CN118135011A - Meteorological station surrounding obstacle elevation angle measuring method based on visual recognition technology - Google Patents

Abstract

The invention relates to the technical field of image processing and machine vision, and discloses a method for measuring the elevation angle of a surrounding obstacle of a weather station based on a visual recognition technology, which comprises the steps of constructing a target detection model of the obstacle and a staff gauge, and training the detection of the staff gauge and the obstacle by adopting a yoloact deep learning model; acquiring scale information in a camera reference position image, extracting a scale contour line, and obtaining the position information of a scale; the camera rotates in the horizontal plane from the reference position to obtain 360-degree panoramic information, and the angle information corresponding to each picture is recorded. And successively inputting the acquired pictures into a trained target detection model to perform target detection, extracting category and contour information of a target obstacle, and calculating the elevation angle of the obstacle by combining the scale information and known scale parameters. Compared with the prior art, the invention obtains the elevation angle information of the obstacle according to the perspective imaging principle and the triangle similarity principle by proportion measurement and calculation, and can reflect the change condition of the obstacle around the observation field in time.

Description

A method for determining the elevation angle of obstacles around a weather station based on visual recognition technology

Technical Field

The invention relates to the technical field of image processing and machine vision, and in particular to a method for measuring the elevation angle of obstacles around a weather station based on visual recognition technology.

Background technique

When selecting a site for a ground-based meteorological observation station, it is necessary to consider the impact of surrounding obstacles on the measurement. The corresponding evaluation index is called the "horizon obstruction elevation angle". For the measurement of the "horizon obstruction elevation angle", the traditional method is to set up the theodolite horizontally at the center of the observation field, with the lens 1.5m above the ground and the azimuth disk 0° facing due north. The maximum obstruction elevation angle of the terrain within the visible range is measured, starting from due north, in a clockwise direction, and the azimuth is measured every 2°, and then the cumulative value of the obstruction viewing angle factor is calculated based on the comprehensive measurement results.

However, traditional measurement methods are time-consuming and labor-intensive, cannot achieve routine detection, and cannot promptly reflect changes in obstacles around the observation field.

Summary of the invention

Purpose of the invention: In view of the problems existing in the prior art, the present invention provides a method for measuring the elevation angles of obstacles around a weather station based on visual recognition technology. The method collects video image data through a camera, extracts the outline of the target obstacle, and obtains the elevation angle information of the obstacle by proportional measurement based on the perspective imaging principle and the triangle similarity principle. The method can obtain the elevation angle information of the surrounding obstacles by performing a 360° rapid scan of the surrounding obstacles.

Technical solution: The present invention provides a method for measuring the elevation angle of obstacles around a weather station based on visual recognition technology, comprising the following steps:

Step 1: Build a target detection model for conventional obstacles and rulers. The target detection model uses the YOLO deep learning model to train the detection of rulers and obstacles.

Step 2: Collect the scale information in the camera reference position image, extract the scale contour line, and obtain the scale position information and height information;

Step 3: The camera rotates in the horizontal plane from the reference position to obtain 360° panoramic information and record the angle information corresponding to each picture;

Step 4: Input the collected images into the target detection model trained in step 1 one by one for target detection, extract the category and contour information of the target obstacle, and calculate the obstacle elevation angle by combining the scale information in step 2 and the known scale parameters.

Furthermore, the step 1 includes the following steps:

Step 1.1: Collect pictures and establish a ruler and obstacle data set; the obstacles include mountains, structures, and trees; set a ruler, and the height of the ruler is consistent with the height of the camera center point;

Step 1.2: Train the target detection model, use labelme software to mark the outlines of the target obstacles and rulers, generate a json file, and then use a script to convert it into a yolo format file;

Step 1.3: The Python script file creates the collected images into a training data set file "train.txt" and a test data set file "test.txt";

Step 1.4: Input the obtained dataset file into the YOLO deep learning model to perform model training for obstacle target detection.

Furthermore, the specific steps of step 2 are:

Step 2.1: Ruler positioning: Select a feature ruler that is highly consistent with the center position of the camera and place it in front of the camera. The distance should be such that a complete image can be presented within the camera's field of view.

Step 2.2: Image preprocessing: After obtaining the image containing the ruler through the camera, take the center position of the image as the calibration, and take one-fifth of the image to the left and right to ensure the center position of the ruler;

Step 2.3: Input the image into the object detection model trained in step 1, perform ruler recognition, extract the ruler contour, and calculate the height c' of the ruler in the image based on the longitudinal contour.

Furthermore, in the step 2.2, after obtaining the central image, the image size is changed without distortion, and the image is changed to a size of 640×640.

Furthermore, the frequency of image acquisition in step 3 is no more than 2°, that is, image information is acquired at least once every 2°.

