CN1453740A - Method of measuring scene and geometric data of bodies inside the scene via single frame of image - Google Patents

Abstract

A method for measuring the geometric information of a scene and objects in the scene from a single image, comprising the steps of: 1) taking one or more images of the scene to be measured and inputting them into a computer; Point or straight line information, parallel line information perpendicular to the reference plane and the two endpoints of the reference distance; 3) use the information of the point or straight line in the reference plane extracted in step 2 and its corresponding relationship with the scene to find the reference plane The mapping relationship matrix H to the image plane; 4) using the point or line information extracted in step 2, calculate the hidden points perpendicular to the reference plane direction and the hidden lines of the reference plane on the image. 5) Extract the boundary information of the object to be measured from the image and fit it into a straight line or a quadratic curve. For the upper and lower surfaces of the object, use the epipolar relationship on the image or the properties of the object surface itself to find the The coordinates of the center point of the surface; 6) using the parameters obtained in the above steps to measure the geometric information of the scene and objects in the scene. These information include: the coordinates of the spatial point, the distance, the angle between the straight line, the distance from the point to the straight line, the area of the plane figure and the volume of the object, etc.

Description

Method for Measuring Geometric Information of Scene and Objects in Scene from a Single Image

technical field

The present invention relates to computer aided automatic measurement.

Background technique

In our daily life and in many fields, measurement is essential, such as traffic accidents, criminal crime scenes, buildings, construction sites, etc., require a lot of measurement work. The traditional method is to use manual measurement, which is very time-consuming and prone to large measurement errors, and this measurement is also prone to unfair results due to the subjective factors of the measurers.

In recent years, with the development and application of computer vision, people have begun to explore the use of computer vision methods to measure through images. In this type of method, it is typical to reconstruct the three-dimensional structure of the scene under test by two or more images using stereo vision, but this method involves a classic problem in the field of computer vision, that is, the matching between multiple images. , so it is difficult to put into practical application. Another method is to manually measure the information of more than 4 control points in the scene plane in advance to calculate the projection matrix between the space plane and the image plane, and then measure through a single image. The measurement accuracy and control of this method The selection accuracy, location and number of points are directly related, but in practical applications, sometimes it is difficult to select control points, and the measurement accuracy of control points cannot be guaranteed.

Contents of the invention

The purpose of the present invention is to provide a measurement method based on a single image that is relatively simple, practical, has high measurement accuracy and robustness.

In order to achieve the above purpose, the method for measuring the geometric information of the scene and the objects in the scene from a single image comprises steps:

1). Take one or more images of the scene under test and input them into the computer;

2). Extract the information of points or straight lines in the reference plane in the scene, the information of parallel lines perpendicular to the reference plane and the two endpoints of the reference distance from the image;

3). Utilize the information of points or straight lines in the reference plane extracted in step 2 and the corresponding relationship with the scene to obtain the mapping relationship matrix H from the reference plane to the image plane;

4). Using the point or line information extracted in step 2, calculate the hidden point in the direction perpendicular to the reference plane and the hidden line of the reference plane on the image.

5). Extract the boundary information of the object to be measured from the image and fit it into a straight line or a quadratic curve. For the upper and lower surfaces of the object, use the epipolar relationship on the image or the properties of the object surface itself to obtain the coordinates of the center point of the surface;

6). Use the parameters obtained in the above steps to measure the geometric information of the scene and the objects in the scene. These information include: the coordinates of the spatial point, the distance, the angle between the straight line, the distance from the point to the straight line, the area of the plane figure and the volume of the object, etc.

The present invention can perform the following measurements: the coordinates of any point in the reference plane in the scene coordinate system and the distance between any two points; the height of any point in space relative to the reference plane, the coordinates in the scene coordinate system and any two points The distance between points; the distance from a point to a straight line in the reference plane; the angle between any two straight lines in the reference plane and the plane parallel to the reference plane in the scene, the area of the plane figure, and some relatively regular and symmetrical objects Volume, such as cube, cylinder, cone, frustum of a cone, etc. At the same time, the error range of each measurement result can be given according to the uncertainty analysis of each step. The measurement method provided by the invention does not need to calibrate the camera parameters, and has the characteristics of simplicity, practicability, high measurement accuracy, good robustness, etc., making the vision-based measurement method practical.

