Road sign automatic statistical method based on image recognition

Abstract

The invention discloses an automatic road sign statistical method based on image recognition. The method comprises the following steps: acquiring an electronic drawing of a traffic engineering graphic design drawing of to-be-counted data; inputting the electronic drawing of the traffic engineering graphic design drawing into the road sign target detection model to obtain sign features; inputting the electronic drawing of the traffic engineering graphic design drawing into the road segmentation model to obtain road features; acquiring a mark name according to the mark feature data type; inputting the sign feature data coordinates and the sign feature image into a road sign character recognition model to obtain a road stake number and a support form; and inputting the sign feature data coordinates, the road feature image, the road feature data coordinates and the road stake number data coordinates into a road sign position discrimination model to obtain a set position, thereby completing automatic statistics of all required information of the road sign. According to the statistical method, omissions possibly occurring in the manual operation process are effectively avoided, so that the working efficiency and the design accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and image recognition, and in particular to an automatic road sign counting method based on image recognition. Background Art

[0002] With the high-quality development of my country's economy and society, the living standards of residents have been significantly improved, and the number of motor vehicles has increased rapidly, which has put forward higher requirements for urban transportation systems. In order to cope with the challenges brought by the increase in traffic volume, the refined design of traffic engineering is particularly important. How to effectively improve the design efficiency and drawing quality in the road construction drawing design stage has become an urgent problem to be solved.

[0003] After the design of the traffic engineering is completed, it is necessary to count the relevant information in the plane design drawing of the traffic engineering, including the sign pile number, sign name, setting location, support form, etc. At present, this process mainly adopts the traditional manual counting method. Designers need to spend a lot of time to check and fill in the information of the sign item by item when drawing. This is not only inefficient, but also easy to cause omissions due to human factors, and it is difficult to meet the current demand for efficient and high-quality design.

[0004] The existing technology lacks practical tools and effective methods for automated processing in this work, and it is difficult to cope with the growing workload of road design. Therefore, how to develop an efficient and accurate road sign counting method to replace the traditional manual counting method and improve design efficiency and drawing quality has become one of the key issues to be solved in road construction drawing design technology. This has important practical significance for improving road design quality, shortening design cycles, and better serving the construction of urban transportation systems. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned background technology and provide an automatic road sign counting method based on image recognition. The method is efficient and accurate and can replace the traditional manual counting method to improve the design efficiency and accuracy.

[0006] To achieve the above object, the present invention provides an automatic road sign counting method based on image recognition, comprising the following steps:

[0007] Obtain electronic drawings of the plane design of the traffic engineering project for which statistics are to be collected;

[0008] Inputting the electronic drawing of the traffic engineering plan design into the trained road sign target detection model to obtain the sign features; the sign features include the sign feature image, the sign feature data type and the sign feature data coordinates;

[0009] Inputting the electronic drawing of the traffic engineering plane design into the trained road segmentation model to obtain road features; the road features include road feature images, road feature data coordinates and road pile number data coordinates;

[0010] Acquire the sign name according to the sign feature data type; input the sign feature data coordinates and the sign feature image into a road sign text recognition model to obtain the road stake number and support form;

[0011] The sign feature data coordinates, road feature image, road feature data coordinates and road stake number data coordinates are input into a road sign position discrimination model to obtain a setting position, thereby completing automatic statistics of all required information of the road sign.

[0012] As a preferred implementation mode, the traffic engineering plan design drawing to be statistically counted includes design information of road routes and road signs.

[0013] As a preferred implementation, the traffic engineering plan design drawing for which the statistical data is to be collected is in the format of an image.

[0014] As a preferred implementation method, the method for obtaining the traffic engineering plan design drawings to be statistically counted is to clean the traffic engineering plan design drawings and delete irrelevant graphic and text data.