Furthermore, the specific steps of calculating the obstacle elevation angle in step 4 are:

1) Obtain the central image according to the image preprocessing method in step 2.2 and change the image size to 640×640;

2) Input the resized image into the target detection model trained in step 1 to identify the category contours of the target obstacles and extract relevant information;

3) Extract characteristic size: Extract the coordinates of the highest point of the obstacle contour line at the vertical center line of each image, calculate the difference between the coordinates of the highest point and the vertical coordinate of the center point of the image, and its absolute value represents the size b' of the obstacle above the ruler in the image;

4) Elevation angle calculation: Combine the actual height b of the ruler and the actual distance a between the ruler and the camera to calculate the elevation angle using the principle of similar triangles. The relevant calculation formula is:

tana＝b/a (2)

a＝arctan(b/a) (3)

Wherein, b' is the size of the obstacle above the ruler in the figure; c' is the height of the ruler in the figure; c is the actual height of the ruler; and a is the actual distance between the ruler and the camera.

Beneficial effects:

The present invention collects video image data through a camera, extracts the circumscribed rectangular frame of the target obstacle, and obtains the elevation angle information of the obstacle by proportional measurement based on the perspective imaging principle and the triangle similarity principle. This method can obtain the elevation angle information of the obstacles around the weather station by performing a 360° rapid scan of the surrounding obstacles. Compared with the traditional manual measurement method, this method can provide higher detection accuracy, is efficient and fast, can realize routine detection, and can timely reflect the changes in obstacles around the observation field.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is an obstacle data set collected by an embodiment of the present invention;

FIG2 is a diagram showing the relationship between the scale and the camera selected in an embodiment of the present invention;

FIG3 is a schematic diagram of camera, scale, obstacle perspective imaging and triangle similarity principle according to an embodiment of the present invention;

FIG4 is a diagram showing the category contours of target obstacles and scales after being detected by a target detection model according to an embodiment of the present invention;

Figure 5 is a framework diagram of the Yolo deep learning model of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

The present invention discloses a method for measuring the elevation angle of obstacles around a weather station based on visual recognition technology. The method collects video image data through a camera, extracts the outline of the target obstacle, and obtains the elevation angle information of the obstacle by proportional measurement based on the perspective imaging principle and the triangle similarity principle. The method can obtain the elevation angle information of the surrounding obstacles by performing a 360° rapid scan of the surrounding obstacles. Specifically, the method comprises the following steps:

1. Obstacle target detection based on YOLO.

Collect conventional obstacle images (mountains, structures, trees, etc.), establish an obstacle dataset, label the obstacle images by category, and build an obstacle target detection model. See Figure 1, which shows the selected obstacle dataset.

Calibrate the horizontal viewing angle of the camera: Select two rulers that are at the same height as the center point of the camera, and adjust the camera angle so that the two points that are at the same height as the camera overlap each other in the camera viewing angle. At this time, the camera is in a horizontal viewing angle. See Figure 2 for details.

When training the target detection model, the specific steps include:

1) Train the target detection model. The target detection model uses the YOLO deep learning model to train the ruler and obstacles, see Figure 5. Use the Labelme software to annotate the contours of the target obstacles and rulers, annotate to generate a JSON file, and then use a script to convert it into a YOLO format file.

2) The Python script file converts the collected images into a training data set file "train.txt" and a test data set file "test.txt".

Collect the ruler information in the camera reference position image, extract the ruler contour line, and obtain the ruler position information and height information, as follows:

Step 2.1: Ruler positioning: Select a feature ruler that is highly consistent with the center position of the camera and place it in front of the camera. The distance should be such that a complete image can be presented within the field of view of the camera.

Step 2.2: Image preprocessing: After obtaining the image containing the ruler through the camera, take the center position of the image as the calibration, and take one-fifth of the image to the left and right to ensure the center position of the ruler. After obtaining the center image, resize the image without distortion and change the image to 640×640 size.

Step 2.3: Input the image into the object detection model trained in step 1, perform ruler recognition, extract the ruler contour, and calculate the height c' of the ruler in the image based on the longitudinal contour.

The camera rotates in the horizontal plane from the reference position to obtain 360° panoramic information and record the angle information corresponding to each picture. The picture acquisition frequency is no more than 2°, that is, image information is collected at least once every 2°.

The collected images are input into the target detection model trained in step 1 one by one for target detection, the category and contour information of the target obstacle are extracted, and the obstacle elevation angle is calculated by combining the scale information and known scale parameters in step 2.

2. Obstacle elevation angle calculation based on target detection

Basic idea: According to the triangle similarity principle, the elevation angle of the obstacle can be calculated using b and a (the distance between the camera and the ruler) in Figure 3 below. b is the characteristic size of the obstacle at the ruler. Through target detection and image feature extraction, a is the distance between the camera and the ruler. The proportional relationship between the size b' of the obstacle above the ruler in the image and the height c' of the ruler in the image can be obtained, and then the elevation angle data can be obtained according to formulas (1) and (2).

tana＝b/a (2)

a＝arctan(b/a) (3)

Wherein, b' is the size of the obstacle above the ruler in the figure; c' is the height of the ruler in the figure; c is the actual height of the ruler; and a is the actual distance between the ruler and the camera.