Description of drawings

Fig. 1 is the flowchart of this measuring method;

Figure 2 The schematic diagram of coordinates and height measurement of points outside the reference plane

Figure 3 Schematic diagram of measuring the distance from a point to a straight line in the reference plane

Fig. 4 is a measurement scene diagram of the embodiment.

Specific Embodiments of the Invention

The method of the present invention requires that the scene contains a reference plane and two or more parallel lines perpendicular to the reference plane and a reference distance, and some known geometric information should be included in the reference plane, such as information of some points or straight lines, This information can be obtained from the window frame, door frame, floor texture, outline of regular objects, etc. in the scene, or by placing a template with known geometric information in the scene. At the same time, there should be a reference distance in the scene. It can be measured by taking an image with an ordinary camera or a digital camera, and there is no need to calibrate the camera parameters. If multiple images can be taken from different angles, you can choose a better one for measurement or comprehensively use the same camera to measure. Measure measurements in multiple images to improve measurement accuracy.

For an image, the method mainly includes image acquisition, key feature extraction, transformation matrix calculation, hidden line and hidden point calculation, object feature extraction and calculation, measurement and other steps, as shown in Figure 1, each step The specific instructions are as follows: 1. Get the image

Take one or more images of the scene to be tested with an ordinary camera or a digital camera, and input the images into the computer by means of a scanner or some special interface. 2. Key feature extraction

Extract the point or line information in the reference plane in the scene, the parallel line information perpendicular to the reference plane, and the two endpoints of the reference distance from the image by automatic or human-computer interaction. If a template is placed in the scene, geometric information of known points or straight lines contained in the template is extracted from the image. Among them, Canny, Robert, Sobel and other edge detection operators can be used to extract edge information, and then Hough transform, least squares, maximum likelihood estimation and other methods can be used to obtain line parameters. 3. Transformation matrix calculation

Using the point or line information in the reference plane extracted in the second step and its corresponding relationship with the scene, the mapping relationship matrix H from the reference plane to the image plane is obtained, where H is a 3×3 matrix. Taking the correspondence of points as an example, assuming that the homogeneous coordinates of points in the scene are X _i , and the corresponding points in the image are m _i , then the mapping relationship matrix H can be obtained by m _i =sHX _i . 4. Calculation of hidden line and hidden point

Use the information of parallel lines perpendicular to the reference plane extracted in the second step to calculate the hidden point V _r of the image in this direction; use the transformation matrix obtained in the third step to calculate the hidden line L of the reference plane in the image _v , or directly use the information in the reference plane extracted in the second step to calculate the hidden line. 5. Extraction and calculation of object features

Using a method similar to step 2, extract the boundary information of the object to be measured from the image through automatic or human-computer interaction methods and fit it segmentally into a straight line or a quadratic curve. For the upper and lower surfaces of the object, the coordinates of the center point of the surface are obtained by using the antipolar relationship on the image or the properties of the object surface itself. 6. Measurement

The present invention can perform the following measurements: the coordinates of any point in the reference plane in the scene coordinate system and the distance between any two points; the height of any point in space relative to the reference plane, the coordinates in the scene coordinate system and any two points The premise of measurement is that the vertical point of the point in the reference plane is known; the reference plane in the scene and the angle between any two straight lines in the plane parallel to the reference plane; the point to the reference plane in the straight line Distance; the area of plane figures and the volume of some relatively regular and symmetrical objects, etc. At the same time, the error range of each measurement result can be given according to the uncertainty analysis of each step. The specific measurement method is as follows:

(1). Measurement of coordinates and distances of points in the reference plane: For any point m on the reference plane image, the coordinate value X of the corresponding point in the scene can be obtained by using the inverse transformation of the mapping relationship matrix obtained in the third step =H ^-1 m, and then the distance between any two points in the reference plane can be calculated.

