A cabinet segmentation method based on traditional image processing

Abstract

The application provides a cabinet segmentation method based on traditional image processing, belongs to the technical field of cabinet image segmentation, and is based on cabinet images collected in a machine room, so that the area occupied by cabinet equipment in the image is segmented, thereby assisting in machine room data analysis. The method can be applied to machine room inspection robots and other products, can be used alone, and can be combined with a deep learning image segmentation method to improve performance and adaptability, and can be used alone to obtain a cabinet segmentation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cabinet image segmentation, and in particular to a cabinet segmentation method based on traditional image processing. BACKGROUND

[0002] In modern data centers and network facilities, cabinets play a crucial role in housing servers, network devices, and other critical infrastructure. As data center scales and equipment quantities increase, the management and maintenance of the interior of the cabinet become increasingly important. In order to achieve more efficient cabinet management, image segmentation technology is introduced for automated and assisted identification and monitoring of equipment within the cabinet. The cabinet image segmentation method can segment the image within the cabinet into different regions, thereby identifying the equipment and components within each region.

[0003] Cabinet image segmentation technology is a research hotspot in the field of computer vision, and has made some progress in data centers, network operation and maintenance, etc. Commonly used image segmentation techniques mainly include two branches based on traditional image processing and deep learning. Traditional image processing requires less data, is faster, but is difficult to optimize and improve accuracy. Deep learning methods require a large amount of data support, are slower, but have more optimization space and are easy to improve accuracy. In the environment of data analysis in the machine room, cabinet image segmentation is usually used in two scenarios. First, the cabinet image needs to be segmented in real time as reference data for other real-time data analysis. Second, the accurate result of the cabinet image is needed for display and non-real-time analysis. SUMMARY

[0004] To solve the above technical problems, the present application provides a cabinet segmentation method based on traditional image processing. It can be directly used in the field of cabinet image segmentation, and the segmentation method can be combined with deep learning image segmentation method to further improve the performance of the cabinet segmentation method, thereby improving the adaptability of the algorithm. The image segmentation method based on deep learning can be run alone to obtain the cabinet segmentation result.

[0005] The technical solution of the present application is:

[0006] In this method, a cabinet segmentation method based on traditional image processing is proposed. First, this method can be directly used in the field of cabinet image segmentation. Second, the segmentation method can be combined with the deep learning image segmentation method to further improve the performance of the cabinet segmentation method, thereby improving the adaptability of the algorithm. The image segmentation method based on deep learning can be run alone to obtain the cabinet segmentation result.

[0007] A cabinet segmentation method based on traditional image processing, comprising:

[0008] Image acquisition, the image acquisition device acquires a picture containing a cabinet;

[0009] The image processing analysis on the picture containing the cabinet includes cabinet picture texture extraction, texture analysis, texture statistics, finding out the cabinet equipment area according to the statistical result, and outputting the cabinet equipment area information;

[0010] The cabinet picture texture extraction includes an image-based edge extraction method, including a sobel operator, a loG operator, a Prewitt operator, a Roberts operator, a Scharr operator and a Canny edge extraction method;

[0011] The texture analysis includes screening the picture texture, including screening the candidate texture area by the connected domain size, the connected domain shape and the texture direction of the texture;

[0012] The texture statistics include the number of candidate texture areas, the position in the image, the number of pixels in the image, and the statistical data obtained by the statistics;

[0013] The statistical data is summarized, and the boundary information of the equipment object is obtained by using data threshold analysis and data value voting;

[0014] The boundary information is processed, the area of the cabinet equipment is output, and the rectangular frame or the cabinet area mask is represented, which provides available information for other data analysis steps;

[0015] Further,

[0016] It can be used alone for data analysis in the machine room, for displaying the cabinet equipment area, and for statistics of the presence or absence of equipment in the machine room;

[0017] It can be used as an input for other machine room data analysis steps, such as equipment detection, equipment model identification, equipment alarm light identification, equipment number identification, and cabinet picture splicing;

[0018] Further,

[0019] It can be used in combination with deep learning-based image methods to improve the applicability and accuracy of the algorithm;

[0020] The deep learning-based image segmentation method includes U-Net, DeepLab, Mask-RCNN, FCN, PSPNet, HRNet, AttentionMechanissm method, etc.;

[0021] The deep learning-based image segmentation method needs an image as a training set, which is labeled and pre-trained with a model, and after obtaining the model, it is used for subsequent segmentation of the cabinet image;

