Escalator drawing design size extraction method

Abstract

The invention relates to the technical field of key point detection, and discloses an escalator drawing design size extraction method, which comprises the following steps: acquiring an escalator drawing, and generating a to-be-detected image according to the escalator drawing; a key point detection data set is constructed, and key point samples in the key point detection data set are composed of real positions, visible conditions and key point types of the key points; performing model training through the key point detection data set to obtain a key point detection model; performing key point detection on the to-be-detected image through the key point detection model to obtain position information of key points; and determining the design size of the escalator drawing according to the position information of the key points and the types of the key points. The key point detection of the escalator drawing can be realized, the key point detection efficiency and accuracy are high, the design size of the escalator drawing is further extracted, and the efficiency and accuracy of the design size extraction of the escalator drawing are effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of key point detection, and in particular to a method for extracting design dimensions from escalator drawings. Background Art

[0002] When conducting a design review of escalator drawings, it is necessary to extract the key parts of the escalator and the design dimensions between the escalator and the building. Since the shapes of the escalator drawings and the surrounding buildings are different, the design dimensions that need to be extracted are also different, and it is impossible to directly use algorithms for detection and identification. For example, the lifting height of the escalator cannot be directly detected by the algorithm. However, if the positions of the upper working point and the lower working point of the escalator are extracted, the design dimensions of the lifting height can be calculated. Therefore, the extraction task of the escalator design dimensions can be transformed into the detection task of the key points corresponding to the design dimensions. However, the existing key point detection method is aimed at targets of fixed shapes, such as human bodies, faces, and hands. Taking the human body as an example, the type and number of key points of each target are the same, and there is only one key point of each type. In the escalator drawings, the number of key points is not fixed. For example, the number of support points and hook points of the escalator varies with the design of the escalator. Therefore, the existing key point detection algorithm cannot be applied to the key point detection task of the escalator drawings, resulting in low efficiency in extracting the design dimensions of the escalator drawings.

[0003] Therefore, how to realize the key point detection of escalator drawings and effectively improve the efficiency and accuracy of escalator drawing design dimension extraction is a problem that needs to be solved urgently.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a method for extracting design dimensions from escalator drawings, aiming to solve the technical problem of how to realize key point detection of escalator drawings, thereby effectively improving the efficiency and accuracy of escalator drawing design dimension extraction.

[0006] To achieve the above purpose, the present application proposes a method for extracting design dimensions of escalator drawings, the method comprising:

[0007] Obtaining an escalator drawing, and generating an image to be detected according to the escalator drawing;

[0008] Constructing a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key point;

[0009] Performing model training using the key point detection data set to obtain a key point detection model;

[0010] Perform key point detection on the image to be detected by using the key point detection model to obtain position information of the key points;

[0011] The design dimensions of the escalator drawing are determined according to the position information of the key points and the types of the key points.

[0012] In one embodiment, the step of obtaining an escalator drawing and generating an image to be detected according to the escalator drawing includes:

[0013] Acquire an escalator drawing, and convert the format of the escalator drawing to obtain an escalator image in a target format;

[0014] Detecting the location and type of the escalator in the escalator image by a target detection algorithm;

[0015] determining a trimming area according to the location and type of the escalator;

[0016] Clipping the escalator image according to the clipping area to obtain a clipped escalator image;

[0017] The cropped escalator image is scaled to a preset pixel value to obtain an image to be detected.

[0018] In one embodiment, constructing a key point detection data set includes:

[0019] Annotate the escalator drawing using a picture annotation tool to obtain annotated key points;

[0020] Obtaining the data coordinates, visibility and key point types of the marked key points, and extracting the data coordinates of all feature points in the escalator drawing;

[0021] Determine the real position of the marked key point according to the data coordinates of the marked key point and the data coordinates of all the feature points;

[0022] The actual position, visibility and key point type of the marked key point are taken as key point samples, and a key point detection dataset is constructed based on the key point samples.

[0023] In one embodiment, the performing model training using the key point detection data set to obtain a key point detection model includes:

[0024] Build an initial detection model based on the human posture estimation model;

[0025] Dividing the key point detection data set according to a preset ratio to obtain a training set and a test set;

[0026] Training the initial key point detection model using the training set to obtain a trained initial key point detection model;

[0027] Testing the trained initial key point detection model using the test set to obtain an average accuracy;

[0028] When the average accuracy reaches a preset accuracy threshold, the trained initial key point detection model is used as a target key point detection model.

[0029] In one embodiment, testing the trained initial key point detection model using the test set to obtain an average accuracy includes:

[0030] Inputting the key point samples in the test set into the trained initial key point detection model to obtain the predicted positions of the key points;

[0031] Determining key point similarity according to the predicted position, the actual position, the key point type, and the visibility condition;

[0032] An average precision is determined based on the keypoint similarities.

[0033] In one embodiment, determining the key point similarity according to the predicted position, the actual position, the key point type and the visibility condition includes:

[0034] Determine the Euclidean distance, normalization factor and scale factor corresponding to each type of key point according to the predicted position, the actual position, the key point type and the visibility condition;

[0035] Key point similarity is determined based on the Euclidean distance, the normalization factor, and the scale factor.

[0036] In one embodiment, the step of performing key point detection on the image to be detected by using the key point detection model to obtain position information of the key points includes:

[0037] Extracting features of the image to be detected to obtain a first feature map, a second feature map, and a third feature map;

[0038] Inputting the first feature map into the backbone network of the key point detection model to obtain a first probability map;

[0039] Performing a deconvolution operation on the first probability map to obtain a second probability map;

[0040] Fusing the second feature map and the second probability map through the residual connection structure of the key point detection model to obtain a fused second probability map;

[0041] Performing a deconvolution operation on the fused second probability map to obtain a third probability map;

[0042] Fusing the third feature map with the third probability map to obtain a target probability map;

[0043] The position information of the key points is determined according to the target probability map.

[0044] In addition, to achieve the above purpose, the present application also proposes an escalator drawing design size extraction device, the escalator drawing design size extraction device comprising:

[0045] An acquisition module, used for acquiring an escalator drawing and generating an image to be detected according to the escalator drawing;

[0046] A construction module is used to construct a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key point;

[0047] A training module, used to perform model training using the key point detection data set to obtain a key point detection model;

[0048] A detection module, used to perform key point detection on the image to be detected by using the key point detection model to obtain position information of the key points;

[0049] The determination module is used to determine the design size of the escalator drawing according to the position information and key point type of the key point.

