithm to match feature values ​​of adjacent frames, filter matching points and calculate relative displacement of matching points, and remove outliers in relative displacement of matching points;

[0008] S3, based on the relative displacement of matching points, using polynomial fitting technology combined with Elastic Net regularization, to build a slump detection model;

[0009] S4. Based on the new video data and combined with the slump detection model, the slump is calculated in real time.

[0010] As a further improvement of the technical solution, in S1, the collected video data is labeled according to the naming method of cement strength-area-slump to form a model training set. The specific steps are as follows:

[0011] S1.1. Collect video data of mixed building materials and classify the video data according to cement strength grade, regional information and actual slump;

[0012] S1.2. Name the video data according to cement strength, regional information and slump;

[0013] S1.3. The video data and its labeling information should be exported into a standardized video dataset format and divided into a model training set and a test set.

[0014] As a further improvement of the present technical solution, in S2, the video image frames of the model training set are skipped and read, the video image frames are converted into grayscale images, the contrast of the grayscale images is enhanced by an image enhancement algorithm, the SURF algorithm is used in combination with the ORB algorithm to extract the feature value point set of the final video image frame, the FlannBasedMatcher algorithm is used to match the feature values ​​of adjacent frames, the matching points are screened and the relative displacement of the matching points is calculated, and the outliers in the relative displacement of the matching points are eliminated. The specific steps are as follows:

[0015] S2.1. Set frame skipping interval , skip frame to read the video image frame of the model training set ;

[0016] S2.2. Converting video image frames into grayscale images , use the histogram equalization algorithm to enhance the contrast of the grayscale image, and use the dilation operation to enhance the contours of objects in the grayscale image to obtain the final video image frame ;

[0017] S2.3, using SURF algorithm to final video image frame , detect SURF feature points through the determinant of the Hessian matrix ;

[0018] SURF feature points The coordinates are:

[0019] ;

[0020] in, is the SURF feature point The horizontal coordinate of the coordinate; is the SURF feature point The ordinate of the coordinate; is the SURF feature point Descriptors of coordinates;

[0021] S2.4. Use ORB algorithm to process the final video image frame , detect ORB feature points through FAST corner detection and BRIEF descriptor ;

[0022] ORB feature points The coordinates are:

[0023] ;

[0024] in, ORB feature point The horizontal coordinate of the coordinate; ORB feature point The ordinate of the coordinate; ORB feature points Descriptors of coordinates;

[0025] S2.5, SURF feature points and ORB feature points Merge into a set of eigenvalue points ;

[0026] ;

[0027] S2.6, use the FlannBasedMatcher algorithm to match adjacent frames and The eigenvalue point set and Matching to get matching point pairs , calculate the matching point pairs The relative displacement between frames ;

[0028] in, for Video image frame at the moment; for Video image frame at the moment; for The set of eigenvalue points at time; for The set of eigenvalue points at time;

[0029] S2.7. Calculate the average displacement and mean square error of all matching points and remove outliers.

[0030] As a further improvement of the technical solution, in S2.6, the FlannBasedMatcher algorithm is used to match adjacent frames. and The eigenvalue point set and Matching to get matching point pairs ; Calculate matching point pairs The relative displacement between frames , the specific method is as follows:

[0031] S2.6.1. At time Frame , matching feature points The coordinates of , the descriptor is ;

[0032] S2.6.2, at time Frame , matching feature points The coordinates of , the descriptor is ;

[0033] S2.6.3. Calculate matching point pairs by Euclidean distance Relative displacement between frames:

[0034] ;

[0035] in, For matching point pairs The relative displacement between frames.

[0036] As a further improvement of the technical solution, in S2.7, the average displacement and mean square error of all matching points are calculated to remove outliers. The specific method is as follows:

[0037] S2.7.1. Calculation The average relative displacement of the matching point pairs :

[0038] ;

[0039] in, is the total number of matching point pairs;

[0040] S2.7.2 Calculation The mean square error of the matching point pairs :

[0041] ;

[0042] S2.7.3. Eliminate abnormal conditions The abnormal matching point pairs of the conditions are obtained to obtain the relative displacement set of the final matching point pairs ;

[0043] ;

[0044] in, is the relative displacement set of the final matching point pair; The threshold for controlling abnormal conditions.