[0015] As a preferred implementation, the road sign target detection model is trained and established in the following manner:

[0016] The road sign target detection model is obtained by training the target detection model using the road sign dataset;

[0017] The road sign dataset is a target detection dataset, and each data is marked with four vertices;

[0018] The annotation range of the road sign data set includes image information of the sign name, text information of the pile number and text information of the supporting form;

[0019] The road sign dataset is annotated using "Road Traffic Signs and Markings Part 2: Road Traffic Signs" (GB5768.2-2022) and traffic engineering plan drawings of multiple engineering projects, including: 48 types of prohibition signs, 43 types of instruction signs, 47 types of warning signs, 77 types of direction signs, 17 types of tourist signs, 10 types of notice signs, and 20 types of auxiliary signs.

[0020] As a preferred implementation, the road segmentation model is trained and established in the following manner:

[0021] The road segmentation model is obtained by training the semantic segmentation model using the road route segmentation dataset;

[0022] The road route segmentation dataset is a semantic segmentation dataset;

[0023] The annotation range of the road route segmentation data set is the center line and the road stake number.

[0024] As a preferred implementation, the road sign character recognition model is trained and established in the following manner:

[0025] The road sign text recognition model consists of three parts: a text recognition model, a similarity text matching algorithm, and a road sign text library;

[0026] After the text recognition model recognizes the sign name and supporting form text, a similarity text matching algorithm is used to match the closest text in the road sign text library;

[0027] The road sign text library includes two types of text: sign name and supporting form.

[0028] As a preferred implementation, the road sign position discrimination model is a vector cross product calculation model.

[0029] As a preferred implementation, the road sign position discrimination model is trained and established in the following manner:

[0030] The road sign position discrimination model first samples four directed line segments close to the position to be judged from the road feature data;

[0031] The positive direction of the directed line segment is determined according to the road stake number, the starting point is a point near a smaller road stake number, the end point is a point near a larger road stake number, and the positive direction is from the starting point to the end point;

[0032] Then, the vector cross products of the four vertices and the four directed line segments of the road sign at the position to be determined are calculated, and a total of 16 cross product results are obtained;

[0033] If the cross product result of more than half of the vertices is greater than 0, the road sign is on the left side of the road, otherwise it is on the right side of the road.

[0034] As a preferred implementation, the road pile number digits are identified from the road feature image and the road pile number data using a text recognition model.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] First, the method of the present invention is a road sign counting method used in the road construction drawing design stage, which can automatically count the relevant information of road signs, including sign pile number, sign name, setting location, support form, etc. This method greatly reduces the time required for designers to conduct sign counting after traffic engineering design, replaces the traditional manual counting method, and effectively avoids possible omissions in the manual operation process, thereby improving work efficiency and design accuracy. Through the application of the present invention, designers can devote more energy to scheme optimization and improvement of design quality, significantly shorten the design cycle, and meet the needs of efficient and high-quality design.

[0037] Secondly, the present invention provides an efficient and automated road sign counting method, which fills the gap in the existing technology in road construction drawing design, and plays an important role in improving the quality of urban road design, optimizing the construction drawing design process, and promoting the construction of urban transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of an automated road sign counting method based on image recognition is provided;

[0039] Figure 2 It is a schematic diagram of the landmark feature extraction method;

[0040] Figure 3 Schematic diagram of the road feature extraction method. DETAILED DESCRIPTION

[0041] The following is a detailed description of the implementation of the present invention in conjunction with the implementation cases, but they do not constitute a limitation of the present invention and are only given as examples. At the same time, the advantages of the present invention will become clearer and easier to understand through the description.

[0042] like Figure 1 As shown, after the traffic engineering design, it is necessary to count the information such as the sign pile number, sign name, setting position, support form, etc. in the plane design drawing of the traffic engineering. At present, the traditional manual counting method is adopted, and it takes a lot of time to complete the sign counting work when the designer draws the drawing, which is inefficient. Therefore, the present invention patent develops an efficient and accurate sign automatic counting method based on full research and combined with actual engineering, so as to replace the traditional manual counting method and improve the design efficiency and accuracy.

