Automobile shape visual inspection system and method based on angle similarity invariance

Abstract

The present invention discloses a vehicle shape visual detection system and method based on angle similarity invariance, aiming to solve the problem of active visual detection of vehicle shape based on angle similarity invariance. The vehicle shape active visual reconstruction system based on angle similarity invariance mainly consists of a camera bracket (1), a camera (2), a connecting piece (3), a line laser (4) and a two-dimensional reference plate (5). The vehicle shape visual detection method based on angle similarity invariance consists of three steps: image acquisition, obtaining a transformation matrix from a projection space to a similarity space according to multiple pairs of parallel straight lines and orthogonal straight lines, and reconstructing a laser projection point where a laser line intersects with a vehicle body surface. A vehicle shape visual detection system and method based on angle similarity invariance with a simple structure and reliable performance is provided.

Description

Technical Field

[0001] The invention relates to a measuring device and a measuring method in the field of automobile shape detection, and more specifically, it is a automobile shape visual detection system and method based on angle similarity invariance. Background Art

[0002] In recent years, machine vision has gradually become a key technology to improve the efficiency and accuracy of automobile inspection. And automobile shape detection is one of the important research contents of automobile passability and driving safety detection. Automobile shape detection based on machine vision can realize automatic identification of whether the vehicle is overloaded or not, and provide a very important basis for the classification of vehicle models, the acquisition and reconstruction of vehicle size parameters, and vehicle information. Therefore, it is of great research significance to study the automobile shape visual detection system and method based on angle similarity invariance, and to enrich automobile performance detection methods, protect people's life and property safety, and realize automobile automatic detection through non-contact measurement methods. The present invention designs a non-contact, fast-measuring, non-strictly constrained detection method suitable for measuring large objects, constructs spatial transformation feature parallel lines and vertical lines, and uses the similar space angle invariance characteristics to construct angle invariants containing laser projection point information, thereby realizing rapid detection of automobile shape using a simple model. Summary of the invention

[0003] The present invention aims to solve the current situation that a non-contact, fast-measuring, non-strictly-constrained detection method suitable for measuring large objects is needed to restore the automobile shape in the process of obtaining the automobile shape, and proposes a method and system with reliable performance, simple structure, easy operation and simple algorithm. The method mainly consists of a camera bracket, a camera, a connector, a line laser and a two-dimensional reference plate. First, multiple pairs of parallel straight lines and orthogonal straight lines on the reference pattern are used to calculate the transformation matrix from the projection space to the similarity space; then the angle similarity invariant composed of the laser projection point and the known line segment on the two-dimensional reference plate is used; finally, the laser projection point in the similarity space is converted into the Euclidean space according to the sine theorem of the triangle, realizing the visual detection of the automobile shape based on the angle similarity invariant.

[0004] In conjunction with the accompanying drawings of the specification, the present invention is implemented using the following technical solutions:

[0005] The automobile shape visual inspection system based on angle similarity invariance includes a camera bracket, a camera, a connecting piece, a line laser and a two-dimensional reference plate;

[0006] The camera bracket is placed on the ground, the camera is fixedly connected to the bolt thread on the top of the camera bracket through the threaded hole at the bottom, the two-dimensional reference plate is placed in the slender groove of the U-shaped steel plate of the connector, the bolts are inserted into the through holes of the connector and the two-dimensional reference plate and are fixedly connected with the nuts thread, the line laser is inserted into the inner hole of the steel pipe of the laser connector, the set screw is screwed into the threaded hole on the side of the steel pipe of the connector, the end of the set screw is in contact with the cylindrical surface of the line laser, and the laser line emitted by the line laser is coplanar with the plane where the two-dimensional reference plate pattern is located;

[0007] The camera bracket described in the technical solution is a tripod bracket with adjustable height;

[0008] The camera described in the technical solution is a wide-angle industrial camera equipped with a narrow-band filter;

[0009] The connecting piece described in the technical solution is a part welded by a U-shaped steel plate and a steel pipe, the side of the U-shaped steel plate is processed with a through hole, and the side of the steel pipe is processed with a threaded hole;

[0010] The line laser described in the technical solution is a cylindrical part that can emit line laser, and the wavelength of the laser emitted by the line laser is consistent with the bandpass wavelength of the narrow-band filter of the camera;

[0011] The two-dimensional reference plate described in the technical solution is a part made of a rectangular steel plate. Regular geometric figures of intersecting straight line families and parallel straight line families that can generate angle features are pasted on the surface of the two-dimensional reference plate, and the two-dimensional reference plate is processed with through holes.

[0012] The specific steps of the automobile shape visual inspection method based on angle similarity invariance are as follows:

[0013] Step 1: Image acquisition based on angle-similarity-invariant car shape visual inspection:

[0014] The camera is fixed on the camera bracket, and the camera bracket is placed on the ground. The line laser is inserted into the inner hole of the steel pipe of the connector. The line laser is turned on and rotated so that the laser line it projects is coplanar with the two-dimensional reference plate. The line laser is fixed by adjusting the screws on the steel pipe of the connector. The camera collects an image, which includes the two-dimensional reference plate and the projected laser point where the line laser intersects with the vehicle surface.

