A method and system for detecting the content of down based on stereovision

Abstract

The application discloses a kind of based on stereovision's down content detection method and system, it is related to down test technical field, down content detection method first reduces down to stereoscopic model, then carries out data comparison classification to it, maximum degree retains the characteristics of down itself, guarantee the accuracy of matching classification result, using algorithm to simulate calculation to the parameter of each down, higher accuracy and higher efficiency;Based on the above-mentioned down content detection method, down content detection system automatically realizes the sorting of down, calculation, greatly reduces the workload of artificial manual sorting, naked eye identification, manual weighing, improves the efficiency of down content detection of down, and the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of the accuracy of

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of down test, in particular to a down content detection method and system based on stereovision. BACKGROUND

[0002] In cold areas, especially in northern areas, down jackets have become essential clothing for people. The warmth retention performance of a down jacket mainly depends on the loftiness of the down jacket, and the loftiness of the down jacket mainly depends on the down content. The down content refers to the content percentage of the down and down silk in all filling down and feather. Generally, the higher the down content, the better the warmth retention performance of the down jacket.

[0003] The domestic down detection mark standard (GB / T 10288 and GB / T 17685) stipulates the characteristics of different down and feather products. At present, the down content detection of the down jacket is mainly carried out in the laboratory. The trained sorting personnel with good discrimination ability use tweezers to classify the down by naked eye. For the down that is difficult to distinguish the category, the sorting personnel will assist by using a projection microscope to project and magnify the shape of the down, and then observe the characteristics of the down by artificial eyes to qualitatively analyze. Finally, the down and down silk sorted by artificial are weighed, and the ratio of the weight of the down and down silk to the weight of the down is the down content.

[0004] However, the above method needs to consume more manpower for detection, and it takes a long time cost to train the sorting personnel with such professional skills. At the same time, since the project belongs to personnel sensory discrimination, the discrimination levels of different sorting personnel are different, and there is a situation that the detection results are greatly different and cannot be consistent. Therefore, for the down content detection of the down, the artificial detection method has the defects of large error and low efficiency, and there is room for improvement. SUMMARY

[0005] In view of the problem that the down content is large in error and low in efficiency by using artificial detection in actual application, the application aims to provide a down content detection method based on stereovision. The down is first restored to a stereoscopic model, and then data comparison and classification are performed thereon, so that the characteristics of the down are retained to the greatest extent, the accuracy of the matching classification result is ensured, the parameters of each down are simulated and calculated by using an algorithm, the accuracy is higher and the efficiency is higher. Based on the above down content detection method based on stereovision, the second purpose is to provide a down content detection system, which automatically realizes the sorting and calculation of the down, greatly reduces the workload of manual sorting, naked eye identification and manual weighing, improves the efficiency of the down content detection of the down, and improves the accuracy of the down content detection result of the down through the feature comparison of the system.

[0006] The specific scheme is as follows:

[0007] A method for detecting the down content based on stereovision, comprising the following steps:

[0008] Obtaining the information related to the down sample and the information related to the database down sample;

[0009] Subdividing the down sample into a limited number of down sub-samples;

[0010] Obtaining the multi-view images of the down sub-samples in a diffusion set space;

[0011] Reconstructing the multi-view images into three-dimensional models by using a set reconstruction algorithm, and generating corresponding reconstructed down models;

[0012] Obtaining the parameter data corresponding to each of the reconstructed down models based on the information related to the database down sample;

[0013] Diffusing the next down sub-sample in the set space and repeating the above steps until the parameter data of the limited number of down sub-samples are obtained;

[0014] Obtaining the down content of the down sample based on the parameter data of all the down sub-samples.

[0015] By using the above technical solution, the down sample to be detected is divided into a limited number of down sub-samples, the down sub-samples are diffused in a set space, and images under multiple views are obtained. For each down, a three-dimensional reconstructed down model is obtained based on the images under multiple views by using a three-dimensional reconstruction algorithm. The reconstructed down model of each down is matched and classified with the down models in the database, and the parameter data of each down model is simulated and calculated based on the classification result. The down content of each down sub-sample is calculated and integrated to obtain the final down content of the down sample. In the traditional manual sorting using tweezers and observation and classification by using a magnifying glass, the down is first restored to a three-dimensional model, and the accuracy of the matching and classification result is guaranteed to the greatest extent when the data of the down is compared and classified. In the manual classification, the down is weighed by using an electronic scale, and the accuracy depends on the accuracy of the electronic scale. In order to guarantee the accuracy of the weighing calculation, more down samples need to be sorted, which further increases the workload of the sorting operator. In the present application, the parameters of each down are simulated and calculated by using an algorithm, which is more accurate and efficient. The down content of each down sub-sample is integrated to obtain the down content result of the entire down sample, which is more accurate. Through accurate classification and accurate calculation of the down, data information of the down sample in different dimensions can be obtained, and the quality of the down can be comprehensively analyzed from multiple dimensions.