[0022] The deep learning-based image segmentation method needs an image as input and outputs a format same as the traditional image processing-based cabinet segmentation method;

[0023] The joint use method includes inputting the result of the traditional image processing-based cabinet segmentation method into the deep learning-based image method, and integrating the results of the two methods;

[0024] The result of the traditional image processing-based cabinet segmentation method is input into the deep learning-based image method, which includes inputting the image mask result as the input of the deep learning method, and obtaining accurate segmentation results through the deep learning-based image segmentation method;

[0025] The deep learning-based image segmentation method can be operated alone to obtain the cabinet segmentation result;

[0026] The integration of the two methods includes that the two methods are operated respectively to obtain the segmentation results, and the intersection or union of the results is obtained as the final segmentation result.

[0027] The beneficial effects of the present application are

[0028] Firstly, the method can be directly used in the field of cabinet image segmentation; secondly, the segmentation method can be combined with the deep learning image segmentation method to further improve the performance of the cabinet segmentation method, thereby improving the adaptability of the algorithm. The method can also be applied to other equipment in the computer room, such as computer room monitoring video analysis. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a working schematic diagram of the present application. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0031] In the present method, a traditional image processing-based cabinet segmentation method is proposed; firstly, the method can be directly used in the field of cabinet image segmentation; secondly, the segmentation method can be combined with the deep learning image segmentation method to further improve the performance of the cabinet segmentation method, thereby improving the adaptability of the algorithm. The deep learning-based image segmentation method can be operated alone to obtain the cabinet segmentation result. As shown in Figure 1 , the method includes the following steps;

[0032] Image acquisition, the image acquisition device acquires a picture containing the cabinet, the acquisition device is a vision system contained in the robot, mainly related to the shooting of 2D image;

[0033] The image processing analysis of the picture containing the cabinet includes: cabinet picture texture extraction, texture analysis, texture statistics, finding out the cabinet equipment area according to the statistical results, and outputting the cabinet equipment area information;

[0034] The cabinet picture texture extraction includes an edge extraction method based on image, including sobel operator, loG operator, Prewitt operator, Roberts operator and Scharr operator and Canny edge extraction method;

[0035] In this paper, the Canny edge extraction method is used to process the image to obtain the edge region;

[0036] The texture analysis includes screening the picture texture, including screening the candidate texture region by the connected domain size, connected domain shape and texture direction of the texture;

[0037] The connected domain size screening removes the region whose connected domain pixel is more than the threshold Max or less than the threshold Min;

[0038] The connected domain shape aspect ratio that does not meet the condition needs to be removed;

[0039] The texture direction that does not meet the slope threshold needs to be removed;

[0040] The texture statistics includes the number of candidate texture regions, the position in the image, and the number of pixels in the image, to obtain the statistical data;

[0041] For the region with too few textures, remove the image region;

[0042] The number of pixels in the image is counted to obtain the statistical value;

[0043] The statistical data is summarized, and the data threshold analysis and data value voting are used to obtain the boundary information of the device object;

[0044] If the statistical value obtained by counting the number of pixels in the image is greater than the threshold H, the voting information is increased, and if it is less than the threshold, the voting information is reduced;

[0045] If the voting information of the candidate region is greater than T, it is the edge region of the device, and the middle line of the candidate region is calculated as the edge of the cabinet equipment, and the boundary information of the device is output according to the edge;

[0046] The boundary information is processed to output the area of the cabinet equipment, and a rectangular frame or a cabinet area mask is used to represent the area, thereby providing available information for other data analysis steps.

[0047] The rectangular frame includes two forms, one is the rectangular frame in the upper left corner coordinate of the image and the length and width of the rectangular frame, and the other is the rectangular frame in the upper left corner coordinate and the lower right corner coordinate of the image.

[0048] The cabinet area mask refers to a single-channel image of the same size as the initial image input, wherein the area of the equipment is greater than zero, and the area of the equipment is equal to zero.

[0049] The cabinet segmentation method based on traditional image processing can be used alone in the data analysis of the machine room, and is used to display the cabinet equipment area, count the information of the presence or absence of the equipment in the machine room, and the like.

[0050] The cabinet segmentation method based on traditional image processing can be used as the input of other machine room data analysis steps, such as equipment detection, equipment model identification, equipment alarm light identification, equipment number identification, and cabinet picture splicing.