[0050] In addition, to achieve the above-mentioned purpose, the present application also proposes an escalator drawing design dimension extraction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the escalator drawing design dimension extraction method as described above.

[0051] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the escalator drawing design dimension extraction method as described above are implemented.

[0052] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the method for extracting design dimensions of escalator drawings as described above are implemented.

[0053] The present application provides a method for extracting design dimensions of an escalator drawing. The present application first obtains an escalator drawing and generates an image to be detected based on the escalator drawing; constructs a key point detection data set, wherein key point samples in the key point detection data set are composed of the real position, visible situation and key point type of the key point; performs model training through the key point detection data set to obtain a key point detection model; performs key point detection on the image to be detected through the key point detection model to obtain the position information of the key points; determines the design dimensions of the escalator drawing based on the position information and key point type of the key points, thereby realizing key point detection of the escalator drawing, and the key point detection efficiency and accuracy are high, thereby effectively improving the efficiency and accuracy of extracting design dimensions of the escalator drawing.

[0054] In summary, the present application performs key point detection on the image to be detected generated based on the escalator drawing through a key point detection model, obtains the position information of the key points, and then determines the design size of the escalator drawing, thereby overcoming the defect that the existing key point detection algorithm cannot be applied to the key point detection task of the escalator drawing, resulting in low efficiency and accuracy in the extraction of the design size of the escalator drawing. The present application can realize the key point detection of the escalator drawing, and the key point detection efficiency and accuracy are high, thereby effectively improving the efficiency and accuracy of the extraction of the design size of the escalator drawing. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0056] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0057] Figure 1 A flow chart of the first embodiment of the method for extracting design dimensions of an escalator drawing of the present application;

[0058] Figure 2 A flow chart of the second embodiment of the method for extracting the design dimensions of an escalator drawing of the present application;

[0059] Figure 3 A schematic diagram of the structure of the human body posture estimation model HigherHRNet provided in Example 2 of the method for extracting design dimensions of an escalator drawing of the present application;

[0060] Figure 4A schematic diagram of the structure of the DrawingHRNet network structure model provided in Embodiment 2 of the method for extracting design dimensions of escalator drawings of the present application;

[0061] Figure 5 This is a schematic diagram of the module structure of the device for extracting dimensions from escalator drawing design according to an embodiment of the present application;

[0062] Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the method for extracting design dimensions from escalator drawings in an embodiment of the present application.

[0063] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0064] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0065] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0066] The main solution of the embodiment of the present application is: obtaining an escalator drawing, and generating an image to be detected based on the escalator drawing; constructing a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visible situation and key point type of the key point; performing model training through the key point detection data set to obtain a key point detection model; performing key point detection on the image to be detected through the key point detection model to obtain the position information of the key points; and determining the design size of the escalator drawing according to the position information and key point type of the key points.

[0067] When conducting a design review of an escalator drawing, it is necessary to extract the key parts of the escalator and the design dimensions between the escalator and the building. Since the shapes of the escalator drawings and the surrounding buildings are different, the design dimensions that need to be extracted are also different, and it is impossible to directly use algorithms for detection and identification. For example, the lifting height of the escalator cannot be directly detected by the algorithm. However, if the positions of the upper working point and the lower working point of the escalator are extracted, the design dimensions of the lifting height can be calculated. However, the existing key point detection method is aimed at targets of fixed shapes, such as human bodies, faces, and hands. Taking the human body as an example, the type and number of key points of each target are consistent, and there is only one key point of each type. In the escalator drawings, the number of key points is not fixed. For example, the number of support points and hook points of the escalator varies with the design of the escalator. Therefore, the existing key point detection algorithm cannot be applied to the key point detection task of the escalator drawings, resulting in low efficiency and accuracy in the extraction of the design dimensions of the escalator drawings. Therefore, how to realize the key point detection of the escalator drawings and effectively improve the efficiency and accuracy of the extraction of the design dimensions of the escalator drawings is a problem that needs to be solved urgently.

[0068] The present application performs key point detection on an image to be detected generated based on an escalator drawing through a key point detection model, obtains the position information of the key points, and then determines the design dimensions of the escalator drawing. This overcomes the technical defect that the existing key point detection algorithm cannot be applied to the key point detection task of the escalator drawing, resulting in low efficiency and accuracy in the extraction of the design dimensions of the escalator drawing. The application can realize key point detection of the escalator drawing, and the key point detection efficiency and accuracy are high, thereby effectively improving the efficiency and accuracy of the extraction of the design dimensions of the escalator drawing.

[0069] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an escalator drawing design dimension extraction device, etc. The following takes the escalator drawing design dimension extraction device as an example to illustrate this embodiment and the following embodiments.

[0070] Based on this, the embodiment of the present application provides a method for extracting the design dimensions of an escalator drawing, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the method for extracting design dimensions of an escalator drawing of the present application.

[0071] In this embodiment, the method for extracting design dimensions of escalator drawings includes steps S10 to S50:

[0072] Step S10, obtaining an escalator drawing, and generating an image to be detected according to the escalator drawing.

[0073] It should be noted that the escalator drawing refers to a detailed design drawing used to describe and plan the installation and design of the escalator, which includes the various components of the escalator and their relative position relationship with the surrounding environment. The escalator drawing can be a paper drawing or an electronic drawing. This embodiment does not specifically limit this. This embodiment takes the electronic drawing as an example for explanation.

[0074] It is understandable that the escalator drawing can be a CAD drawing, a PDF document or any other electronic file format containing escalator design information, and this embodiment does not specifically limit this. For subsequent processing, the escalator drawing needs to be converted into an image format that can be processed and recognized by a computer by scanning, converting or directly reading.

[0075] In a feasible implementation, step S10 may include steps A11 to A15:

[0076] Step A11, obtaining an escalator drawing, and converting the format of the escalator drawing to obtain an escalator image in a target format.

[0077] It should be noted that, in this embodiment, the escalator drawing is a CAD drawing, that is, the format of the escalator drawing is dwg format, and it is impossible to directly use computer vision methods to extract information. Therefore, the escalator drawing needs to be converted into a format.

[0078] It is understandable that the target format may be PNG format, JPEG format or any other general format suitable for image processing, and this embodiment does not specifically limit this, and this embodiment uses PNG format as an example for illustration. PNG format can better preserve the lines, text and shape information in the drawing, and is convenient for subsequent image processing and key point detection.