[0045] As a further improvement of the technical solution, in S3, based on the relative displacement set of the final matching point pair , using polynomial fitting technology combined with Elastic Net regularization, a slump detection model is constructed. The specific steps are as follows:

[0046] S3.1. Relative displacement based on matching point pairs ,The slump detection model was constructed using nonlinear polynomial fitting technology;

[0047] S3.2. Introduce Elastic Net regularization in fitting and constructing slump detection model, and establish Elastic Net regularization loss function ;

[0048] S3.3. Minimize the regularized loss function using gradient descent ;

[0049] S3.4. Use the model training set to train the slump detection model and use the model test set to test the slump detection model, and adjust the estimated coefficient of the slump detection model , order and Elastic Net regularization parameter and , generate the final slump detection model.

[0050] As a further improvement of the technical solution, in S3.1, based on the relative displacement of the matching point pair , the slump detection model is constructed using nonlinear polynomial fitting technology. The specific method is as follows:

[0051] ;

[0052] in, is the estimated coefficient of the slump detection model; is the order of the slump detection model;

[0053] S3.2 introduces Elastic Net regularization in fitting and constructing the slump detection model, and establishes the ElasticNet regularization loss function , the specific method is as follows:

[0054] ;

[0055] in, is the actual slump value; Based on Relative displacement Calculated predicted slump value; is the relative displacement set The number of matching point pairs in ; is the strength coefficient of the L1 regularization term; is the strength coefficient of the L2 regularization term; is the mixing parameter in Elastic Net.

[0056] As a further improvement of the technical solution, in S3.4, the slump detection model is trained using the model training set and the slump detection model is tested using the model test set, and the estimated coefficient of the slump detection model is adjusted. The order and Elastic Net regularization parameter and , generate the final slump detection model, the specific method is as follows:

[0057] S3.4.1. Use the model training set to solve the estimated coefficients of the slump detection model ;

[0058] S3.4.2. Finding the Order of the Optimal Slump Test Model and Elastic Net regularization parameter and , use the model test set to test the slump detection model and generate the final slump detection model:

[0059] ;

[0060] in, is the input relative displacement data; The slump value predicted by the slump detection model.

[0061] As a further improvement of the technical solution, in S4, the slump is calculated in real time based on the new video data and in combination with the slump detection model. The specific method is as follows:

[0062] Calculate the relative displacement of the new video data , and input into the slump detection model In the prediction, the real-time slump value is obtained:

[0063] ;

[0064] in, is the relative displacement of the new video data; is the predicted real-time slump value.

[0065] On the other hand, the present invention provides a device for detecting the slump of mixed building materials based on machine vision, comprising a high-definition industrial camera, an embedded image processing unit, a SURF and ORB feature point detection unit, a feature point matching and displacement calculation module, and a slump calculation processor, characterized in that: the high-definition industrial camera is used to collect video data of mixed building materials in a mixer truck in real time;

[0066] The embedded image processing unit is used to process the video data collected by the high-definition industrial camera;

[0067] The SURF and ORB feature point detection unit is used to extract the SURF and ORB feature points in each frame image;

[0068] The feature point matching and displacement calculation module is used to match the feature points in adjacent frames and calculate the relative displacement between the matching points;

[0069] The slump calculation processor is used to calculate the slump in real time using the trained slump detection model.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. In the method and system for detecting the slump of mixed building materials based on machine vision, the feature point extraction and matching technology based on the combination of SURF and ORB algorithms can quickly and accurately capture the dynamic state of mixed building materials in the mixer truck, achieve higher image recognition accuracy and robustness in complex construction environments, and calculate the slump of materials in real time.

[0072] 2. In the method and system for detecting the slump of mixed building materials based on machine vision, a slump detection model combining polynomial fitting technology and Elastic Net regularization is used to achieve accurate prediction of material slump, and the sparsity and complexity of the model can be adaptively adjusted to avoid overfitting and improve the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] Embodiment 1:

[0076] See also Figure 1 As shown, this embodiment provides a method for detecting the slump of mixed building materials based on machine vision, comprising the following steps:

[0077] S1. Label the collected video data according to the naming method of cement strength-area-slump to form a model training set and a test set;

[0078] In this embodiment S1, the collected video data is labeled according to the naming method of cement strength-area-slump to form a model training set. The specific steps are as follows:

[0079] S1.1. Collect video data of mixed building materials and classify the video data according to cement strength grade, regional information and actual slump;