[0043] An automatic road sign counting method based on image recognition of the present invention comprises the following steps:

[0044] S1: Obtain the plane design drawing of the traffic engineering project to be counted;

[0045] Among them, the schematic diagram of the traffic engineering plan design for which statistics are to be collected is as follows: Figure 2, Figure 3 The traffic engineering plan design drawings to be counted are in the form of images. The method for obtaining the traffic engineering plan design drawings to be counted is to clean the traffic engineering plan design drawings, delete some irrelevant graphic data information, and only retain key design information such as road routes, road signs (including corresponding columns), etc.

[0046] S2: Input the traffic engineering plan design drawing into the trained road sign target detection model to obtain the sign features. Input the traffic engineering plan design drawing into the trained road segmentation model to obtain the road features;

[0047] Among them, Figure 2 As shown, the marker features include a marker feature image, a marker feature data type, and a marker feature data coordinate. Figure 3 As shown, the road feature includes a road feature image and road feature data coordinates.

[0048] The trained road sign target detection model in this embodiment is the YoloV8 model (You Only LookOnce version 8); the road sign dataset is annotated using "Road Traffic Signs and Markings Part 2: Road Traffic Signs" (GB5768.2-2022) and traffic engineering plan drawings of multiple engineering projects. The road sign dataset is a target detection dataset, and each data is annotated with four vertices; the road sign dataset categories include: 48 types of prohibition signs, 43 types of instruction signs, 47 types of warning signs, 77 types of guide signs, 17 types of tourist signs, 10 types of notice signs, 20 types of auxiliary signs, etc.; before training, the annotated drawings are cropped to meet the size input of the model. The drawings are cropped into 512*512 pixel blocks.

[0049] In this embodiment, the trained road segmentation model adopts the Unet model; the data set is annotated with the traffic engineering plan design drawings of multiple engineering projects. Before training, the annotated drawings are cropped to meet the size input of the model. The drawings are cropped into 512*512 pixel blocks.

[0050] In this embodiment, YoloV8 is used as the target detection model and Unet is used as the semantic segmentation model, but other models and data may be used in other implementation schemes.

[0051] S3: Obtaining a sign name according to the sign feature data type; inputting the sign feature data coordinates and the sign feature image into a road sign text recognition model to obtain a road stake number and a support form;

[0052] In this embodiment, the sign text recognition model consists of three parts: a text recognition model, a similarity text matching algorithm, and a road sign text library.

[0053] The text recognition model in this embodiment is the easyOCR model. The similarity text matching algorithm in this embodiment adopts the vector output by bert-base-chinese and the cosine similarity algorithm. The road sign text library in this embodiment is established based on the text information statistically collected in the sign list of multiple engineering projects.

[0054] In this embodiment, the text recognition model is the easyOCR model, the similarity text matching algorithm uses the vector output by bert-base-chinese and the cosine similarity algorithm, and the road sign text library is based on the text information statistically collected in the sign list of multiple engineering projects, but other models and data can be used in other implementation plans.

[0055] S4: inputting the sign feature data coordinates, the road feature image, the road feature data coordinates and the road stake number data coordinates into a road sign position discrimination model to obtain the sign position, thereby completing automatic statistics of all required information of the road sign;

[0056] The principle of the road sign position discrimination model in this embodiment is as follows:

[0057] First, define the vertices of the rectangle. Assume that the four vertices of the rectangle are P1, P2, P3, and P4. Then, sample four line segments close to the position to be judged from the road feature data, and define the two endpoints of each line segment as A1 and B1, A2 and B2, A3 and B3, and A4 and B4. Take a certain line segment as an example, and assume that the two endpoints of the line segment are A(ax,ay) and B(bx,by). Then, calculate the direction of each rectangle vertex relative to the line segment: for each rectangle vertex P(px,py), calculate the cross product of vectors AB and AP. The cross product formula is: AB xAP=(bx-ax)*(py-ay)-(by-ay)*(px-ax). Then, determine the result of the cross product. If the cross product results of more than 8 vertices in all 16 vertices are greater than 0, it means that all vertices of the rectangle are on the left side of the line segment. Otherwise, it means that all vertices of the rectangle are on the right side of the line segment.