[0016] Preferably, the step of obtaining the parameter data corresponding to each of the reconstructed down models based on the information related to the database down sample comprises:

[0017] The database comprises a sub-database of standard feature information of different types of down;

[0018] The reconstructed down model is classified based on the sub-database;

[0019] Parameter data corresponding to the reconstructed down model is obtained based on the classification result.

[0020] By using the above technical solution, the standard feature information of each type of down in the sub-database is compared with the reconstructed down model. If the comparison is successful, the down is classified into the sub-class. The sub-class specifically includes the type of bird to which the down belongs, the color of the down, and the specific type of the down. The down is classified in detail, which facilitates subsequent data analysis.

[0021] Preferably, the classification of the reconstructed down model based on the sub-database comprises:

[0022] The relevant feature information of the reconstructed down model is analyzed and obtained;

[0023] The reconstructed down model is classified based on a preset reconstructed model analysis rule and in combination with the relevant feature information of the reconstructed down model.

[0024] By using the above technical solution, the relevant information features of the down reconstructed model are obtained, and the standard feature information in the corresponding sub-database is compared and classified. The relevant information features include color features, size features, and morphological features.

[0025] Preferably, the preset reconstructed model analysis rule is:

[0026] When the relevant feature information contains standard feature information, high-precision analysis data corresponding to the standard feature information is obtained from the database;

[0027] When the high-precision analysis data cannot be obtained, corresponding simulation analysis data is obtained through an auxiliary processing scheme.

[0028] By using the above technical solution, when the relevant feature information of the down contains the standard feature information of a certain type of down, that is, the down belongs to the sub-class, when the relevant features of the down only partially match the labeled feature information, the down cannot be classified with high precision, and high-precision analysis data cannot be obtained. At this time, corresponding simulation analysis data is obtained through an auxiliary processing scheme to perform as accurate simulation analysis as possible on the down that cannot be classified with high precision, thereby ensuring the integrity of the down classification.

[0029] Preferably, the auxiliary processing scheme at least includes an automatic fitting processing scheme and a manual intervention processing scheme.

[0030] By adopting the technical scheme, the auxiliary processing scheme includes an automatic fitting processing scheme realized by an algorithm and a manual intervention processing scheme by experienced operators, and the classification result is intervened by manual intervention, thereby ensuring the accuracy of the down classification result.

[0031] Preferably, the automatic simulation processing scheme includes the following steps:

[0032] Respective similar models of the reconstructed down model in each of the sub-class databases are obtained;

[0033] Based on each similar model, simulation analysis data corresponding to the reconstructed down model is obtained by a preset simulation algorithm.

[0034] By adopting the technical scheme, the reconstructed down model that cannot be classified with high precision can be classified into a sub-class with the highest feature coincidence degree by a simulation algorithm, and if the coincidence degrees are the same, the reconstructed down model is processed by averaging.

[0035] Preferably, the manual intervention processing scheme includes the following steps:

[0036] Respective similar models of the reconstructed down model in each of the sub-class databases are obtained;

[0037] Based on each similar model, the reconstructed down model is manually analyzed;

[0038] Based on the manual analysis result, simulation analysis data corresponding to the reconstructed down model is obtained according to a preset parameter acquisition algorithm.

[0039] By adopting the technical scheme, the similar model of the reconstructed down model in each of the sub-class databases is provided for manual intervention, a part of interference items is excluded for manual sorting, and the accuracy and speed of manual sorting are improved.

[0040] Preferably, the method further includes the following steps:

[0041] A supplementary correlation between the reconstructed down model and the manual analysis result is established;

[0042] The supplementary correlation is stored in the corresponding sub-class database.

[0043] By adopting the technical scheme, the classification result of manual intervention is stored in the database as a basis for subsequent feature comparison and automatic classification, and the down model in each of the sub-class databases is continuously expanded and corrected, so that more and more accurate high-precision analysis data is obtained.