[0080] The video data should cover the entire process of mixed materials from loading and transporting to unloading in the mixer truck; the video data acquisition equipment should ensure that the resolution and frame rate are high enough;

[0081] The strength grades of cement are 32.5MPa, 42.5MPa, 52.5MPa, etc. These strength grades directly affect the slump of the mixed materials;

[0082] S1.2. Name the video data according to cement strength, regional information and slump; this ensures that the training data is highly structured, which is convenient for subsequent data reading and model training; the naming method is as follows:

[0083] "Region_01_C32.5_Slump_150_Video_01.mp4"

[0084] This naming method can not only provide a clear slump for subsequent models, but also expand the application scenarios of the model through differences in cement strength and region, thereby improving its generalization ability.

[0085] S1.3. The video data and its labeling information should be exported into a standardized video dataset format and divided into a model training set and a test set.

[0086] S2, skip frame reading model training set video image frames, convert the video image frames into grayscale images, use image enhancement algorithm to enhance the contrast of grayscale images, use SURF algorithm combined with ORB algorithm to extract feature value point set of final video image frames, use FlannBasedMatcher algorithm to match feature values ​​of adjacent frames, filter matching points and calculate relative displacement of matching points, and remove outliers in relative displacement of matching points;

[0087] In this embodiment S2, the video image frames of the model training set are skipped and converted into grayscale images. The grayscale image contrast is enhanced by using an image enhancement algorithm. The SURF algorithm is used in combination with the ORB algorithm to extract the feature value point set of the final video image frame. The FlannBasedMatcher algorithm is used to match the feature values ​​of adjacent frames, the matching points are screened and the relative displacement of the matching points is calculated, and the abnormal values ​​in the relative displacement of the matching points are eliminated. The specific steps are as follows:

[0088] S2.1. Set frame skipping interval , skip frame to read the video image frame of the model training set ;

[0089] Set frame skip interval The purpose is to reduce the amount of data processed without sacrificing the overall information quality, thereby improving the processing efficiency of the system. For the slump detection scenario of mixed building materials, the changes in the mixed materials will not be significantly different in each frame, so it is not necessary to process every frame of data. Frame skipping can reduce the computational load while retaining enough image data to ensure the training and detection accuracy of the model.

[0090] If the state of mixed building materials in the video data changes slowly, you can choose a larger value to reduce unnecessary data processing; on the contrary, if the state of the material changes quickly, it is recommended to choose a smaller value to ensure that changes in key states are captured in a timely manner;

[0091] S2.2. Converting video image frames into grayscale images , use the histogram equalization algorithm to enhance the contrast of the grayscale image, and use the dilation operation to enhance the contours of objects in the grayscale image to obtain the final video image frame ;

[0092] S2.3, using SURF algorithm to final video image frame , detect SURF feature points through the determinant of the Hessian matrix ;

[0093] SURF feature points The coordinates are:

[0094] ;

[0095] in, is the SURF feature point The horizontal coordinate of the coordinate; is the SURF feature point The ordinate of the coordinate; is the SURF feature point Descriptors of coordinates;

[0096] The SURF algorithm detects salient points in an image through the determinant of the Hessian matrix, which can reflect local areas with unique information in the image. The SURF algorithm detects strong feature points in an image by calculating the determinant of the Hessian matrix. These feature points are often located at the edge of the image or in areas with strong textures, and can better represent important information in the image.

[0097] is a vector used to describe the local image structure information around the feature point;

[0098] S2.4. Use ORB algorithm to process the final video image frame , detect ORB feature points through FAST corner detection and BRIEF descriptor ;

[0099] ORB feature points The coordinates are:

[0100] ;

[0101] in, ORB feature points The horizontal coordinate of the coordinate; ORB feature points The ordinate of the coordinate; ORB feature points Descriptors of coordinates;

[0102] It is a binary vector calculated by the BRIEF method and is used to describe the local image structure information around the feature point;

[0103] The ORB algorithm combines FAST corner detection and BRIEF descriptor, which can greatly improve computational efficiency while maintaining high detection accuracy;

[0104] S2.5, SURF feature points and ORB feature points Merge into a set of eigenvalue points ;

[0105] ;

[0106] S2.6, use the FlannBasedMatcher algorithm to match adjacent frames and The eigenvalue point set and Matching to get matching point pairs , calculate the matching point pairs The relative displacement between frames ;

[0107] in, for Video image frame at the moment; for Video image frame at the moment; for The set of eigenvalue points at time; for The set of eigenvalue points at time;

[0108] S2.7. Calculate the average displacement and mean square error of all matching points and remove outliers.