[0079] In the specific implementation, the image plug-in is used to convert the corresponding area in the dwg format CAD escalator drawing into a PNG format image according to the specified range selected by the user, and the starting and ending coordinates of the area, as well as the coordinates of all points in the area, are uniformly saved as json data. Among them, the image plug-in is an image conversion tool that converts dwg format to PNG format developed using .NET secondary development technology.

[0080] Step A12: Detect the position and type of the escalator in the escalator image by using a target detection algorithm.

[0081] It should be noted that the target detection algorithm refers to an algorithm that can identify specific objects and their positions in an image, such as using a convolutional neural network (CNN) for target detection. In the present embodiment, the target detection algorithm is used to identify the escalator area in the escalator image and determine its type. Since the escalator to be extracted accounts for too small a proportion in the drawing, if key point detection is performed directly, it is limited by hardware conditions. Therefore, the image needs to be scaled to a certain pixel. However, direct scaling will cause blurred lines in the drawing, and the key point position of the escalator cannot be accurately detected. Therefore, it is necessary to use a target detection algorithm to detect the position and type of the escalator in the generated PNG format escalator image, and then cut and scale it.

[0082] It is understandable that the image preprocessing technology can be used to enhance the contrast and clarity of the escalator image to highlight the difference between the escalator lines and the background, for example, through grayscale, threshold segmentation, edge enhancement and other operations to improve the image quality.

[0083] Step A13, determining a trimming area according to the location and type of the escalator.

[0084] It should be noted that an initial rectangular cropping box is marked on the escalator image based on the position information of the escalator detected by the object detection algorithm. This cropping box should surround the escalator as closely as possible while avoiding cropping out any key design elements such as handrails, hooks, etc.

[0085] Step A14, cropping the escalator image according to the cropping area to obtain a cropped escalator image.

[0086] It should be noted that the escalator image is cropped according to the cropping area to remove redundant background information and focus on the design details of the escalator itself. The cropped image should clearly show the key parts of the escalator and related buildings, such as handrails, hooks, etc., to provide high-quality image materials for subsequent design size extraction.

[0087] Step A15, scaling the cropped escalator image to a preset pixel value to obtain an image to be detected.

[0088] It should be noted that the preset pixel value can be 500×500 pixels. Under this preset pixel value, the image to be detected can clearly present all the key features of the escalator, while reducing unnecessary calculations and improving processing speed. The scaling operation should use a suitable interpolation method. Experiments have shown that compared with bilinear interpolation and bicubic interpolation, regional interpolation can better preserve the clarity of lines and the accuracy of key features in the drawing image.

[0089] Step S20, constructing a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key point.

[0090] It should be noted that key points usually represent important structural features on escalators and related buildings, such as support points, hook points, upper working points, lower working points, etc. They are crucial for subsequent dimension measurement, type identification and design analysis. The real position of the key point refers to the real coordinate position corresponding to the key point in the drawing or the cropped escalator image. The visibility of the key point reflects its visibility in the image, which may be caused by factors such as occlusion, change in viewing angle or image quality, including visible and invisible. The key point type is pre-divided according to the function of the key point. For example, all support points and all hook points of the escalator are each a category of key points. After the division, the type of the escalator key point is fixed, and the number of key points in each category is not fixed. For example, after the escalator is divided, there are 13 categories of key points, and there are several key points in each category.

[0091] It can be understood that the key point detection data set is composed of a large number of key point samples, and each key point sample is obtained by processing key point annotation and real position calculation on CAD drawings.

[0092] In a feasible implementation manner, step S20 may include steps B11 to B14:

[0093] Step B11, annotating the escalator drawing by using a picture annotation tool to obtain annotated key points.

[0094] It should be noted that an image annotation tool refers to a software tool that can assist users in accurately marking specific elements or areas on an image, such as labelme annotation software. Labelme annotation software is used to annotate key points on escalator drawings. However, the positions of key points annotated manually often cannot completely fit the true positions of the key points, thus affecting the performance of the trained algorithm. Therefore, it is necessary to further calculate the true positions of the annotated key points.

[0095] Step B12, obtaining the data coordinates, visibility and key point types of the marked key points, and extracting the data coordinates of all feature points in the escalator drawing.

[0096] It should be noted that the data coordinates of the annotated key points refer to the pixel coordinates. In the image annotation process, the pixel coordinates are obtained by the user directly clicking or dragging the marking point in the image annotation tool. The visibility of the key points is annotated by the annotator based on the actual visibility of the key points in the drawing, such as visible or invisible. The key point type is a label assigned to each key point by the annotator based on the predefined key point classification system.

[0097] It can be understood that when the user selects the drawing of the area corresponding to the specified range for format conversion, the starting and ending coordinates of the area and the coordinates of all points in the area will be saved as json data. By extracting the saved json data, the data coordinates of all feature points in the escalator drawing can be obtained.

[0098] Step B13, determining the real position of the marked key point according to the data coordinates of the marked key point and the data coordinates of all the feature points.

[0099] It should be noted that in order to solve the problem that the positions of key points annotated manually cannot completely fit the real positions of key points, thus affecting the performance of the trained algorithm, the coordinate data of all points in the drawing are extracted in the CAD drawing information extraction part, combining the characteristics of drawing key point detection. After the key points are annotated with labelme, the distance between the annotated key points and each point in the drawing is calculated, and the point with the smallest distance is taken as the real value of this annotated point.

[0100] It can be understood that the distance between the marked key point and each feature point is calculated based on the data coordinates of the marked key point and the data coordinates of all feature points, and the data coordinates of the feature point with the smallest distance are used as the true value of this marked point.

[0101] Step B14: taking the real position, visibility and key point type of the marked key point as key point samples, and constructing a key point detection data set based on the key point samples.

[0102] It should be noted that, in conventional key point detection datasets, such as the COCO2017 dataset, since there is only one key point of each type in the target, the key point format in its label is set to [x, y, v, x, y, v, ...], each group of x, y, v represents a key point sample, x, y represents the key point position, v represents whether the key point is visible, and the position of the key point in the array represents the type of the key point. However, the number of key points of each type in the escalator drawings is different and there is more than one, so the key point label format of the conventional dataset cannot be used. In this embodiment, a new key point label format [x, y, v, k, x, y, v, k, ...] is proposed, each group of x, y, v, k represents a key point sample, x, y, v represents the key point position and whether it is visible, and k represents the key point type.

[0103] Step S30, performing model training using the key point detection data set to obtain a key point detection model.