1) Object Detection:

As shown in FIG4 , the bounding box of the ruler and obstacles can be detected in the image through the target detection model, and the coordinate position and length and width data of the bounding box can be obtained through feature extraction.

2) Feature size extraction

The vertical coordinate of the center point of the image is the position of the camera horizontal plane. The difference between the vertical coordinate of the highest point of the obstacle boundary box and the vertical coordinate of the center point is the value of b' in the calculation formula (1); and the height of the ruler boundary box in the image is c' in formula (1).

3) Elevation angle calculation

Formula (1) can be used to calculate the perspective height b of the obstacle at the location of the scale; then formula (2) is used to calculate the tangent value of the elevation angle, and formula (3) is used to calculate the elevation angle value.

The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims (6)

1. A method for measuring the elevation angle of a surrounding obstacle of a weather station based on a visual recognition technology is characterized by comprising the following steps:

Step 1: constructing a target detection model of a conventional obstacle and a ruler, wherein the target detection model adopts a yolo deep learning model to train the detection of the ruler and the obstacle;

step 2: acquiring scale information in a camera reference position image, extracting a scale contour line, and obtaining position information and height information of a scale;

step 3: the camera rotates in a horizontal plane from a reference position to obtain 360-degree panoramic information, and angle information corresponding to each picture is recorded;

Step 4: and (3) sequentially inputting the acquired pictures into the target detection model trained in the step (1) to perform target detection, extracting the category and contour information of the target obstacle, and calculating the elevation angle of the obstacle by combining the scale information and the known scale parameters in the step (2).

2. The method for measuring the elevation angle of the obstacle around the weather station based on the visual recognition technology according to claim 1, wherein the step 1 comprises the steps of:

step 1.1: collecting pictures, and establishing a scale and obstacle data set; the obstacle comprises a mountain, a structure and a tree; setting a scale, wherein the height of the scale is consistent with the height of the center point of the camera;

step 1.2: training a target detection model, performing contour labeling on a target obstacle and a scale by using labelme software, generating a json file by labeling, and then converting the json file into a yolo-format file by using a script;

step 1.3: the Python script file makes the collected pictures into a training data set file 'train. Txt' and a test data set file 'test. Txt';

Step 1.4: and inputting yolo the obtained data set file into a deep learning model, and performing model training of obstacle target detection.

3. The method for measuring the elevation angle of the obstacle around the weather station based on the visual recognition technology according to claim 1, wherein the specific steps of the step 2 are as follows:

Step 2.1: positioning a scale: selecting a characteristic scale with the height consistent with the central position of the camera, and placing the characteristic scale in front of the camera, wherein the distance is based on the fact that a complete picture can be presented in the view range of the camera image;

Step 2.2: preprocessing a picture: after the picture containing the scale is obtained through the camera, taking the center position of the picture as a calibration, and taking one fifth of the picture left and right to ensure the center position of the scale;

Step 2.3: inputting the image into the target detection model trained in the step 1, carrying out scale recognition, extracting the scale outline, and calculating the height c' of the scale in the figure according to the longitudinal outline.

4. The method for measuring elevation angle of obstacle around weather station based on visual recognition technology according to claim 3, wherein in step 2.2, after the center picture is obtained, the picture is changed to 640 x 640 size by changing the image size without distortion.

5. The method for measuring the elevation angle of the obstacle around the weather station based on the visual recognition technology according to claim 1, wherein the picture acquisition frequency in the step 3 is not more than 2 degrees, namely, the image information is acquired at least every 2 degrees.

6. The method for measuring the elevation angle of the obstacle around the weather station based on the visual recognition technology according to claim 3, wherein the specific step of calculating the elevation angle of the obstacle in the step 4 is as follows:

1) Acquiring a center picture according to the image preprocessing method in the step 2.2, and changing the picture into 640 multiplied by 640;

2) Inputting the image with the modified size into the target detection model trained in the step 1, identifying the category contour of the target obstacle, and extracting relevant information;

3) Extracting characteristic dimensions: extracting the highest point coordinate of the barrier contour line at the vertical center line of each picture, and calculating the difference value between the highest point coordinate and the ordinate of the image center point, wherein the absolute value of the difference value represents the dimension b' of the barrier higher than the scale in the picture;

4) Elevation angle calculation: the elevation angle is calculated by combining the actual height b of the scale and the actual distance a between the scale and the camera through a similar triangle principle, and the related calculation formula is as follows:

tana＝b/a (2)

a＝arctan(b/a) (3)

Wherein b' is the dimension of the obstacle higher than the scale in the figure; c' is the height of the scale in the figure; c is the actual height of the scale; a is the actual distance between the scale and the camera.