[0109] In this embodiment S2.6, the FlannBasedMatcher algorithm is used to match adjacent frames. and The eigenvalue point set and Matching to get matching point pairs ; Calculate matching point pairs The relative displacement between frames , the specific method is as follows:

[0110] S2.6.1. At time Frame , matching feature points The coordinates of , the descriptor is ;

[0111] S2.6.2, at time Frame , matching feature points The coordinates of , the descriptor is ;

[0112] S2.6.3. Calculate matching point pairs by Euclidean distance Relative displacement between frames:

[0113] ;

[0114] in, For matching point pairs The relative displacement between frames.

[0115] In S2.7 of this embodiment, the average displacement and mean square error of all matching points are calculated to remove outliers. The specific method is as follows:

[0116] S2.7.1. Calculation The average relative displacement of the matching point pairs :

[0117] ;

[0118] in, is the total number of matching point pairs;

[0119] S2.7.2 Calculation The mean square error of the matching point pairs :

[0120] ;

[0121] S2.7.3. Eliminate abnormal conditions The abnormal matching point pairs of the conditions are obtained to obtain the relative displacement set of the final matching point pairs ;

[0122] ;

[0123] in, is the relative displacement set of the final matching point pair; The threshold for controlling abnormal conditions.

[0124] S3, based on the relative displacement of matching points, using polynomial fitting technology combined with Elastic Net regularization, to build a slump detection model;

[0125] In this embodiment S3, based on the relative displacement set of the final matching point pair , using polynomial fitting technology combined with Elastic Net regularization, a slump detection model is constructed. The specific steps are as follows:

[0126] S3.1. Relative displacement based on matching point pairs ,The slump detection model was constructed using nonlinear polynomial fitting technology;

[0127] S3.2. Introduce Elastic Net regularization in fitting and constructing slump detection model, and establish Elastic Net regularization loss function ;

[0128] S3.3. Minimize the regularized loss function using gradient descent ;

[0129] Through multiple iterations of the gradient descent method, the loss function is gradually reduced until the convergence condition is reached, and finally the optimal slump detection model is obtained;

[0130] S3.4. Use the model training set to train the slump detection model and use the model test set to test the slump detection model, and adjust the estimated coefficient of the slump detection model , order and Elastic Net regularization parameter and , generate the final slump detection model.

[0131] In this embodiment S3.1, based on the relative displacement of the matching point pair , the slump detection model is constructed using nonlinear polynomial fitting technology. The specific method is as follows:

[0132] ;

[0133] in, is the estimated coefficient of the slump detection model; is the order of the slump detection model;

[0134] In this embodiment S3.2, Elastic Net regularization is introduced in the fitting and construction of the slump detection model, and an Elastic Net regularization loss function is established. , the specific method is as follows:

[0135] ;

[0136] in, is the actual slump value; Based on Relative displacement Calculated predicted slump value; is the relative displacement set The number of matching point pairs in ; is the strength coefficient of the L1 regularization term; is the strength coefficient of the L2 regularization term; is the mixing parameter in Elastic Net.

[0137] when When , Elastic Net is equivalent to Lasso regression;

[0138] when When , Elastic Net is equivalent to Ridge regression;

[0139] In this embodiment S3.4, the slump detection model is trained using the model training set and the slump detection model is tested using the model test set, and the estimated coefficient of the slump detection model is adjusted. The order and Elastic Net regularization parameter and , generate the final slump detection model, the specific method is as follows:

[0140] S3.4.1. Use the model training set to solve the estimated coefficients of the slump detection model ;

[0141] S3.4.2. Finding the Order of the Optimal Slump Test Model and Elastic Net regularization parameter and , use the model test set to test the slump detection model and generate the final slump detection model:

[0142] ;

[0143] in, is the input relative displacement data; The slump value predicted by the slump detection model.

[0144] S4, based on the new video data and combined with the slump detection model, calculate the slump in real time;

[0145] In this embodiment S4, the slump is calculated in real time based on the new video data and in combination with the slump detection model. The specific method is as follows:

[0146] Calculate the relative displacement of the new video data , and input into the slump detection model In the prediction, the real-time slump value is obtained:

[0147] ;

[0148] in, is the relative displacement of the new video data; is the predicted real-time slump value.