[0104] It should be noted that the initial detection model is trained using a key point detection data set, and the performance indicators of the trained model are evaluated. When the model performance meets the requirements, a key point detection model is obtained.

[0105] It is understandable that during the model training process, the key point detection dataset needs to be preprocessed to ensure data quality and consistency. The preprocessing steps include data cleaning, data enhancement, and data standardization. Data cleaning mainly removes invalid and erroneous data to avoid negative impact on model training. Data enhancement increases data diversity and improves the generalization ability of the model through methods such as rotation, scaling, and translation. Data standardization ensures that the input data is within the same scale range, which helps the model converge faster.

[0106] It is worth noting that in the process of model training, it is also necessary to select a suitable loss function and optimization algorithm. The loss function is used to measure the difference between the model prediction value and the true value. The optimization algorithm is used to adjust the model parameters to minimize the loss function, for example, stochastic gradient descent (SGD), Adam and RMSprop, etc., which are not specifically limited in this embodiment.

[0107] In addition, to further improve model performance, regularization techniques such as L1 and L2 regularization can be used to prevent model overfitting. Early stopping can also be used to avoid overfitting, that is, stop training when the performance on the validation set no longer improves.

[0108] After the model training is completed, the model needs to be evaluated. The evaluation indicators include average precision (AP), which can be used to fully understand the performance of the model on the key point detection task. If the model performance does not meet the expected requirements, the model structure, parameters or training strategy can be adjusted, and retraining and evaluation can be performed until the performance indicators are met.

[0109] Step S40: performing key point detection on the image to be detected by using the key point detection model to obtain position information of the key points.

[0110] It should be noted that this application designs a key point detection model for the key point detection task of amorphous objects in drawings, and the model is named DrawingHRNet. The key point detection model will receive the image to be detected as input, and use its internally learned feature representation and mapping relationship to automatically identify and mark the precise location and type of key points in the image. This process involves image processing and deep learning algorithms, including but not limited to feature extraction in convolutional neural networks, the application of attention mechanisms, and the implementation of regression or classification techniques.

[0111] It is understandable that in order to improve the accuracy and robustness of key point detection, non-maximum suppression (NMS) can be used to eliminate redundant detection points, key point positions can be fine-tuned to optimize positioning accuracy, or graphical models can be used to optimize the structural relationship between key points, etc. This embodiment does not impose specific restrictions on this.

[0112] In a feasible implementation, step S40 may include steps C11 to C17:

[0113] Step C11, extracting features from the image to be detected to obtain a first feature map, a second feature map, and a third feature map.

[0114] It should be noted that the first feature map, the second feature map and the third feature map are feature maps of 1 / 4, 1 / 2 and 1 / 1 of the size of the image to be detected, respectively. The image to be detected is subjected to feature extraction through three layers of convolution to obtain feature maps of three sizes, namely, feature maps of 1 / 4, 1 / 2 and 1 / 1 of the size of the image to be detected, respectively.

[0115] Step C12: input the first feature map into the backbone network of the key point detection model to obtain a first probability map.

[0116] It should be noted that the first feature map is a feature map of 1 / 4 size, and the first probability map is a probability map of 1 / 4 size. The feature map of 1 / 4 size is input into the backbone network of the key point detection model to obtain a probability map of 1 / 4 size.

[0117] It is understandable that the backbone network of the key point detection model uses a high-resolution network HRNet. Feature extraction through the high-resolution network HRNet can effectively retain the high-resolution information of the image and extract rich semantic features to obtain the first feature map. The structural design of HRNet enables it to extract and fuse features at different scales at the same time, thus performing well in the key point detection task.

[0118] Step C13, performing a deconvolution operation on the first probability map to obtain a second probability map;

[0119] It should be noted that the second probability map, i.e., the probability map of size 1 / 2, performs a deconvolution operation on the probability map of size 1 / 4 to obtain a probability map of size 1 / 2. Deconvolution, also known as transposed convolution, achieves upsampling by mapping a smaller input to a larger output.

[0120] Step C14, fusing the second feature map and the second probability map through the residual connection structure of the key point detection model to obtain a fused second probability map;

[0121] It should be noted that the second feature map, that is, the 1 / 2-sized feature map, is fused with the 1 / 2-sized feature map and the 1 / 2-sized probability map through the residual connection structure of the key point detection model to obtain a fused 1 / 2-sized probability map.

[0122] It can be understood that by fusing the 1 / 2-sized feature map and the 1 / 2-sized probability map obtained through the deconvolution operation, feature information of different scales can be fully utilized to enhance the model's ability to detect key points.

[0123] In the specific implementation, the residual connection structure adds the 1 / 2 size feature map initially generated by the network and the 1 / 2 size probability map generated by deconvolution, thereby integrating the shallow features with clear line information with the deep features containing global information, so that the network can obtain the category and approximate position of the key points based on the global information, and the network can obtain the precise position of the key points based on the local information.

[0124] Step C15: performing a deconvolution operation on the fused second probability map to obtain a third probability map.

[0125] It should be noted that the third probability map, that is, the probability map of size 1 / 1, is obtained by performing a deconvolution operation on the fused probability map of size 1 / 2 to obtain a probability map of size 1 / 1.

[0126] It can be understood that in order to further improve the accuracy of key point position detection, a deconvolution operation is performed again on the basis of the probability map of 1 / 2 size to generate a probability map of 1 / 1 size.

[0127] Step C16, fusing the third feature map with the third probability map to obtain a target probability map;

[0128] It should be noted that the third feature map, that is, the feature map of size 1 / 1, fuses the feature map of size 1 / 1 and the probability map of size 1 / 1 to obtain the target probability map.

[0129] It can be understood that the 1 / 1 size probability map is added to the 1 / 1 size feature map initially generated by the network through the residual connection to obtain the final probability map, that is, the target probability map.

[0130] Step C17, determining the position information of the key points according to the target probability map.

[0131] It should be noted that the probability map (Heatmap), also known as a heat map, is used to describe the position of key points in the image. The probability map is a two-dimensional matrix whose size is the same as the input image or reduced by a certain proportion. In this matrix, the value at each position represents the probability that the position is a certain joint point. This probability value usually takes a value between 0 and 1. The higher the value, the more likely it is that the position is the position of the key point. A multi-resolution supervision strategy can be used to calculate the mean square error between all the generated resolution probability maps and the corresponding GT probability maps, thereby obtaining the loss function, and optimizing the model parameters through the back propagation algorithm to improve the accuracy of key point detection. Finally, the specific location information of the key point is determined based on the position with the highest probability value in the probability map.