[0044] Preferably, the multi-view image is reconstructed by using a set reconstruction algorithm to generate a corresponding reconstructed down model, which comprises:

[0045] The gray scale information and image distribution space information on the multi-view image are obtained.

[0046] Based on the gray scale information and image distribution space information, it is determined whether the down sample distribution on the multi-view image is uniform.

[0047] If the distribution is uniform, three-dimensional reconstruction is performed by using a set reconstruction algorithm.

[0048] If the distribution is not uniform, the multi-view image of the down sample in the diffusion set space is re-obtained, and the above determination is performed.

[0049] By using the above technical solution, the gray scale information and image distribution space information on the multi-view image are used to determine whether the down distribution in the set space is uniform from multiple angles. If the down distribution is not uniform and overlaps, it will cause deviation of the three-dimensional reconstruction result and actual result, and further cause error of the down content calculation result. The uniformity of the down distribution is determined by using the above image processing and recognition technology, so as to ensure the accuracy of the subsequent three-dimensional reconstruction result and down content calculation result.

[0050] A down content detection system based on stereovision, comprising a shooting space for dispersing down samples, and further comprising:

[0051] An information acquisition unit, comprising an image acquisition module for obtaining a down multi-view image, an information input module for obtaining an artificial classification result, and an information connection module for obtaining down sample related information in a cloud server database;

[0052] An information storage unit, which is connected with the information acquisition unit and is used for storing a reconstructed down model and related algorithms;

[0053] A central processing unit, which is connected with the information acquisition unit and the information storage unit and is used for obtaining a down content of a down sample by using related algorithms;

[0054] An execution unit, which is connected with the central processing unit and is used for receiving and responding to a control signal output by the central processing unit.

[0055] By adopting the technical scheme, the image acquisition module acquires the down feather multi-view images, the central processing unit calls the three-dimensional reconstruction related algorithm stored in the information storage unit, reconstructs the down feather into a three-dimensional model, compares and classifies the reconstructed down feather model with the related information stored in the cloud server, an artificial can intervene in the classification result through the information input module, and the corresponding parameter data of the down feather model is acquired, and finally the down feather sample content data is obtained. The whole process is mainly realized automatically by the system, greatly reducing the workload of manual sorting, naked eye identification and manual weighing, and greatly improving the efficiency of down feather content detection. The feature comparison of the model is realized by the system, errors caused by low identification level of the operator are avoided, and the accuracy of the down feather content detection result is improved.

[0056] Preferably, the related algorithm comprises:

[0057] An image processing algorithm for acquiring related information of the down feather on the image;

[0058] A three-dimensional reconstruction algorithm for reconstructing the acquired multi-view images into a corresponding reconstructed down feather model;

[0059] A reconstructed model analysis rule for analyzing and matching the reconstructed down feather model with the related information of the down feather sample in the database to obtain analysis data;

[0060] A parameter acquisition algorithm for acquiring parameter data of the corresponding down feather based on the analysis data.

[0061] By adopting the technical scheme, the central processing unit can call the above algorithm to acquire the related information of the down feather on the multi-view image, reconstruct the two-dimensional image into a three-dimensional model, simulate and analyze to obtain the classification result of the down feather, and acquire the parameter data of the reconstructed down feather model, which is high in accuracy.

[0062] Preferably, the execution unit comprises:

[0063] At least two industrial cameras arranged in a shooting space and set at different angles, connected with the image acquisition module, for shooting multi-view images;

[0064] A blower arranged at the bottom of the shooting space for dispersing the down feather sample in the shooting space;

[0065] A vacuum pump connected with a vacuum pipeline, the vacuum pipeline being in communication with the shooting space, for pumping the down feather sample after shooting out of the shooting space;

[0066] A mechanical arm for conveying the down feather sample into the shooting space.

[0067] Through the technical scheme, the subdivision, diffusion, image acquisition and extraction of the down sample can be automatically realized under the control of the central processing unit through the execution unit, the manual operation amount is reduced, and the detection efficiency is improved.

[0068] Preferably, the system further comprises an on-site display module and a remote display module.

[0069] The on-site display module is in data connection with the central processing unit and comprises a display screen for displaying real-time analysis data of the system and a reconstructed down model, so as to facilitate on-site intervention of an operator in analysis and matching results.

