A target detection accuracy evaluation method based on relative position

Abstract

The present invention belongs to the field of aviation technology, and in particular relates to a method for evaluating target detection accuracy based on relative position, comprising: establishing a pixel coordinate system for an image in a real-time video acquired by a forward-looking camera of an aircraft; acquiring an internal parameter matrix of the forward-looking camera; acquiring the coordinates of pixel points of a runway bottom point in real time; testing the time from image acquisition to acquisition of the pixel points of the runway bottom point by the forward-looking camera of a drone; acquiring a first latitude and longitude and a first altitude of a GPS on the drone; acquiring the latitude and longitude of a landing point of the drone and the latitude and longitude of the GPS, and calculating the relative position of the drone relative to the runway landing point; calculating an ideal vertical coordinate of the pixel points of the runway bottom point; and determining an error mean and variance of the vertical coordinates of the pixel points of the runway bottom point detected in real time.

Description

Technical Field

[0001] The present invention belongs to the field of aviation technology, and in particular relates to a target detection accuracy evaluation method based on relative position. Background Art

[0002] With the rapid development of computer vision technology, the application scenarios of computer vision for target detection are becoming more and more extensive. In the fields of robotics and drones, there are very high requirements for the accuracy of target detection. In some special application scenarios, drones need to use the vertical coordinate of the runway bottom point pixel as the control quantity. The accuracy of the runway bottom point pixel value determines whether the aircraft can land safely. How to evaluate the detection accuracy of the runway bottom point pixel value is of great significance. Summary of the Invention

[0003] The technical problem solved by the present invention is as follows: by testing the target detection time and the image transmission delay, the latitude and longitude time scale of the GPS on the UAV is aligned with the time scale of the runway bottom point pixels detected in real time, the attitude correction matrix is ​​applied to eliminate the influence of the aircraft attitude change on the runway bottom point pixels, and the ideal runway bottom point longitudinal pixel value is calculated through perspective projection, so as to calculate the accuracy (error and error variance) of the runway bottom point pixel longitudinal coordinate.

[0004] The technical solution of the present invention:

[0005] A method for evaluating target detection accuracy based on relative position, the method comprising:

[0006] S1, establishes a pixel coordinate system for the image in the real-time video obtained by the aircraft's forward-looking camera;

[0007] S2, obtain the intrinsic parameter matrix of the front-view camera;

[0008] S3, real-time acquisition of pixel coordinates of the runway bottom point;

[0009] S4, the time it takes for the drone's front-view camera to capture an image and obtain the pixel points of the runway bottom point;

[0010] S5, obtaining the first latitude and longitude and the first altitude of the GPS on the drone;

[0011] S6, obtaining the latitude and longitude of the UAV landing point and the latitude and longitude of the GPS, and calculating the relative position of the UAV with respect to the runway landing point;

[0012] S7, calculates the ideal pixel vertical coordinate of the runway bottom point based on the relative position of the UAV to the runway landing point;

[0013] S8, determining the error mean and variance of the real-time detected runway bottom point pixel vertical coordinates based on the ideal runway bottom point pixel vertical coordinates and the real-time detected runway bottom point pixel vertical coordinates.

[0014] Furthermore, S1 is specifically as follows: the upper left corner of the image is the origin of the pixel coordinate system, the positive direction of the U axis points to the right, and the positive direction of the V axis points directly downward.

[0015] Furthermore, S2 is specifically as follows: using the camera annotation tool to obtain the camera's intrinsic parameter matrix K:

[0016]

[0017] Among them, f is the focal length of the camera, dx is the physical size of each pixel in the U-axis direction, dy is the physical size of each pixel in the V-axis direction, and (u0, v0) is the pixel coordinate of the center point of the image.

[0018] Furthermore, S3 is specifically as follows: obtaining the video information of the drone's front-view camera in real time, inputting the video information into the target detection network, and obtaining the pixel coordinates (u1, v1) of the runway bottom point in real time.

[0019] Furthermore, S4 is specifically: testing the time T0 from when the drone's front-view camera collects the image to when the target detection network obtains the pixel point of the runway bottom point.