[0132] It is understandable that after the probability map is generated, the probability map can be refined to improve the accuracy of key point positioning. For example, the probability map can be smoothed using a Gaussian filter to reduce the impact of noise and discontinuity. Non-Maximum Suppression (NMS) technology can also be used to ensure that the response area of ​​each key point has a unique peak, thereby improving the accuracy of key point positioning.

[0133] Step S50, determining the design dimensions of the escalator drawing according to the position information and key point types of the key points.

[0134] It should be noted that the design dimensions of the escalator drawings are calculated and set based on the location information and key point types of key points. The location information of key points provides the precise location of each component of the escalator, while the key point type defines the component types corresponding to these positions, such as upper and lower working points, overlap points, opening points, etc. With this information, key design parameters such as the lifting height, inclination angle, and opening size of the escalator can be calculated.

[0135] It can be understood that the design dimensions such as the lifting height and inclination angle of the escalator can be calculated based on the position information and type information of the key points. For example, the lifting height of the escalator is determined by the vertical distance between the upper and lower working points, while the inclination angle is determined by calculating the ratio of the height difference between the upper and lower working points and the horizontal distance.

[0136] The present embodiment provides a method for extracting design dimensions of an escalator drawing. The present embodiment first obtains an escalator drawing and generates an image to be detected based on the escalator drawing; constructs a key point detection data set, wherein key point samples in the key point detection data set are composed of the real position, visible situation and key point type of the key points; performs model training through the key point detection data set to obtain a key point detection model; performs key point detection on the image to be detected through the key point detection model to obtain the position information of the key points; determines the design dimensions of the escalator drawing based on the position information and key point type of the key points, thereby realizing key point detection of the escalator drawing, and the key point detection efficiency and accuracy are high, thereby effectively improving the efficiency and accuracy of extracting design dimensions of the escalator drawing.

[0137] In summary, this embodiment uses a key point detection model to perform key point detection on the image to be detected generated based on the escalator drawing, obtains the position information of the key points, and then determines the design size of the escalator drawing, overcomes the technical defect that the existing key point detection algorithm cannot be applied to the key point detection task of the escalator drawing, resulting in low efficiency and accuracy in the extraction of the design size of the escalator drawing, and can realize the key point detection of the escalator drawing, with high key point detection efficiency and accuracy, thereby effectively improving the efficiency and accuracy of the extraction of the design size of the escalator drawing.

[0138] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can refer to the above introduction, and will not be repeated later. Figure 2 , the step S30 further includes steps S301-S305:

[0139] Step S301, constructing an initial detection model based on a human posture estimation model.

[0140] It should be noted that the human posture estimation model, namely the HigherHRNet network structure model, is used for scale-aware representation learning of bottom-up human posture estimation. It uses a high-resolution network (HRNet) as the backbone network and combines deconvolution operations to generate high-resolution feature maps for probability map prediction, achieving more accurate key point positioning, especially when dealing with small individuals. Figure 3 As shown, Figure 3This is a schematic diagram of the structure of the human pose estimation model HigherHRNet. The left side of the dotted line in the figure represents the high-resolution network (HRNet), which is used for feature extraction. The right side of the dotted line is the improved part of the human pose estimation model HigherHRNet, that is, by adding deconvolution operations to improve the resolution of the feature map generated by HRNet, thereby improving the detection accuracy of small target key points. The HigherHRNet loss function adopts a multi-resolution supervision strategy to calculate the mean square error between the predicted heat map of each scale and its related GT heat map. The final loss of the heat map is the sum of the mean square errors of all resolutions. And add a labeling loss to achieve the grouping of key points.

[0141] It can be understood that the initial detection model, namely the DrawingHRNet network structure model, is obtained by adding residual connections to the HigherHRNet network structure model. Figure 4 As shown, Figure 4 This is a structural diagram of the DrawingHRNet network structure model. The backbone network adopts HRNet, and draws on the idea of ​​HigherHRNet. The 1 / 4 probability map obtained by the backbone network is deconvolved to generate a 1 / 2 probability map. At the same time, a residual connection is added to add the 1 / 2 feature map initially generated by the network to the 1 / 2 probability map generated by deconvolution, so that the shallow features with clear line information and the deep features containing global information are integrated, so that the network can obtain the category and approximate position of the key points based on the global information, and the network can obtain the precise position of the key points based on the local information. Subsequently, on the basis of the 1 / 2 probability map, a deconvolution operation is performed again to generate a 1 / 1 probability map, and it is added to the 1 / 1 feature map initially generated by the network through a residual connection to obtain the final probability map.

[0142] It is worth mentioning that the escalator drawings are mainly composed of lines, and the precise positions of the key points are determined by the local lines, which account for a small proportion of the escalator drawings. The size of the heatmap generated by the conventional key point detection algorithm is often 4 times smaller than the original input image size, resulting in the loss of local line features, which makes the detected key point positions have large deviations. Therefore, the idea of ​​generating high-resolution feature maps by HigherHRNet is borrowed to improve the accuracy of key point position detection. However, the high-resolution feature maps of HigherHRNet are generated by deconvolution operations. Although this operation obtains a high-resolution feature map, the features of the line elements in the feature map are blurred, resulting in errors in the detected key point positions. Therefore, in this embodiment, residual connections are added to the HigherHRNet network structure model to obtain the DrawingHRNet network structure model, that is, the initial detection model.

[0143] Step S302, dividing the key point detection data set according to a preset ratio to obtain a training set and a test set.

[0144] It should be noted that the preset ratio can be adjusted according to actual needs, for example, the training set accounts for 80% of the data set and the test set accounts for 20%. Such a division can ensure that the model has enough data for training, while retaining a portion of the data for verifying the generalization ability of the model.

[0145] Step S303: training the initial key point detection model using the training set to obtain a trained initial key point detection model.

[0146] It should be noted that during the training process, a variety of optimization strategies can be used to improve the convergence speed and performance of the model. For example, the stochastic gradient descent (SGD) optimizer can be used in combination with the momentum mechanism to accelerate the training process. At the same time, the learning rate decay strategy can be adopted to gradually reduce the learning rate as the training progresses to avoid overfitting and enable the model to adjust parameters more finely in the later stages of training. In addition, weight decay (L2 regularization) can also be introduced to prevent the model weight from being too large, thereby improving the generalization ability of the model.

[0147] Step S304, testing the trained initial key point detection model using the test set to obtain an average accuracy.