[0051] The cabinet segmentation method based on traditional image processing can be used in combination with the image method based on deep learning to improve the applicability and accuracy of the algorithm.

[0052] The image segmentation method based on deep learning includes U-Net, DeepLab, Mask-RCNN, FCN, PSPNet, HRNet, and AttentionMechanissm methods.

[0053] The image segmentation method based on deep learning requires images as a training set, marking and model pre-training, and after obtaining the model, the model is used for subsequent segmentation of the cabinet image.

[0054] The image segmentation method based on deep learning requires images as input, and the output is in the same format as the cabinet segmentation method based on traditional image processing.

[0055] The Mask-RCNN method is used in the embodiments of the present application.

[0056] The combined use method includes inputting the result of the cabinet segmentation method based on traditional image processing into the image method based on deep learning, and also includes integration of the results of the two methods.

[0057] The result of the cabinet segmentation method based on traditional image processing is input into the image method based on deep learning, which includes inputting the image mask result as the input of the deep learning method, and obtaining accurate segmentation results through the image segmentation method based on deep learning.

[0058] The integration of the two methods includes running the segmentation results of the two methods respectively, performing intersection or union of the results as the final segmentation result.

[0059] The above merely describes the preferred embodiments of the present application, which are used for illustrating the technical solutions of the present application, but not used for limiting the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A cabinet segmentation method based on traditional image processing, characterized in that, Image acquisition: Image acquisition equipment captures images including those of the server rack. Image processing and analysis of images containing server racks includes: texture extraction from server rack images, texture analysis, texture statistics, identifying server rack equipment areas based on statistical results, and outputting server rack equipment area information; in, Rack image texture extraction includes image-based edge extraction methods, including Sobel operator, LoG operator, Prewitt operator, Roberts operator, Scharr operator, and Canny edge extraction methods; Texture analysis includes filtering image textures, such as filtering candidate texture regions by the size, shape, and direction of connected components. Texture statistics include counting the number of candidate texture regions, their location in the image, and the number of pixels they occupy in the image to obtain statistical data; The statistical data is summarized, and the boundary information of the device object is obtained by using data threshold analysis and data value voting. The boundary information is processed to output the area of ​​the cabinet equipment, and represented in the form of a rectangle or a cabinet area mask to provide usable information for other data analysis steps; Deep learning-based image segmentation methods include U-Net, DeepLab, Mask-RCNN, FCN, PSPNet, HRNet, and Attention Mechanism. Images are needed as a training set for labeling and model pre-training. Once the model is obtained, it will be used for subsequent segmentation of cabinet images. An image is required as input, and the output format is the same as that of the segmented output of the rack image; This includes running the segmentation results using two methods, and then finding the intersection or union of the results to obtain the final segmentation result; If the statistical value obtained by counting the number of pixels in an image is greater than the threshold H, then voting information is added; if it is less than the threshold, voting information is reduced. For candidate regions with voting information greater than T, the edge region of the device is determined. The median line of the candidate region is used as the edge of the cabinet device, and the boundary information of the device is output based on the edge. There are two types of rectangles: one is the rectangle's coordinates at the top left corner of the image and its length and width, and the other is the rectangle's coordinates at the top left corner and bottom right corner of the image. A rack area mask is a single-channel image of the same size as the initial image input, in which areas that are devices are set to values ​​greater than zero, and areas that are not devices are set to values ​​equal to zero.

2. The method according to claim 1, characterized in that, Used alone in data analysis of computer rooms, it is used to display the equipment area of ​​the server rack and to provide information on the presence or absence of equipment in the computer room.

3. The method according to claim 1, characterized in that, As input for other data analysis steps in the data center, it includes equipment detection, equipment model identification, equipment alarm light identification, equipment number identification, and rack image stitching.

4. The method according to claim 1, characterized in that, It can be used in conjunction with deep learning-based segmentation methods to improve the applicability and accuracy of the algorithm.

5. The method according to claim 1, characterized in that, The results of the rack segmentation method are input into a deep learning-based image method, including results from both methods: The results of the rack segmentation method are input into the deep learning-based image method, including using the image mask results as input to the deep learning method. After passing through the deep learning-based image segmentation method, accurate segmentation results are obtained. The image segmentation method based on deep learning was run alone to obtain the cabinet segmentation results.

Images