[0149] Embodiment 2:

[0150] This embodiment provides a device for detecting the slump of mixed building materials based on machine vision, comprising a high-definition industrial camera, an embedded image processing unit, a SURF and ORB feature point detection unit, a feature point matching and displacement calculation module, and a slump calculation processor, characterized in that: the high-definition industrial camera is used to collect video data of mixed building materials in a mixer truck in real time;

[0151] The embedded image processing unit is used to process the video data collected by the high-definition industrial camera;

[0152] The SURF and ORB feature point detection unit is used to extract the SURF and ORB feature points in each frame image;

[0153] The feature point matching and displacement calculation module is used to match the feature points in adjacent frames and calculate the relative displacement between the matching points;

[0154] The slump calculation processor is used to calculate the slump in real time using the trained slump detection model.

[0155] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for detecting the slump of mixed building materials based on machine vision, characterized in that: The following steps are involved: S1. Label the collected video data according to the naming method of cement strength-area-slump to form a model training set and a test set; S2, skip frame reading model training set video image frames, convert the video image frames into grayscale images, use image enhancement algorithm to enhance the contrast of grayscale images, use SURF algorithm combined with ORB algorithm to extract feature value point set of final video image frames, use FlannBasedMatcher algorithm to match feature values ​​of adjacent frames, filter matching points and calculate relative displacement of matching points, and remove outliers in relative displacement of matching points; S3, based on the relative displacement of matching points, using polynomial fitting technology combined with Elastic Net regularization, to build a slump detection model; S4, based on the new video data and combined with the slump detection model, calculate the slump in real time; In S3, based on the relative displacement set of the final matching point pair , using polynomial fitting technology combined with Elastic Net regularization, a slump detection model is constructed. The specific steps are as follows: S3.

1. Relative displacement based on matching point pairs ,The slump detection model was constructed using nonlinear polynomial fitting technology; S3.

2. Introduce Elastic Net regularization in fitting and constructing slump detection model, and establish Elastic Net regularization loss function ; S3.

3. Minimize the regularized loss function using gradient descent ; S3.

4. Use the model training set to train the slump detection model and use the model test set to test the slump detection model, and adjust the estimated coefficient of the slump detection model , order and Elastic Net regularization parameter and , generate the final slump detection model; In S3.1, based on the relative displacement of the matching point pair , the slump detection model is constructed using nonlinear polynomial fitting technology. The specific method is as follows: ； in, is the estimated coefficient of the slump detection model; is the order of the slump detection model; S3.2 introduces Elastic Net regularization in fitting and constructing the slump detection model, and establishes the Elastic Net regularization loss function , the specific method is as follows: ； in, is the actual slump value; Based on Relative displacement Calculated predicted slump value; is the relative displacement set The number of matching point pairs in ; is the strength coefficient of the L1 regularization term; is the strength coefficient of the L2 regularization term; is the mixing parameter in Elastic Net.

2. The method for detecting the slump of mixed building materials based on machine vision according to claim 1, characterized in that: In S1, the collected video data is labeled according to the naming method of cement strength-area-slump to form a model training set. The specific steps are as follows: S1.

1. Collect video data of mixed building materials and classify the video data according to cement strength grade, regional information and actual slump; S1.

2. Name the video data according to cement strength, regional information and slump; S1.

3. The video data and its labeling information should be exported into a standardized video dataset format and divided into a model training set and a test set.

3. The method for detecting the slump of mixed building materials based on machine vision according to claim 1, characterized in that: In S2, the video image frames of the model training set are read by skipping frames, the video image frames are converted into grayscale images, the contrast of the grayscale images is enhanced by using an image enhancement algorithm, the feature value point set of the final video image frame is extracted by using the SURF algorithm in combination with the ORB algorithm, the feature value matching of adjacent frames is performed by using the FlannBasedMatcher algorithm, the matching points are screened and the relative displacement of the matching points is calculated, and the outliers in the relative displacement of the matching points are eliminated. The specific steps are as follows: S2.