[0148] It should be noted that in the evaluation stage, multiple indicators can be used to measure the performance of the model, including accuracy, precision, recall and F1 score, etc., which are not specifically limited in this embodiment. Accuracy reflects the proportion of key points correctly predicted by the model, while precision and recall evaluate the performance of the model from the two aspects of accuracy and completeness of the prediction results respectively. The F1 score is the harmonic mean of precision and recall, which can comprehensively reflect the prediction performance of the model. Through the comprehensive evaluation of these indicators, the performance of the model in the key point detection task can be fully understood.

[0149] It can be understood that in this embodiment, average precision is used as the evaluation indicator. Average Precision (AP) is a commonly used indicator for measuring algorithm performance. It comprehensively considers precision and recall, and can perform an overall evaluation of the algorithm prediction results.

[0150] In a feasible implementation, step S304 may include steps D11 to D13:

[0151] Step D11, inputting the key point samples in the test set into the trained initial key point detection model to obtain the predicted positions of the key points.

[0152] It should be noted that the input of key point samples is preprocessed to ensure consistency with the data format used in the training phase. The preprocessing steps may include image normalization, resizing, and enhancement to improve the adaptability and robustness of the model to different input conditions.

[0153] It can be understood that the image of the key point sample is input into the DrawingHRNet network structure model, and the predicted position is processed by multiple convolutional layers and deconvolution layers within the network, combined with the optimization of the residual connection, and finally outputs the local maximum points on the high-resolution probability map. These local maximum points represent the predicted positions of the key points.

[0154] Step D12, determining key point similarity according to the predicted position, the actual position, the key point type and the visibility condition.

[0155] It should be noted that Object Keypoint Similarity (OKS) is similar to the Intersection over Union (IoU) in object detection, which measures the similarity between the predicted keypoint positions and the true keypoint positions. OKS calculates the keypoint prediction error and weights it according to the scale of the target and the visibility of the keypoint.

[0156] It can be understood that the predicted position is compared with the true position, and the prediction error is calculated by using the Euclidean distance to measure the deviation between the predicted position and the true position. By calculating the Euclidean distance of each key point, the performance of the model on a single key point can be evaluated, and further weighted by key point type and visibility to determine key point similarity.

[0157] In a feasible implementation, step D12 may include: determining the Euclidean distance, normalization factor and scale factor corresponding to each type of key point based on the predicted position, the actual position, the key point type and the visibility situation; determining the key point similarity based on the Euclidean distance, the normalization factor and the scale factor.

[0158] It should be noted that the Euclidean distance is the direct distance between the predicted position and the true position, which represents the prediction error; the normalization factor represents the standard deviation between the manually labeled value and the true value, which is used to adjust the similarity measure between objects of different scales; the scale factor takes into account the impact of the target size on the similarity of key points.

[0159] It can be understood that in the DrawingHRNet network structure model, a multi-resolution supervision strategy is adopted to calculate the mean square error between the probability maps of all generated resolutions and the corresponding true annotation (GT) probability maps. Adding group loss, that is, the labels of all key points of a single escalator should be consistent, thereby introducing global supervision for key point detection. For example, if the escalator span is large, the number of hooks and support points should be increased accordingly. In addition, since the position of each escalator has been detected by the target detection algorithm before the key point detection algorithm, the number of escalators input to the key point detection algorithm is single. Since the number of key points of each category in the escalator is different and there is more than one, the group loss should make all key points of a single escalator belong to the same group, that is, the labels of all key points should be consistent; instead of making each key point of each category of detected key points belong to a different group like previous methods. The group loss is as follows:

[0160]

[0161] T={0},k=1,...,,K

[0162]

[0163] Among them, h k is the predicted tag of the kth keypoint, and h(x) is the tag value at pixel position x ， T represents the true position of all key points of the escalator, x k represents the true position of the kth keypoint, and h represents the labeling mean of all keypoints of a single escalator. The final loss function uses the sum of the mean square error of the resolution probability map and the labeling loss.

[0164] The existing key point detection uses AP value as the evaluation index. AP is determined by the target key point similarity (OKS). The OKS calculation formula for a single target is as follows:

[0165]

[0166] Among them, OKS is the key point similarity; k represents the key point category; d k Represents the detection value x of the k-th key point k and the true value (Groundtruth) x k 'Euclidean distance; S represents the scale factor of the current target, w and h represent the width and height of the current target, which can be calculated from the true positions of all key points of the target; σk represents the normalization factor of the kth key point, which represents the standard deviation between the manual annotation value and the true value; v kRepresents the visibility of the k-th key point in the groundtruth, v k = 0 means the key point does not exist in the image, v k =1 means that the key point exists in the image; δ(*)=1 means that if the * condition is met, then δ(*)=1, otherwise it is 0, that is, only the OKS of the key points existing in the image are calculated.

[0167] In the above formula, d k It represents the Euclidean distance between the detection value of the k-th key point and the true value (Groundtruth). In conventional key point detection tasks, each target has only one key point of each type, while the number of key points of each type for an amorphous target such as an escalator is different and there is more than one key point. Therefore, it is impossible to know the one-to-one correspondence between the detection value of each type of key point and the true value, and it is impossible to use the above formula to calculate the Euclidean distance between the detection value of each type of key point and the Groundtruth. Therefore, in this embodiment, a key point similarity OKS calculation method for amorphous targets is proposed, as follows:

[0168]

[0169] Among them, d k For the k-th key point, assuming that the number of Groundtruth key points is M, the Groundtruth key points are represented by X, the number of detected key points is N, and the detected key points are represented by X', then is an M×N matrix, each row of which is the Euclidean distance between a key point in Groundtruth and the N key points detected. max(*[m,:]) means that the maximum value of the mth row in the matrix is ​​taken as the Euclidean distance between the true value and the detected value of the mth key point in the kth class of key points. Since the manual annotation results are corrected with the points in the drawing in the dataset production method and label format stage, σ k Take 0.001.

[0170] Step D13, determining the average precision according to the key point similarity.

[0171] It should be noted that the average precision (AP) is obtained by calculating the area under the precision and recall curves at different thresholds. The higher the AP value, the better the performance of the model in the key point detection task.