[0058] The text recognition model in this embodiment is the easyOCR model. Note that the text recognition model in this embodiment is the easyOCR model, but other models and data may be used in other implementation schemes.

[0059] Typical examples of automatically collected information are shown in Table 1:

[0060] Table 1 Typical cases of automatically counted information

[0061]

[0062] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the field without departing from the purpose of the present invention, which should also be regarded as the protection scope of the present invention.

[0063] The above are only specific embodiments of the present invention. It should be pointed out that any changes or substitutions that can be easily thought of by any technician familiar with the field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention, and the rest not described in detail belong to the prior art.

Claims

1. An automatic road sign counting method based on image recognition, characterized in that: The steps include: Obtain electronic drawings of the plane design of the traffic engineering project for which statistics are to be collected; Inputting the electronic drawing of the traffic engineering plan design into the trained road sign target detection model to obtain the sign features; the sign features include the sign feature image, the sign feature data type and the sign feature data coordinates; Inputting the electronic drawing of the traffic engineering plane design into the trained road segmentation model to obtain road features; the road features include road feature images, road feature data coordinates and road pile number data coordinates; Acquire the sign name according to the sign feature data type; input the sign feature data coordinates and the sign feature image into a road sign text recognition model to obtain the road stake number and support form; The sign feature data coordinates, road feature image, road feature data coordinates and road stake number data coordinates are input into a road sign position discrimination model to obtain a setting position, thereby completing automatic statistics of all required information of the road sign.

2. The method for automatic road sign counting based on image recognition according to claim 1, characterized in that: The traffic engineering plan design drawing to be statistically collected includes design information of road routes and road signs.

3. The method for automatic road sign counting based on image recognition according to claim 2, characterized in that: The traffic engineering plan drawings to be statistically counted are in the format of images.

4. The method for automatic road sign counting based on image recognition according to claim 3, characterized in that: The method for obtaining the traffic engineering plan design drawings for which statistical data is to be collected is to perform data cleaning on the traffic engineering plan design drawings.

5. The method for automatic road sign counting based on image recognition according to any one of claims 1 to 4, characterized in that: The road sign target detection model is trained and established in the following way: The road sign target detection model is obtained by training the target detection model using the road sign dataset; The road sign dataset is a target detection dataset, and each data is marked with four vertices; The annotation scope of the road sign data set includes image information of the sign name, text information of the pile number and text information of the supporting form.

6. The method for automatic road sign counting based on image recognition according to claim 5, characterized in that: The road segmentation model is trained and established in the following way: The road segmentation model is obtained by training the semantic segmentation model using the road route segmentation dataset; The road route segmentation dataset is a semantic segmentation dataset; The annotation range of the road route segmentation data set is the center line and the road stake number.

7. The method for automatic road sign counting based on image recognition according to claim 6, characterized in that: The road sign text recognition model is trained and established in the following manner: The road sign text recognition model consists of three parts: a text recognition model, a similarity text matching algorithm, and a road sign text library; After the text recognition model recognizes the sign name and supporting form text, a similarity text matching algorithm is used to match the closest text in the road sign text library; The road sign text library includes two types of text: sign name and supporting form.

8. The method for automatic road sign counting based on image recognition according to claim 7, characterized in that: The road sign position discrimination model is a vector cross product calculation model.

9. The method for automatic road sign counting based on image recognition according to claim 8, characterized in that: The road sign position discrimination model is trained and established in the following manner: The road sign position discrimination model first samples four directed line segments close to the position to be judged from the road feature data; The positive direction of the directed line segment is determined according to the road stake number, the starting point is a point near a smaller road stake number, the end point is a point near a larger road stake number, and the positive direction is from the starting point to the end point; Then, the vector cross products of the four vertices and the four directed line segments of the road sign at the position to be determined are calculated; If the cross product result of more than half of the vertices is greater than 0, the road sign is on the left side of the road, otherwise it is on the right side of the road.

10. The method for automatic road sign counting based on image recognition according to claim 9, characterized in that: The road pile number digits are identified from the road feature image and the road pile number data using a text recognition model.

Images