[0020] Furthermore, S5 is specifically as follows: obtaining the time scale of the longitude and latitude of the GPS on the drone and the altitude information, and subtracting T0 to obtain the first longitude and latitude and the first altitude (long, lat, height) of the GPS on the drone.

[0021] Furthermore, S6 is specifically as follows: using the latitude and longitude height (lon0, lat0, height0) of the drone landing point and the latitude and longitude height (lon, lat, height) of the GPS, the relative position (tx, ty, tz) of the drone relative to the runway landing point is calculated, where tx is the longitudinal distance of the drone relative to the runway landing point, ty is the lateral distance of the drone relative to the runway, and tz is the height of the drone relative to the runway. The calculation method is as follows:

[0022] First intermediate variable:

[0023] Second intermediate variable:

[0024] Third intermediate variable: y_earth = reh573 × (lat - lat0)

[0025] Fourth intermediate variable: x_earth = reh573 × (long - lon0) × ye_correct

[0026] tx=x_earth×cos(runway_heading)+y_earth×sin(runway_heading)

[0027] ty=y_earth×cos(runway_heading)-x_earth×sin(runway_heading)

[0028] tz=height0-height_in

[0029] Where runway_heading is the runway heading.

[0030] Furthermore, S7 specifically includes: eliminating the influence of the drone's attitude on the pixel value of the runway bottom point through attitude correction. The ideal runway bottom point pixel vertical coordinate v_ideal satisfies the following relationship:

[0031]

[0032] The technical solution of the present invention designs a relative position target detection accuracy evaluation method. By testing the target detection time and image transmission delay, the latitude and longitude time scales of the GPS on the drone are aligned with the time scales of the runway bottom point pixels detected in real time. An attitude correction matrix is ​​applied to eliminate the impact of aircraft attitude changes on the runway bottom point pixels. The ideal runway bottom point longitudinal pixel value is calculated through perspective projection, thereby calculating the accuracy of the runway bottom point pixel longitudinal coordinate (error and error variance). BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of the pixel coordinates of the runway bottom point provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0035] The embodiment of the present invention provides a method for evaluating target detection accuracy based on relative position. Figure 1 As shown, the specific technical solutions are as follows:

[0036] 1) Define the pixel coordinate system in the image. The upper left corner of the image is the origin of the pixel coordinate system. The positive direction of the U axis points to the right, and the positive direction of the V axis points directly downward. Establish the pixel coordinate system.

[0037] 2) Use the camera annotation tool to obtain the camera's intrinsic parameter matrix K

[0038]

[0039] Where f is the focal length of the camera, dx is the physical size of each pixel in the U-axis direction, dy is the physical size of each pixel in the V-axis direction, and (u0, v0) is the pixel coordinate of the center point of the image;

[0040] 3) Obtain the video information from the drone’s front-view camera in real time, input the video information into the target detection network, and obtain the pixel coordinates (u1, v1) of the runway bottom point in real time;

[0041] 4) Test the time T0 from the drone's front-view camera capturing the image to the target detection network acquiring the runway bottom pixel;

[0042] 5) Compare the GPS latitude, longitude, and altitude information with the time scale of the runway bottom pixel: Subtract T0 from the time scale of the GPS latitude, longitude, and altitude information on the drone to obtain the GPS latitude, longitude, and altitude (long, lat, height);

[0043] 6) Using the latitude and longitude of the drone landing point (lon0, lat0, height0) and the latitude and longitude of the GPS (lon, lat, height), calculate the relative position of the drone relative to the runway landing point as (tx, ty, tz), where tx is the longitudinal distance of the aircraft relative to the runway landing point, ty is the lateral distance of the aircraft relative to the runway, and tz is the height of the aircraft relative to the runway. The calculation method is as follows:

[0044]

[0045]

[0046] y_earth = reh573 × (lat - lat0)

[0047] x_earth=reh573×(long-lon0)×ye_correct

[0048] tx=x_earth×cos(runway_heading)+y_earth×sin(runway_heading)

[0049] ty=y_earth×cos(runway_heading)-x_earth×sin(runway_heading)

[0050] tz=height0-height_in

[0051] Where runway_heading is the runway heading.