[0070] The remote display module is in data connection with the central processing unit and comprises a remote computer control end or a mobile device operation end, so as to facilitate remote intervention of an operator in analysis and matching results.

[0071] Through the technical scheme, the on-site display module facilitates the operator to monitor the classification results and parameter data results on site, and also facilitates the operator to intervene in the classification and matching based on the options provided on the display screen; the remote display module can facilitate experienced operators to simultaneously review and control the sorting results of multiple systems, and further ensure the accuracy of the detection results.

[0072] Compared with the prior art, the application has the following beneficial effects:

[0073] (1) By reducing the down to a three-dimensional model first, the characteristics of the down are maximally retained when the down is data-processed and classified, and the accuracy of the classification results is ensured;

[0074] (2) In manual classification, the down is weighed by using an electronic scale, and the accuracy depends on the accuracy of the electronic scale. In order to ensure the accuracy of the weighing calculation, more down samples need to be sorted, which further increases the workload of the sorting operators. In the application, the parameters of each down are simulated and calculated by using an algorithm, and the accuracy is higher and the efficiency is higher;

[0075] (3) The down content of each down sample is comprehensively obtained, and the down content of the entire down sample is more accurate;

[0076] (4) Through accurate classification and accurate calculation of the down, different dimensional data information of the down sample can be obtained, and the quality of the down can be comprehensively analyzed from multiple dimensions;

[0077] (5) The sorting results are supplemented by automatic simulation and manual intervention, the accuracy of the sorting results is ensured, and the accuracy of the down content detection results is further ensured through pre-dispersion uniformity detection and post-manual review and control. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 This is a schematic diagram illustrating the steps of the fiber content detection method in this application;

[0079] Figure 2 This is a flowchart illustrating step 400 of this application;

[0080] Figure 3 This is a flowchart illustrating the steps of the model reconstruction analysis rules in this application.

[0081] Figure 4 This is a schematic diagram of the modules of the down content detection system of this application.

[0082] Reference numerals in the attached diagram: 1. Information acquisition unit; 2. Information storage unit; 3. Central processing unit; 4. Execution unit; 5. On-site display module; 6. Remote display module. Detailed Implementation

[0083] The present application will be further described in detail below with reference to the embodiments and figures, but the implementation of the present application is not limited thereto.

[0084] like Figure 1 As shown, a method for detecting down content based on stereo vision includes the following steps:

[0085] S100, acquire information related to down samples and information related to down samples in a database. The database includes sub-databases of standard feature information for different types of down, wherein the feature information includes information such as the color, size, and presence of a stem of the down, which can classify down into sub-categories.

[0086] Specifically, taking down as an example, down is divided into goose down and duck down. Goose down is generally larger than duck down and of superior quality. Duck down can be further divided into white duck down and grey duck down, with white duck down being of superior quality. Down clusters have a thin stem at the center, forming a down ball, and another thin stem on one side, forming an umbrella-shaped down ball. By meticulously classifying these standard characteristics, down can be accurately and comprehensively categorized, yielding multi-dimensional data that allows for a more detailed analysis of down quality.

[0087] S200, the down sample is further subdivided into a finite number of down sub-samples. This subdivision can be uniform or non-uniform. In this embodiment, since uniform subdivision has higher requirements, this method uses non-uniform subdivision, that is, it is only necessary to divide the down sample into a finite number of non-overlapping sub-samples that can be dispersed in a set space, and it is not required that the mass of each down sub-sample be equal.

[0088] S300, acquire multi-view images of the down sample within a diffusion set space, and use an industrial camera to capture multi-view images of the down sample within a diffusion set space from multiple angles. The number of industrial cameras is set to at least two. When the number of industrial cameras is set to two, the obtained multi-view images are binocular images.

[0089] S400, the multi-view images are used to perform three-dimensional reconstruction using a set reconstruction algorithm, and a corresponding reconstructed down model is generated. The set algorithm refers to an existing three-dimensional reconstruction algorithm based on multi-view images, such as KinectFusion or the more sophisticated Kintinuous algorithm, which also includes a distortion correction part.

[0090] like Figure 2 As shown, S400 specifically includes the following steps:

[0091] S401, acquire the grayscale information and image distribution spatial information of the multi-view image. The GetPixel function can be used to obtain the color values ​​of pixels sequentially, and the spatial distribution characteristics of the down feathers on the image can be obtained based on the color values ​​of each pixel.