(2). Measurement of the coordinates and height of points outside the reference plane: as shown in Figure 2, _Lv and _Vr are the hidden lines and hidden points obtained in the fourth step, respectively; S ₀ and E ₀ are the reference distances , then the three points S ₀ , E ₀ and V _r are collinear L _r ; S ₁ and E ₁ are the two endpoints of the height to be measured, then the point S ₁ can be mapped to a straight line with the help of the hidden line L _v L _r on the point H ₁ that has the same height as it. In this way, the spatial distance corresponding to S ₁ and E ₁ can be obtained according to the property that the projective transformation keeps the cross ratio unchanged. On the other hand, since the point E ₁ is in the reference plane, its coordinates can be obtained by (1), thus the coordinates of the spatial point corresponding to the point S ₁ can be obtained.

(3). Measurement of the distance between any two points in the scene: From (1) and (2), the distance between any two points in the reference plane and the height of the space point can be obtained through simple geometric algebra operations The distance between any two points in the scene.

(4). Measurement of the included angle between any two straight lines in the reference plane: Let the two intersection points of the two straight lines and the blanking line L _v be P ₁ and P ₂ , and the mapping relationship matrix H obtained in the third step can be The images I and J of the two circle points are obtained, then the four points P ₁ , P ₂ and I and J are collinear, and the two The angle between the lines. For example: if the cross ratio is Cr, then the included angle θ=1n(Cr)/(2*i). It is also possible to map this straight line to a space plane through a transformation matrix, and directly calculate the angle between them according to the equations of the two straight lines.

(5). The measurement of the area of any plane figure in the reference plane and the plane parallel to the reference plane: for a general plane figure, detect its edge information on the image and map it to the space plane through the mapping relationship matrix H, Then obtain the approximate value of its area by integral method; for rectangle and triangle, can directly obtain its side length by (1) and utilize rectangle and triangle area formula to calculate area; for circle, extract its edge on the image and approximate it with quadratic curve Combine its equation, use the antipolar relationship between the quadratic curve and the hidden line to find the coordinates of its center, and calculate the distance between the center and any edge point by (1) as the radius, so that the area of the circle can be used formula to get its area.

(6). Measurement of object volume: the bottom area and cross-sectional area of the object can be measured by (5), and the height of the object can be measured by (2). In this way, the volume calculation formula can be obtained by simple algebraic or geometric operations Find the volume of the object. For example: for a cone, the boundary information of the bottom is extracted from step 5 and fitted into a quadratic curve and the coordinates of its center point are obtained, so that the area and height of the bottom can be calculated to obtain the volume; for the circular frustum, respectively Extract the boundary information of the upper and lower surfaces and fit them into a quadratic curve, calculate the area of the upper and lower surfaces, and find the coordinate center of its center point, then the distance between the two center points is the height, so you can use the volume calculation formula to find out of volume.

(7). Measurement of the distance from a point to a straight line in the reference plane: the measurement method is shown in Figure 3, and the distance from point A to straight line BC in the reference plane is calculated. Let the intersection point of BC and the blanking line L _v be D, the two ring points on L _v be I, J, calculate the point E on L _v with the intersection ratio of D, I, J -1, connect A, E and BC intersect at point H, and the distance between A and H calculated by (1) is the distance from point A to straight line BC. To calculate the distance from a point S in space to the straight line BC, you can first measure the height of point S by (2), and then calculate the distance from the vertical point of point S in the reference plane to the straight line BC, then use simple algebraic or geometric operations The distance from point S to line BC can be calculated. It is also possible to map the straight line to the space plane through the transformation matrix, calculate the coordinates of the point in the scene coordinate system by (2), and then directly calculate the distance from the point to the straight line according to the equation of the straight line and the coordinates of the space point.

Example

Figure 4 shows an image of a tested scene. We use edge detection and line fitting algorithms to extract the known floor information and information of parallel lines perpendicular to the floor, and calculate the mapping relationship matrix by least squares method , blanking line and blanking point, the reference distance is the height H ₀ of the iron cabinet. According to the method of the present invention, select two points with the mouse in the image, and the distance between these two points in the scene can be directly obtained from the image, and the measurement results are as shown in Figure 4. Among them, V is the volume of the round iron drum. It can be seen that the distances D ₁ and D ₂ are difficult to measure in real space. The measured values obtained by the method of the present invention are shown in FIG. 4 . The real values of the measured values in the figure are as follows: human height H ₁ =171.5cm; control cabinet height H ₂ =170.4cm; distance D ₁ =302.7cm; D ₂ =389.2cm; diameter of round iron drum is 24cm, height is 59.3cm, and the volume is V=26826.7cm ³ . The measurement relative error of the present invention is generally within 4%, which can fully meet the measurement accuracy requirements.