1. Set frame skipping interval , skip frame to read the video image frame of the model training set ; S2.

2. Converting video image frames into grayscale images , use the histogram equalization algorithm to enhance the contrast of the grayscale image, and use the dilation operation to enhance the contours of objects in the grayscale image to obtain the final video image frame ; S2.3, using SURF algorithm to final video image frame , detect SURF feature points through the determinant of the Hessian matrix ; SURF feature points The coordinates are: ； in, is the SURF feature point The horizontal coordinate of the coordinate; is the SURF feature point The ordinate of the coordinate; is the SURF feature point Descriptors of coordinates; S2.

4. Use ORB algorithm to process the final video image frame , detect ORB feature points through FAST corner detection and BRIEF descriptor ; ORB feature points The coordinates are: ； in, ORB feature point The horizontal coordinate of the coordinate; ORB feature points The ordinate of the coordinate; ORB feature points Descriptors of coordinates; S2.5, SURF feature points and ORB feature points Merge into a set of eigenvalue points ; ; S2.6, use the FlannBasedMatcher algorithm to match adjacent frames and The eigenvalue point set and Matching to get matching point pairs , calculate the matching point pairs The relative displacement between frames ; in, for Video image frame at the moment; for Video image frame at the moment; for The set of eigenvalue points at time; for The set of eigenvalue points at time; S2.

7. Calculate the average displacement and mean square error of all matching points and remove outliers.

4. The method for detecting the slump of mixed building materials based on machine vision according to claim 3, characterized in that: In S2.6, the FlannBasedMatcher algorithm is used to match adjacent frames. and The eigenvalue point set and Matching to get matching point pairs ; Calculate matching point pairs The relative displacement between frames , the specific method is as follows: S2.6.

1. At time Frame , matching feature points The coordinates of , the descriptor is ; S2.6.2, at time Frame , matching feature points The coordinates of , the descriptor is ; S2.6.

3. Calculate matching point pairs by Euclidean distance Relative displacement between frames: ;in, For matching point pairs The relative displacement between frames.

5. The method for detecting the slump of mixed building materials based on machine vision according to claim 3, characterized in that: In S2.7, the average displacement and mean square error of all matching points are calculated to remove outliers. The specific method is as follows: S2.7.

1. Calculation The average relative displacement of the matching point pairs : ； in, is the total number of matching point pairs; S2.7.2 Calculation The mean square error of the matching point pairs : ； S2.7.

3. Eliminate abnormal conditions The abnormal matching point pairs of the conditions are obtained to obtain the relative displacement set of the final matching point pairs ; ； in, is the relative displacement set of the final matching point pair; The threshold for controlling abnormal conditions.

6. The method for detecting the slump of mixed building materials based on machine vision according to claim 1, characterized in that: In S3.4, the slump detection model is trained using the model training set and the slump detection model is tested using the model test set, and the estimated coefficient of the slump detection model is adjusted. The order and Elastic Net regularization parameter and , generate the final slump detection model, the specific method is as follows: S3.4.

1. Use the model training set to solve the estimated coefficients of the slump detection model ; S3.4.

2. Finding the Order of the Optimal Slump Test Model and Elastic Net regularization parameter and , use the model test set to test the slump detection model and generate the final slump detection model: ； in, is the input relative displacement data; The slump value predicted by the slump detection model.

7. The method for detecting the slump of mixed building materials based on machine vision according to claim 1, characterized in that: In S4, the slump is calculated in real time based on the new video data and in combination with the slump detection model. The specific method is as follows: Calculate the relative displacement of the new video data , and input into the slump detection model In the prediction, the real-time slump value is obtained: ； in, is the relative displacement of the new video data; is the predicted real-time slump value.

8. A device for detecting the slump of mixed building materials based on machine vision, using the method for detecting the slump of mixed building materials based on machine vision as described in any one of claims 1 to 7 for detection, comprising a high-definition industrial camera, an embedded image processing unit, a SURF and ORB feature point detection unit, a feature point matching and displacement calculation module, and a slump calculation processor, characterized in that: The high-definition industrial camera is used to collect video data of mixed building materials in the mixer truck in real time; The embedded image processing unit is used to process the video data collected by the high-definition industrial camera; The SURF and ORB feature point detection unit is used to extract the SURF and ORB feature points in each frame image; The feature point matching and displacement calculation module is used to match the feature points in adjacent frames and calculate the relative displacement between the matching points; The slump calculation processor is used to calculate the slump in real time using the trained slump detection model.

Images