[0092] S402, based on the grayscale information and image distribution spatial information, determine whether the distribution of down sub-samples on the multi-view image is uniform. Set a basic grayscale threshold; when the grayscale value at a certain point in the image exceeds the threshold, down overlap and occlusion occur at that point.

[0093] S403, if the distribution is uniform, perform 3D reconstruction by setting a reconstruction algorithm.

[0094] S404, if the distribution is uneven, re-acquire multi-view images of the down sub-sample within the diffusion setting space, and perform the above judgment.

[0095] S500, obtain parameter data corresponding to each of the reconstructed down models based on the information related to the down samples in the database.

[0096] Image processing technology is used to acquire grayscale information and spatial distribution information from multiple viewpoints, and the information from these multiple viewpoints is compared to determine whether the down distribution within a given space is uniform. If the down distribution is uneven and overlaps, the grayscale value at that location will be greater than that of normally distributed down. Uneven down distribution will cause deviations between the 3D reconstruction results and reality, and will further cause errors in the calculation of down content. By judging the uniformity of down distribution before 3D reconstruction, the accuracy of subsequent 3D reconstruction results and down content calculation results is ensured.

[0097] Specifically, S500 includes the following steps:

[0098] S510, classifying the reconstructed down model based on the sub-class database.

[0099] S511, analyzing the relevant feature information of the reconstructed down model.

[0100] S512, classifying the reconstructed down model based on the preset reconstructed model analysis rule and the relevant feature information of the reconstructed down model.

[0101] As shown in Figure 3 the reconstructed model analysis rule is:

[0102] When the relevant feature information contains standard feature information, obtain high-precision analysis data corresponding to the standard feature information from the database;

[0103] When the high-precision parameter data cannot be obtained, obtain corresponding simulation analysis data through an auxiliary processing scheme. The auxiliary processing scheme at least includes an automatic fitting processing scheme and a manual intervention processing scheme.

[0104] The automatic simulation processing scheme includes the following steps: obtaining corresponding similar models of the reconstructed down model in each of the sub-class databases, respectively, based on each similar model, obtaining simulation analysis data corresponding to the reconstructed down model through a preset simulation algorithm. The reconstructed down model that cannot be classified with high precision can be classified into the sub-class with the highest relevant feature coincidence degree through the simulation algorithm. If the coincidence degree is the same, the average processing is performed, that is, the quality is evenly divided into each similar sub-class.

[0105] The manual intervention processing scheme includes the following steps: obtaining corresponding similar models of the reconstructed down model in each of the sub-class databases, respectively, based on each similar model, manually analyzing the reconstructed down model. Providing similar models in each sub-class database to the manual classification for the reconstructed down model eliminates most interference terms, improves the accuracy and speed of manual classification. Compared with the traditional planar observation under the projection microscope, the similar models are directly presented to the operator, which further improves the accuracy of manual classification.

[0106] In order to continuously expand and correct the down models in each sub-class database to obtain more and more accurate high-precision analysis data, the manual intervention processing scheme further includes the following steps:

[0107] Establishing a supplementary correlation between the reconstructed down model and the manual analysis result;

[0108] Storing the supplementary correlation into the corresponding sub-class database.

[0109] Through the above steps, the artificial intervention classification result is stored in the database as the basis for subsequent feature comparison automatic simulation analysis or high-precision analysis.

[0110] S520, based on the classification result, obtaining parameter data corresponding to the reconstructed down model. The parameter specifically refers to down quality in the embodiments of the present application, and the acquisition method is to calculate the volume of the down based on the reconstructed three-dimensional model

[0111] S600, diffuse the next down sample in the set space and repeat the above steps until the parameter acquisition of a limited number of down samples is completed.

[0112] S700, obtaining the down content of the down sample based on the parameter data of all the down samples.

[0113] By comparing and averaging the down contents of each down sample, the down content of the down sample can be obtained. Compared with the one-time sorting in the prior art, the detection results of multiple times are integrated in the present application, which on the one hand increases the fault tolerance of single detection, and on the other hand ensures the accuracy of the final down sample down content detection result.

