An artificial intelligence-based image control point automatic piercing method

Abstract

The application discloses an artificial intelligence-based automatic photocontrol point pricking method, which comprises the following steps: collecting a photocontrol point image, wherein the shapes of the photocontrol points are different, and specific coordinates of the photocontrol points are set based on the shapes; pre-processing the photocontrol point image to obtain photocontrol point sample data; constructing an initial photocontrol point pricking model, training the photocontrol point pricking model based on the photocontrol point sample data, and obtaining a photocontrol point pricking model; and performing automatic photocontrol point pricking based on the photocontrol point pricking model. The application makes the photogrammetric pricking work free from the dependence on artificial experience, can not only improve the photocontrol point pricking speed, but also ensure the consistency of the pricking precision, improve the pricking precision, and thus improve the quality of surveying and mapping results.

Description

Technical Field

[0001] The present invention relates to the field of intelligent image recognition, and in particular to an automatic image control point puncture method based on artificial intelligence. Background Art

[0002] With the rapid development of drone technology, aerial photogrammetry has become increasingly mature. Oblique photogrammetry has become the main method for solving large-scale topographic map measurement, which is not only highly efficient but also can meet the accuracy requirements.

[0003] Currently, oblique photogrammetry is mainly carried out by using drones carrying five-lens cameras (or other multi-lens cameras) to shoot the ground. The data is then used to generate a real-life 3D model using Context Capture modeling software, which is then used for design, planning, and other work.

[0004] Currently, advanced drone technology makes it very easy to acquire photos, allowing us to collect large amounts of image data in a very short time. Regarding data processing, advances in computer science have led to the widespread use of technologies such as GPUs and asynchronous computing, which is a boon for photogrammetric data processing. Current computer technology is fully capable of handling the massive amounts of data collected by photogrammetry.

[0005] However, there is a process of image control points between data collection and data processing, which is to assign coordinates to specific locations in the photo. This process requires manual participation, which has become a bottleneck technology that hinders the development of aerial photogrammetry. At present, the industry mainly uses "L", "X", Image control points of uniform shape are used throughout the entire area. During internal work, the coordinates of specific shape symbols in the photograph are identified and assigned to those locations. This process requires technical expertise, and the accuracy of the points is significantly affected by the quality of the operator, which can easily lead to deviations in the image control point assignment results. Furthermore, manual processing is inefficient. Automating the image control point process would undoubtedly bring significant changes to the photogrammetry industry. Summary of the Invention

[0006] The purpose of the present invention is to provide an automatic puncture method for image control points based on artificial intelligence to solve the problems existing in the above-mentioned prior art, so that the photogrammetry puncture work can get rid of the dependence on manual experience. It can not only improve the speed of image control point puncture, but also ensure the consistency of puncture accuracy, improve puncture accuracy, and thus improve the quality of surveying and mapping results.

[0007] To achieve the above-mentioned purpose, the present invention provides the following solution: The present invention provides an automatic point-puncturing method based on artificial intelligence, comprising the following steps:

[0008] Acquiring an image of image control points, wherein the image control points have different shapes, and setting specific coordinates of the image control points based on the shapes;

[0009] Preprocessing the image control point image to obtain image control point sample data;

[0010] Constructing an initial image-controlled point prick point model, and training the image-controlled point prick point model based on the image-controlled point sample data to obtain an image-controlled point prick point model;

[0011] Based on the image control point puncture model, automatic image control point puncture is performed.

[0012] Optionally, the center position of the image control point is coordinate information.

[0013] Optionally, acquiring the image of the control point includes:

[0014] Constructing a basic image, wherein the background of the basic image is black, and setting a first image unit in a target area of ​​the basic image;

[0015] A second image unit is constructed, and according to the preset shape of the image control points, the basic image is filled with the second image unit to obtain the image control point image and store it.

[0016] Optionally, preprocessing the image of the control point includes:

[0017] The image control point images are randomly rotated, scaled, stretched, brightness adjusted, and noise added, the number of the image control point images is increased, and the image control point sample data is obtained.

[0018] Optionally, the image control point puncture point model adopts a deep learning network.

[0019] Optionally, the image-controlled point puncture model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer, and the input layer, the convolutional layer, the pooling layer, the fully connected layer and the output layer are connected in this order.