[0052] 7) Attitude correction is used to eliminate the influence of the drone's attitude on the pixel value of the runway bottom point. The ideal vertical coordinate v_ideal of the runway bottom point pixel satisfies the following relationship:

[0053]

[0054] 8) Based on the real-time ideal runway bottom point longitudinal pixel coordinate v_ideal and the real-time detected runway bottom point pixel longitudinal coordinate v1, calculate the error mean △ and variance σ of v1.

[0055] The technical solution of the present invention designs a relative position target detection accuracy evaluation method. By testing the target detection time and image transmission delay, the latitude and longitude time scales of the GPS on the drone are aligned with the time scales of the runway bottom point pixels detected in real time. An attitude correction matrix is ​​applied to eliminate the impact of aircraft attitude changes on the runway bottom point pixels. The ideal runway bottom point longitudinal pixel value is calculated through perspective projection, thereby calculating the accuracy of the runway bottom point pixel longitudinal coordinate (error and error variance).

Claims

1. A method for evaluating target detection accuracy based on relative position, characterized in that: The method comprises: S1, establishes a pixel coordinate system for the image in the real-time video obtained by the aircraft's forward-looking camera; S2, obtain the intrinsic parameter matrix of the front-view camera; S3, real-time acquisition of pixel coordinates of the runway bottom point; S4, the time it takes for the drone's front-view camera to capture an image and obtain the pixel points of the runway bottom point; S5, obtaining the first latitude and longitude and the first altitude of the GPS on the drone; S6, obtaining the latitude and longitude of the UAV landing point and the latitude and longitude of the GPS, and calculating the relative position of the UAV with respect to the runway landing point; S7, calculates the ideal pixel vertical coordinate of the runway bottom point based on the relative position of the UAV to the runway landing point; S8, determining the error mean and variance of the real-time detected runway bottom point pixel vertical coordinates based on the ideal runway bottom point pixel vertical coordinates and the real-time detected runway bottom point pixel vertical coordinates.

2. The target detection accuracy evaluation method based on relative position according to claim 1, characterized in that: Specifically, S1 is as follows: the upper left corner of the image is the origin of the pixel coordinate system, the positive direction of the U axis points to the right, and the positive direction of the V axis points directly downward.

3. The target detection accuracy evaluation method based on relative position according to claim 2, characterized in that: S2 is specifically: using the camera annotation tool to obtain the camera's intrinsic parameter matrix K: in, f is the camera focal length, dx is the physical size of each pixel in the U-axis direction, dy is the physical size of each pixel in the V-axis direction, ( , ) is the pixel coordinate of the center point of the image.

4. The target detection accuracy evaluation method based on relative position according to claim 3 is characterized in that: S3 is specifically: real-time acquisition of drone front camera video information, and input of video information into the target detection network, real-time acquisition of runway bottom point pixel coordinates ( , ).

5. The target detection accuracy evaluation method based on relative position according to claim 4 is characterized in that: S4 is specifically: testing the time T0 from the drone's front-view camera collecting pictures to the target detection network obtaining the pixel point of the runway bottom.

6. The target detection accuracy evaluation method based on relative position according to claim 5, characterized in that: S5 is specifically as follows: obtaining the time scale of the GPS latitude and longitude and altitude information on the drone, and subtracting T0 to obtain the first latitude and longitude and the first altitude of the GPS on the drone ( long , lat , height ).

7. The target detection accuracy evaluation method based on relative position according to claim 6, characterized in that: S6 is specifically: using the latitude and longitude of the drone landing point ( lon 0, lat 0, height 0) and GPS latitude and longitude ( long , lat , height ) Calculate the relative position of the drone relative to the runway landing point ( tx , ty , tz ), tx is the longitudinal distance of the aircraft relative to the landing point on the runway, ty is the lateral distance of the aircraft relative to the runway, tz is the height of the aircraft relative to the runway, calculated as follows: First intermediate variable: Second intermediate variable: The third intermediate variable: Fourth intermediate variable: in, is the runway heading.

8. The target detection accuracy evaluation method based on relative position according to claim 7, characterized in that: S7 specifically: eliminate the influence of the drone's attitude on the pixel value of the runway bottom point through attitude correction, the ideal runway bottom point pixel vertical coordinate The following relationship is satisfied: 。

Images