[0172] It is understandable that in the key point detection task, the average precision (AP) is an important evaluation indicator, which comprehensively considers the accuracy and recall rate of detection. In order to calculate AP, it is first necessary to determine the key point similarity (OKS) of each target, and then determine the true positive, false positive and false negative of the detection result according to the OKS threshold. For each target, if its key point similarity is greater than a certain threshold (for example, 0.5), the target is considered to be correctly detected; otherwise, the detection is considered to have failed. By changing the threshold, the precision-recall curve under different thresholds can be drawn, and the area under the curve can be calculated, which is the AP value.

[0173] Step S305, when the average accuracy reaches a preset accuracy threshold, the trained initial key point detection model is used as a target key point detection model.

[0174] It should be noted that when the average accuracy reaches the preset accuracy threshold, it indicates that the current model performance has met the requirements, and the initial key point detection model in the current training phase can be used as the target key point detection model. When the average accuracy does not reach the preset accuracy threshold, it is necessary to continue to optimize the model and adjust the model's hyperparameters, such as learning rate, batch size, and optimizer type, to find a parameter combination that is more suitable for the current data set, thereby improving the accuracy of the model.

[0175] In this embodiment, an initial detection model is constructed based on the human posture estimation model, and then the initial detection model is trained and tested until the average accuracy reaches a preset accuracy threshold, thereby obtaining a key point detection model, which effectively improves the model accuracy and generalization ability.

[0176] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the method for extracting design dimensions from escalator drawings of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0177] This application also provides an escalator drawing design size extraction device, please refer to Figure 5 , the escalator drawing design dimension extraction device comprises:

[0178] The acquisition module 10 is used to acquire the escalator drawing and generate the image to be detected according to the escalator drawing.

[0179] The construction module 20 is used to construct a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key point.

[0180] The training module 30 is used to perform model training using the key point detection data set to obtain a key point detection model.

[0181] The detection module 40 is used to perform key point detection on the image to be detected by using the key point detection model to obtain the position information of the key points.

[0182] The determination module 50 is used to determine the design size of the escalator drawing according to the position information and key point type of the key point.

[0183] The present embodiment provides a device for extracting design dimensions of an escalator drawing. The present embodiment first obtains an escalator drawing and generates an image to be detected based on the escalator drawing; constructs a key point detection data set, wherein key point samples in the key point detection data set are composed of the real position, visible situation and key point type of the key points; performs model training through the key point detection data set to obtain a key point detection model; performs key point detection on the image to be detected through the key point detection model to obtain the position information of the key points; determines the design dimensions of the escalator drawing based on the position information and key point type of the key points, thereby realizing key point detection of the escalator drawing, and the key point detection efficiency and accuracy are high, thereby effectively improving the efficiency and accuracy of design dimension extraction of the escalator drawing.

[0184] In summary, this embodiment uses a key point detection model to perform key point detection on the image to be detected generated based on the escalator drawing, obtains the position information of the key points, and then determines the design size of the escalator drawing, overcomes the technical defect that the existing key point detection algorithm cannot be applied to the key point detection task of the escalator drawing, resulting in low efficiency and accuracy in the extraction of the design size of the escalator drawing, and can realize the key point detection of the escalator drawing, with high key point detection efficiency and accuracy, thereby effectively improving the efficiency and accuracy of the extraction of the design size of the escalator drawing.

[0185] Optionally, the acquisition module 10 is also used to acquire an escalator drawing, and convert the format of the escalator drawing to obtain an escalator image in a target format; detect the position and type of the escalator in the escalator image through a target detection algorithm; determine a cropping area according to the position and type of the escalator; crop the escalator image according to the cropping area to obtain a cropped escalator image; scale the cropped escalator image to a preset pixel value to obtain an image to be detected.

[0186] Optionally, the construction module 20 is also used to annotate the escalator drawing through a picture annotation tool to obtain annotated key points; obtain the data coordinates, visibility and key point types of the annotated key points, and extract the data coordinates of all feature points in the escalator drawing; determine the true position of the annotated key points according to the data coordinates of the annotated key points and the data coordinates of all feature points; use the true position, visibility and key point type of the annotated key points as key point samples, and construct a key point detection data set based on the key point samples.

[0187] Optionally, the training module 30 is also used to construct an initial detection model based on a human posture estimation model; divide the key point detection data set according to a preset ratio to obtain a training set and a test set; train the initial key point detection model with the training set to obtain a trained initial key point detection model; test the trained initial key point detection model with the test set to obtain an average accuracy; when the average accuracy reaches a preset accuracy threshold, use the trained initial key point detection model as the target key point detection model.

[0188] Optionally, the training module 30 is also used to input the key point samples in the test set into the trained initial key point detection model to obtain the predicted position of the key point; determine the key point similarity based on the predicted position, the actual position, the key point type and the visibility; and determine the average accuracy based on the key point similarity.

[0189] Optionally, the training module 30 is also used to determine the Euclidean distance, normalization factor and scale factor corresponding to each type of key point based on the predicted position, the actual position, the key point type and the visibility situation; and determine the key point similarity based on the Euclidean distance, the normalization factor and the scale factor.

[0190] Optionally, the detection module 40 is also used to perform feature extraction on the image to be detected to obtain a first feature map, a second feature map and a third feature map; input the first feature map into the backbone network of the key point detection model to obtain a first probability map; perform a deconvolution operation on the first probability map to obtain a second probability map; fuse the second feature map and the second probability map through the residual connection structure of the key point detection model to obtain a fused second probability map; perform a deconvolution operation on the fused second probability map to obtain a third probability map; fuse the third feature map with the third probability map to obtain a target probability map; and determine the location information of the key point according to the target probability map.

[0191] The automatic escalator drawing design dimension extraction device provided by the present application adopts the automatic escalator drawing design dimension extraction method in the above-mentioned embodiment, which can solve the technical problem of how to realize the key point detection of the automatic escalator drawing, thereby effectively improving the efficiency and accuracy of the automatic escalator drawing design dimension extraction. Compared with the prior art, the beneficial effects of the automatic escalator drawing design dimension extraction device provided by the present application are the same as the beneficial effects of the automatic escalator drawing design dimension extraction method provided by the above-mentioned embodiment, and the other technical features in the automatic escalator drawing design dimension extraction device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0192] The present application provides an escalator drawing design dimension extraction device, which includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the escalator drawing design dimension extraction method in the above-mentioned embodiment 1.