Claims (13)

1. measure the method for scene and scene geometric data of bodies inside by single image for one kind, comprise step:

1) takes the image of the tested scene of one or more and import computing machine;

2) from image, extract in the scene point in the reference planes or the information of straight line, two end points of parallel lines information vertical and reference distance with reference planes;

3) utilize point in the reference planes that step 2 extracts or straight line information and with the corresponding relation of scene, obtain the mapping relations matrix H of reference planes to the plane of delineation.

4) utilize the point that step 2 extracts or the information of line, at the blanking line that calculates on the image perpendicular to vanishing point on the reference planes direction and reference planes.

5) from image, extract the boundary information of the object of need measuring and its piecewise fitting is in line or quafric curve,, utilize character on the image to obtain the coordinate of this centre of surface point utmost point relation or body surface self for the upper and lower surface of object;

6) utilize the geological information of object in parameter measurement scene that above-mentioned steps obtains and the scene, these information comprise: coordinates of spatial points, distance, included angle of straight line, the distance of putting straight line, planar graph area and object volume etc.

2. by the described method of claim 1, it is characterized in that in scene, placing template, to calculate mapping relations matrix, vanishing point and blanking line by single image measurement scene and scene geometric data of bodies inside.

3. measure the method for scene and scene geometric data of bodies inside by claim 1 is described by single image, it is characterized in that utilizing in the scene to be measured reference planes and measure perpendicular to the information of consult straight line.

4. by the described method of measuring scene and scene geometric data of bodies inside by single image of claim 1, it is characterized in that the point of height to be measured is mapped on the straight line at reference distance place, keep the constant character of double ratio can obtain its corresponding spatial altitude according to projective transformation, and in conjunction with its vertical its corresponding volume coordinate that measures of putting in reference planes.

5. by the described method of claim 1, it is characterized in that utilizing the distance of any point-to-point transmission in the reference planes and the height of spatial point, try to achieve the distance between any 2 in the scene by single image measurement scene and scene geometric data of bodies inside.

6. by the described method of measuring scene and scene geometric data of bodies inside by single image of claim 1, it is characterized in that two intersection points and two double ratios that annulus point constituted, keep the constant character of double ratio to calculate angle between two straight lines in conjunction with projective transformation according to two straight lines and blanking line.

7. measure the method for scene and scene geometric data of bodies inside by claim 1 is described by single image, it is characterized in that straight line being mapped on the space plane, directly calculate its angle according to the equation of two straight lines by transformation matrix.

8. by the described method of measuring scene and scene geometric data of bodies inside by single image of claim 1, it is characterized in that for being positioned at reference planes or being parallel to the planar graph on the plane of reference planes, detecting its marginal information on the image and it being mapped on the space plane, obtain the approximate value of its area then by integral method by the mapping relations matrix.

9. measure the method for scene and scene geometric data of bodies inside by claim 1 is described by single image, thus it is characterized in that utilizing with two annulus point double ratios be that-1 method is asked for some the distance of the intersection point calculation level of straight line in the reference planes to straight line.

10. by claim 1 or the 4 described methods of measuring scene and scene geometric data of bodies inside by single image, it is characterized in that straight line being mapped on the space plane, according to the equation of straight line and the direct calculation level of coordinates of spatial points distance to straight line by transformation matrix.

11., it is characterized in that utilizing the utmost point is concerned between quafric curve and blanking line to obtain round centre coordinate by the described method of claim 1 by single image measurement scene and scene geometric data of bodies inside.

12., it is characterized in that utilizing the centre coordinate of justifying pairing image quafric curve to calculate the height of objects such as cylinder, circular cone, round platform by claim 1 or 11 described methods by single image measurement scene and scene geometric data of bodies inside.

13. by claim 4 or 12 described methods, it is characterized in that calculating floorage, sectional area and the height of object, thereby calculate the volume of object by single image measurement scene and scene geometric data of bodies inside.

Images