[0114] A down content detection system based on stereovision, comprising a shooting space for dispersing down samples, as shown in Figure 4 It also includes an information acquisition unit 1 for acquiring down samples and related information in the cloud database, an information storage unit for storing reconstructed down models and related algorithms, a central processing unit 3 for obtaining the down content of the down sample using related algorithms, and an execution unit 4. The information acquisition unit 1 acquires down multi-view images, the central processing unit 3 calls the related algorithms stored in the information storage unit, reconstructs the down into a three-dimensional model, and then compares and classifies the reconstructed down model with the related information stored in the cloud server. Artificial intervention can be performed on the classification result, and the corresponding parameter data of the down model can be obtained. Finally, the down content data of the down sample is obtained. The whole process is automatically realized under the control of the central processing unit 3 by the execution unit 4, greatly reducing the workload of manual sorting, naked eye identification and manual weighing, and greatly improving the efficiency of down content detection.

[0115] Specifically, as shown in Figure 4As shown, the information acquisition unit 1 includes an image acquisition module for acquiring down multi-view images, an information input module for acquiring artificial classification results, and an information connection module for down sample related information in the cloud server database. The information input module is configured as a man-machine interface to facilitate the operator to view the reconstructed three-dimensional model from multiple angles and accurately intervene in the system sorting results. The information connection module includes but is not limited to a 4G / 5G communication module, a WIFI communication module / GPRS communication module, and a wireless data connection with the cloud server database.

[0116] In this application, the central processing unit 3 is the core unit of the down content detection system, which can be configured as an FPGA module, a single-chip microcomputer module or a customized DSP chip module with a set program. Correspondingly, the information storage unit includes a storage medium itself and an interface for data reading and writing. In this application, the above-mentioned information storage unit is configured as a readable and writable storage chip for storing the reconstructed down model and related algorithms.

[0117] The related algorithms at least include:

[0118] An image processing algorithm for obtaining the related information of down on the image;

[0119] A three-dimensional reconstruction algorithm for reconstructing the obtained multi-view image into a corresponding reconstructed down model;

[0120] A reconstructed model analysis rule for analyzing and matching the reconstructed down model with the down sample related information in the database and obtaining analysis data;

[0121] A parameter acquisition algorithm for obtaining the parameter data of the corresponding down based on the above analysis data.

[0122] The central processing unit 3 calls the above algorithms respectively to perform image recognition analysis, three-dimensional reconstruction, classification matching and parameter acquisition of down.

[0123] The execution unit 4 includes at least two industrial cameras, which are configured as two industrial cameras in the embodiment of the application, and are detachably arranged in the shooting space through bolts and mounting plates, and are arranged at different angles towards the center of the shooting space. The two industrial cameras are connected with the image acquisition module for shooting multi-view images, and are connected with the central processing unit 3 for control, and are synchronized for shooting in response to the control signal output by the central processing unit 3.

[0124] The shooting space is further provided below with a blower, which is arranged towards the center of the shooting space. The shooting space is provided above with a negative pressure pump, which is connected with the shooting space through a vacuum pipeline and is started in response to the control signal output by the central processing unit 3 to draw the down sample out of the shooting space. The shooting space is surrounded by a black backlight plate, which is selected to facilitate the identification and image processing of the white or gray down.

[0125] A sample inlet is formed on the backlight plate at one side of the shooting space, and an automatically opened and closed sealing door is arranged at the sample inlet. A mechanical arm is slidably arranged in the sample inlet and is used to transport the down sample into the shooting space. The above are all prior art and will not be described again.

[0126] A live display module 5 is arranged below the shooting space and includes a display screen for displaying real-time analysis data of the system and a reconstructed down model, which facilitates the operators to monitor the classification results and parameter data results of the down in the whole process and also facilitates the intervention of the classification matching based on the options provided on the display screen.

[0127] In addition to the live display module 5, a remote display module 6 is also configured, which is connected with the central processing unit 3 and includes a remote computer control end or a mobile device operation end, so that experienced operators can simultaneously audit and control the sorting results of multiple systems to further ensure the accuracy of the detection results.

[0128] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the scope of the present application shall be considered as falling within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

Claims

1. A method for detecting the content of down based on stereovision, characterized in that, The method comprises the following steps: acquiring a down sample and information related to the down sample in a database; subdividing the down sample into a plurality of down sub-samples; acquiring multi-view images of the down sub-samples in a diffusion space; reconstructing the multi-view images into a three-dimensional down model using a preset reconstruction algorithm; acquiring parameter data corresponding to the down model based on the information related to the down sample in the database; repeating the steps of acquiring multi-view images and generating a corresponding down model until the parameter data of all the down sub-samples are acquired; acquiring the down content of the down sample based on the parameter data of all the down sub-samples; the step of acquiring parameter data corresponding to the down model based on the information related to the down sample in the database comprises: the database comprises a plurality of sub-databases of standard feature information of different types of down; classifying the down model based on the sub-databases; acquiring parameter data corresponding to the down model based on the classification results.