[0020] Optionally, during the training of the GCP prick point model, the loss function adopts the categorical_crossentropy function, and the optimization function adopts the Adam function.

[0021] Optionally, performing automatic puncture of image control points based on the image control point puncture model includes:

[0022] The image of the image control points to be detected is input into the image control point puncturing model, the pattern of the image control points to be detected is identified, the specific coordinates of the image control points to be detected are extracted, and the image control points are automatically punctured.

[0023] The present invention discloses the following technical effects:

[0024] The present invention provides an artificial intelligence-based automatic image control point puncturing method. By constructing a batch of image control points of different shapes, each of which carries specific coordinate information, the image control point recognition method is completely completed automatically by a computer, and no human intervention is required throughout the process. This method can not only improve the image control point puncturing speed, but also ensure the consistency of puncturing accuracy, improve puncturing accuracy, and thus improve the quality of surveying and mapping results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 Schematic diagram of the process of the automatic point puncturing method based on artificial intelligence in an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of image control point patterns in an embodiment of the present invention;

[0028] Figure 3 Schematic diagram of a method for constructing image control points in an embodiment of the present invention;

[0029] Figure 4 Schematic diagram of the result after image control point preprocessing in an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of a network structure in an embodiment of the present invention;

[0031] Figure 6 Schematic diagram of the intelligent recognition process in an embodiment of the present invention;

[0032] Figure 7 Schematic diagram of automatic point verification of image control points in an embodiment of the present invention, wherein Figure (a) is a pattern of image control points to be detected; Figure (b) is a schematic diagram of the verification result. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] The present invention provides an automatic point-puncturing method based on artificial intelligence. Figure 1 As shown, the following steps are included:

[0036] S1. Construct an image data set of image control points and set corresponding coordinate data for the image control points.

[0037] Different from the traditional industry which mainly uses image control points in the shape of "L", "X", "", etc., this embodiment uses the image control point construction method to effectively expand the shape of the image control points and construct 124 image control points of different shapes. The center position of each image control point represents the position information of the image control point, and each image control point of different shapes is assigned specific coordinate data. In addition, the 124 images are numbered from 0 to 123, which is also the file name when the image is stored. The constructed image control point style is as follows: Figure 2 shown.

[0038] The construction orientation of each image control point is as follows Figure 3 As shown, the following steps are included:

[0039] S101 . Create a background image of a threshold size and fill the background image with black. In this embodiment, the size of the background image is 160 dpi*160 dpi.

[0040] S102: Fill a first image unit in a target area of ​​a background image.

[0041] In this embodiment, the target area is a 10*10 white square.

[0042] S103: Construct a second image unit, fill the second image unit into a corresponding area in the background image according to a preset shape of the image control points, obtain an image of the image control points, and store the image of the image control points.

[0043] The second image unit uses a 20*20 white square. In the background image, plan the overall area that can be filled by the second image unit, such as Figure 3 The area inside the green frame shown in the left image is 140*140. The second image unit is filled in the corresponding position of the area inside the green frame according to the shape of the image control point to be constructed. Taking image control point 0 as an example, the second position of the leftmost column and the bottom row of the area inside the green frame is filled according to the shape of image control point 0, as shown in the following example: Figure 3 (right picture) shown.

[0044] After the image acquisition of the control points is completed, the coordinates of each control point image are assigned, that is, the coordinate information is matched with the control point photo, for example:

[0045] GCP 1: x coordinate: 13234.00, y coordinate: 32313.00;

[0046] GCP 2: x coordinate: 13234.00, y coordinate: 12333.00;

[0047] GCP 3: x coordinate: 13234.00, y coordinate: 32323.00;

[0048] GCP 4: x-coordinate: 13234.00, y-coordinate: 32353.00.

[0049] S2. Preprocess the collected image control point images to obtain image control point sample data.

[0050] The number of image control point samples of each image control point is increased by rotating, scaling, stretching, adjusting brightness, adding noise and other technical means to form a training sample data set.

[0051] The image control points are randomly rotated, scaled, stretched, brightness adjusted, and noise added. One or more methods are used to process the image control points to a preset number of times. After processing, each image control point style will be composed of a corresponding number of types of expressions. In this example, the number of processing times is 500, and each image control point style will eventually have 500 different expressions. Taking image control point 0 as an example, the result after preprocessing is as follows: Figure 4 shown.