[0193] Reference below Figure 6 , which shows a schematic diagram of the structure of an escalator drawing design dimension extraction device suitable for implementing the embodiment of the present application. The escalator drawing design dimension extraction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The escalator drawing design dimension extraction device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0194] like Figure 6As shown, the automatic escalator drawing design size extraction device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the automatic escalator drawing design size extraction device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the automatic escalator drawing design dimension extraction device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows an automatic escalator drawing design dimension extraction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0195] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0196] The automatic escalator drawing design dimension extraction device provided by the present application adopts the automatic escalator drawing design dimension extraction method in the above-mentioned embodiment, which can solve the technical problem of how to realize the key point detection of the automatic escalator drawing, thereby effectively improving the efficiency and accuracy of the automatic escalator drawing design dimension extraction. Compared with the prior art, the beneficial effects of the automatic escalator drawing design dimension extraction device provided by the present application are the same as the beneficial effects of the automatic escalator drawing design dimension extraction method provided by the above-mentioned embodiment, and the other technical features in the automatic escalator drawing design dimension extraction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0197] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0198] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0199] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the method for extracting design dimensions of escalator drawings in the above-mentioned embodiment.

[0200] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0201] The computer-readable storage medium may be included in the escalator drawing design dimension extraction device; or may exist independently without being assembled into the escalator drawing design dimension extraction device.

[0202] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the escalator drawing design dimension extraction device, the escalator drawing design dimension extraction device: obtains the escalator drawing, and generates an image to be detected based on the escalator drawing; constructs a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key points; performs model training through the key point detection data set to obtain a key point detection model; performs key point detection on the image to be detected through the key point detection model to obtain the position information of the key points; and determines the design size of the escalator drawing according to the position information and key point type of the key points.

[0203] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0204] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0205] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0206] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for extracting the design dimensions of escalator drawings, and can solve the technical problem of how to realize key point detection of escalator drawings, thereby effectively improving the efficiency and accuracy of escalator drawing design dimension extraction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the method for extracting the design dimensions of escalator drawings provided by the above-mentioned embodiments, and will not be elaborated here.

[0207] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned method for extracting design dimensions of escalator drawings when executed by a processor.

[0208] The computer program product provided by the present application can solve the technical problem of how to realize key point detection of escalator drawings, thereby effectively improving the efficiency and accuracy of escalator drawing design dimension extraction. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the escalator drawing design dimension extraction method provided by the above embodiment, which will not be repeated here.

[0209] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for extracting design dimensions from escalator drawings, characterized in that: The method comprises: Obtaining an escalator drawing, and generating an image to be detected according to the escalator drawing; Constructing a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key point; Performing model training using the key point detection data set to obtain a key point detection model; Perform key point detection on the image to be detected by using the key point detection model to obtain position information of the key points; The design dimensions of the escalator drawing are determined according to the position information of the key points and the types of the key points.

2. The method according to claim 1, characterized in that The step of obtaining the escalator drawing and generating the image to be detected according to the escalator drawing includes: Acquire an escalator drawing, and convert the format of the escalator drawing to obtain an escalator image in a target format; Detecting the location and type of the escalator in the escalator image by a target detection algorithm; determining a trimming area according to the location and type of the escalator; Clipping the escalator image according to the clipping area to obtain a clipped escalator image; The cropped escalator image is scaled to a preset pixel value to obtain an image to be detected.

3. The method according to claim 1, characterized in that The step of constructing a key point detection data set includes: Annotate the escalator drawing using a picture annotation tool to obtain annotated key points; Obtaining the data coordinates, visibility and key point types of the marked key points, and extracting the data coordinates of all feature points in the escalator drawing; Determine the real position of the marked key point according to the data coordinates of the marked key point and the data coordinates of all the feature points; The actual position, visibility and key point type of the marked key point are taken as key point samples, and a key point detection dataset is constructed based on the key point samples.

4. The method according to claim 1, characterized in that The step of performing model training using the key point detection data set to obtain a key point detection model includes: Build an initial detection model based on the human posture estimation model; Dividing the key point detection data set according to a preset ratio to obtain a training set and a test set; Training the initial key point detection model using the training set to obtain a trained initial key point detection model; Testing the trained initial key point detection model using the test set to obtain an average accuracy; When the average accuracy reaches a preset accuracy threshold, the trained initial key point detection model is used as a target key point detection model.

5. The method according to claim 4, characterized in that The testing of the trained initial key point detection model using the test set to obtain an average accuracy includes: Inputting the key point samples in the test set into the trained initial key point detection model to obtain the predicted positions of the key points; Determining key point similarity according to the predicted position, the actual position, the key point type, and the visibility condition; An average precision is determined based on the keypoint similarities.

6. The method according to claim 5, characterized in that The determining the key point similarity according to the predicted position, the actual position, the key point type and the visibility condition comprises: Determine the Euclidean distance, normalization factor and scale factor corresponding to each type of key point according to the predicted position, the actual position, the key point type and the visibility condition; Key point similarity is determined based on the Euclidean distance, the normalization factor, and the scale factor.

7. The method according to claim 1, characterized in that The step of performing key point detection on the image to be detected by using the key point detection model to obtain position information of the key points includes: Extracting features of the image to be detected to obtain a first feature map, a second feature map, and a third feature map; Inputting the first feature map into the backbone network of the key point detection model to obtain a first probability map; Performing a deconvolution operation on the first probability map to obtain a second probability map; Fusing the second feature map and the second probability map through the residual connection structure of the key point detection model to obtain a fused second probability map; Performing a deconvolution operation on the fused second probability map to obtain a third probability map; Fusing the third feature map with the third probability map to obtain a target probability map; The position information of the key points is determined according to the target probability map.

8. An escalator design dimension extraction device, characterized in that: The escalator drawing design dimension extraction device comprises: An acquisition module, used for acquiring an escalator drawing and generating an image to be detected according to the escalator drawing; A construction module is used to construct a key point detection data set, wherein the key point samples in the key point detection data set are composed of the real position, visibility and key point type of the key point; A training module, used to perform model training using the key point detection data set to obtain a key point detection model; A detection module, used to perform key point detection on the image to be detected by using the key point detection model to obtain position information and type information of the key points; The determination module is used to extract the design dimensions of the escalator drawing according to the position information and key point types of the key points.

9. An escalator design dimension extraction device, characterized in that: The escalator drawing design dimension extraction device includes: a memory, a processor, and an escalator drawing design dimension extraction program stored in the memory and executable on the processor, wherein the escalator drawing design dimension extraction program is configured to implement the escalator drawing design dimension extraction method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores an escalator drawing design dimension extraction program, and when the escalator drawing design dimension extraction program is executed by the processor, the escalator drawing design dimension extraction method according to any one of claims 1 to 7 is implemented.

Images