2. The stereovision-based pile content detection method according to claim 1, characterized in that, the step of classifying the down model based on the sub-databases comprises: analyzing the feature information of the down model; classifying the down model based on a preset analysis rule and the feature information of the down model.

3. The stereovision-based pile content detection method according to claim 2, characterized in that, the preset analysis rule comprises: when the feature information includes standard feature information, acquiring high-precision analysis data corresponding to the standard feature information from the database; when the high-precision analysis data cannot be acquired, acquiring corresponding simulation analysis data through an auxiliary processing scheme.

4. The stereovision-based pile content detection method according to claim 3, characterized in that, the auxiliary processing scheme comprises at least an automatic fitting processing scheme and a manual intervention processing scheme.

5. The stereovision-based pile content detection method according to claim 4, characterized in that, the automatic fitting processing scheme comprises the following steps: acquiring corresponding similar models of the down model in each of the sub-databases; acquiring simulation analysis data corresponding to the down model through a preset simulation algorithm based on the similar models.

6. The stereovision-based pile content detection method according to claim 4, wherein, the manual intervention processing scheme comprises the following steps: acquiring corresponding similar models of the down model in each of the sub-databases; manually analyzing the down model based on the similar models; acquiring simulation analysis data corresponding to the down model based on the results of the manual analysis of the down model according to a preset parameter acquisition algorithm.

7. The stereovision-based pile content detection method according to claim 6, characterized in that, the method further comprises the following steps: establishing a supplementary correlation between the down model and the results of the manual analysis; storing the supplementary correlation in the corresponding sub-database.

8. The stereovision-based pile content detection method according to claim 1, wherein, the step of reconstructing the multi-view images into a three-dimensional down model using a preset reconstruction algorithm comprises: acquiring grayscale information and image distribution space information of the multi-view images; determining whether the distribution of the down sub-samples in the multi-view images is uniform based on the grayscale information and the image distribution space information; if the distribution is uniform, performing three-dimensional reconstruction through a preset reconstruction algorithm; if the distribution is not uniform, reacquiring multi-view images of the down sub-samples in a diffusion space and performing the above determination.

9. A stereovision-based pile content detection system, characterized in that, The system comprises a shooting space for dispersing down sample, and further comprises: an information acquisition unit (1) comprising an image acquisition module for acquiring multi-view images of down, an information input module for acquiring artificial classification results, and an information connection module for acquiring information related to down sample in a cloud server database; an information storage unit connected with the information acquisition unit (1) for storing reconstructed down model and related algorithms; a central processing unit (3) connected with the information acquisition unit (1) and the information storage unit for obtaining the down content of the down sample by using the related algorithms; an execution unit (4) connected with the central processing unit (3) for receiving and responding to the control signal output by the central processing unit (3); the related algorithms comprise: an image processing algorithm for obtaining information related to down on the image; a three-dimensional reconstruction algorithm for reconstructing the obtained multi-view images into corresponding reconstructed down model; a reconstructed model analysis rule for analyzing and matching the reconstructed down model with the information related to down sample in the database to obtain analysis data; a parameter acquisition algorithm for obtaining parameter data of corresponding down based on the analysis data.

10. The stereovision-based pile content detection system according to claim 9, wherein, The execution unit (4) comprises: at least two industrial cameras arranged in the shooting space and set at different angles, connected with the image acquisition module, for shooting multi-view images; an air blower arranged at the bottom of the shooting space for dispersing down sample in the shooting space; a vacuum pump connected with a vacuum pipeline in communication with the shooting space for pumping the down sample out of the shooting space after shooting; a mechanical arm for transporting the down sample into the shooting space.

11. The stereovision-based pile content detection system according to claim 9, wherein, Further comprising an on-site display module (5) and a remote display module (6); the on-site display module (5) connected with the central processing unit (3) comprises a display screen for displaying real-time analysis data and reconstructed down model of the system, so as to facilitate on-site intervention of the analysis and matching results by the operator; the remote display module (6) connected with the central processing unit (3) comprises a remote computer control end or a mobile device operation end, so as to facilitate remote intervention of the analysis and matching results by the operator.

Images