[0052] S3. Construct an initial image-controlled point prick point model, and train the image-controlled point prick point model using the image-controlled point sample data to obtain the image-controlled point prick point model.

[0053] A deep learning network model was built using Keras, with TensorFlow as the backend for computation. The image-controlled point prick model consists of an input layer, a convolutional layer, a max-pooling layer, and an output layer. The image-controlled point sample data (image) is fed into the input layer. The convolutional layer performs convolution operations on the image-controlled point sample data (image) to extract a feature map of the image-controlled point samples. This feature map is then fed into the max-pooling layer, which downsamples the feature map, filters the features within the receptive field, and extracts the largest feature within the receptive field. The resulting feature map containing the largest feature is flattened using the Flatten layer, which converts the multidimensional feature map into a one-dimensional image. The reduced image is then fed into a fully connected layer, which performs feature mapping on the image to obtain the final image-controlled point prick recognition image. This image is then fed into the output layer, where the final result is obtained. Key parameters of the image-controlled point prick model include a 5*5 convolution kernel size, the categorical_crossentropy loss function, and the Adam (Adaptive momentum) optimization function. The model is run (model iteration) 100 times, and the model parameters are automatically adjusted according to the loss function during the running process. Finally, the model with the optimal parameters is saved.

[0054] The training process is as follows Figure 5 shown.

[0055] S4. Input the image of the image control points to be detected into the image control point puncturing model, perform automatic puncturing of the image control points, and complete the image control point puncturing work.

[0056] Input the image of the control points to be detected into the trained image control point puncture model, identify the shape of the control points, identify the shape type, and obtain the corresponding coordinate data to complete the image control point puncture task, such as Figure 6 shown.

[0057] by Figure 7 Taking the image of the control point shown in (a) as an example, the image is input into the image control point puncture point model, and the recognition result is as follows: Figure 7 As shown in (b), the probability that the image control point in the image is image control point No. 12 is 0.96. It is judged (predict_label) to be No. 12. The image control point is manually marked (label) as No. 12, and the coordinates of No. 12 are matched with the image to complete the point puncture.

[0058] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0059] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. However, such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. An automatic point-puncturing method based on artificial intelligence, characterized by: The following steps are involved: Acquiring an image of image control points, wherein the image control points have different shapes, and setting specific coordinates of the image control points based on the shapes; Preprocessing the image of the control points to obtain sample data of the control points; Constructing an initial image-controlled point prick point model, and training the initial image-controlled point prick point model based on the image-controlled point sample data to obtain an image-controlled point prick point model; Based on the image control point puncture model, automatically puncture the image control points; Acquiring the image of the control point includes: Constructing a basic image, wherein the background of the basic image is black, and setting a first image unit in a target area of ​​the basic image; A second image unit is constructed, and according to the preset shape of the image control point, the corresponding position area in the basic image is filled with the second image unit to obtain the image control point image and store it.

2. The method for automatic image control point puncture based on artificial intelligence according to claim 1, characterized in that: The center position of the image control point is coordinate information.

3. The method for automatic image control point puncture based on artificial intelligence according to claim 1, characterized in that: Preprocessing the image of the control points includes: The image control point images are randomly rotated, scaled, stretched, brightness adjusted, and noise added, the number of the image control point images is increased, and the image control point sample data is obtained.

4. The method for automatic point marking based on image control points based on artificial intelligence according to claim 1, characterized in that: The image-controlled point-puncture model adopts a deep learning network.

5. The method for automatic image control point puncture based on artificial intelligence according to claim 1, characterized in that: The image-controlled point-puncture model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer, and the input layer, the convolutional layer, the pooling layer, the fully connected layer and the output layer are connected in this order.

6. The method for automatic image control point puncture based on artificial intelligence according to claim 1, characterized in that: During the training of the GCP model, the loss function uses the categorical_crossentropy function, and the optimization function uses the Adam function.

7. The method for automatic image control point puncture based on artificial intelligence according to claim 1, characterized in that: Based on the image control point puncture model, automatic puncture of image control points includes: The image of the image control points to be detected is input into the image control point puncturing model, the pattern of the image control points to be detected is identified, the specific coordinates of the image control points to be detected are extracted, and the image control points are automatically punctured.

Images