[0015] Step 2: Obtain the transformation matrix from the projection space to the similarity space based on multiple pairs of parallel lines and orthogonal lines:

[0016] According to the image captured by the camera, the image coordinates l of the i-th (i=1, 2...m, m≥2) pair of parallel lines are extracted using Hough transform. i，1 , l i，2 , find the image of the i-th point at infinity as

[0017] x i = l i，1 × i，2

[0018] The image of a point at infinity is drawn on the straight line at infinity l ∞ =(0 0 1) T Like

[0019]

[0020] in The image of the infinite line can be obtained by SVD decomposition. Then we can get the transformation matrix from projection space to affine space And use H F Multiply the image point coordinates to transform the image from the projective space to the affine space;

[0021] Extract p (p = 1, 2...n, n ≥ 2) pairs of orthogonal lines l from the image after affine correction. q and m q ,but

[0022]

[0023] in for l q and m q The image coordinates of KK can be obtained by SVD decomposition. T , and then use Cholesky decomposition KK T Get the upper triangular matrix K, let Get the matrix that corrects the image to a similar space

[0024] V=H F H A

[0025] Step 3: The laser projection point where the laser line intersects the body surface is reconstructed in the Euclidean coordinate system:

[0026] Get an arbitrary line segment L from a 2D reference board AB Euclidean coordinates of the two endpoints A and B Extract its image coordinates from the image obtained in the first step And the image coordinates x of the laser projection point on the vehicle surface j =(x j y j 1) T , the Euclidean coordinates of the laser projection point on the vehicle surface to be determined are

[0027] According to the transformation matrix V obtained in the second step, the feature points are corrected from the projection space to the similarity space

[0028] A s =V -1 A 1

[0029] B s =V -1 B 1

[0030]

[0031] In similar space, line segment L AB And from point A s and B s and The coordinates of the straight lines are

[0032] l AB =V -1 L AB

[0033]

[0034]

[0035] straight line With straight line The angle is θ A ,straight line With straight line The angle is θ B ,straight line With straight line The angle is θ X , for the jth group of straight lines, the angles are

[0036]

[0037]

[0038]

[0039] According to the angle invariance of similar space, the coordinates of known points A and B in Euclidean space and the sine theorem of triangle, we can get

[0040]

[0041] Substitute the coordinate elements to get

[0042]

[0043]

[0044] make Substituting into the above formula, we get

[0045]

[0046]

[0047] Get the horizontal coordinate of the Euclidean coordinate of the feature point on the body surface and the vertical coordinate for

[0048]

[0049]

[0050] The beneficial effects of the present invention are:

[0051] (1) The method of the present invention uses the images of parallel lines and perpendicular lines to obtain the transformation matrix from Euclidean space to similarity space, uses known points and laser projection points to construct angle similarity invariants, uses the sine theorem to construct the vehicle body point reconstruction solution equation, and uses points that coincide with the coordinate axes to realize the linearization of the quadratic equation, thereby avoiding solving complex nonlinear equations and realizing the reconstruction of the vehicle body points.

[0052] (2) The present invention only requires the camera to obtain an image including the two-dimensional reference plate 5 and the projected laser point where the laser 4 intersects with the vehicle surface, so as to solve the real coordinates of the vehicle body feature points in the coordinate system of the two-dimensional reference plate 5, and further solve the Euclidean coordinates of the vehicle body feature points.

[0053] (3) The camera 2 used in the system of the present invention is a wide-angle industrial camera equipped with a narrow-band filter, which can obtain a wider field of view, so that the wavelength of the laser emitted by the line laser 4 used is consistent with the bandpass wavelength of the narrow-band filter of the camera 2, which can effectively reduce the interference of the ambient light source on the measurement process and greatly improve the measurement accuracy. The steel pipe of the connecting piece 3 is processed with threaded holes on the top, bottom, left and right sides to adjust and fix the position of the line laser 4, so that the laser line projected by the line laser 4 can be coplanar with the two-dimensional reference plate 5. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is an axonometric diagram of the automobile shape vision inspection system based on angle similarity invariance;

[0055] Figure 2 is an axonometric diagram of a camera bracket 1 in a vehicle shape visual inspection system based on angle similarity invariance;

[0056] Figure 3 It is the axonometric view of camera 2 in the automobile shape visual inspection system based on angle similarity invariance;

[0057] Figure 4 It is an axonometric view of the connector 3 in the automobile shape visual inspection system based on angle similarity invariance;

[0058] Figure 5 is an axonometric view of a two-dimensional reference plate 5 in a vehicle shape visual inspection system based on angle similarity invariance;

[0059] Figure 6 It is an axonometric diagram of a connector 3, a line laser 4, and a two-dimensional reference plate 5 in a vehicle shape visual inspection system based on angle similarity invariance;

[0060] Figure 7 This is a schematic diagram of a method for visually inspecting automobile shape based on angle similarity invariance;

[0061] Figure 8 This is a flowchart for solving the transformation matrix from the projection space to the affine space in the car shape visual inspection method based on angle similarity invariance;

[0062] Fig. 9 This is a flowchart for solving the transformation matrix of the projection space to the similarity space in the automobile shape visual inspection method based on angle similarity invariance;

[0063] Fig.10 This is a flowchart of the reconstruction of vehicle body feature points using angle-invariant vehicle shape visual inspection method based on angle similarity invariant;

[0064] In the figure: 1. Camera bracket, 2. Camera, 3. Connector, 4. Line laser, 5. Two-dimensional reference plate. DETAILED DESCRIPTION

[0065] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0066] See also Figures 1 to 6 The automobile shape visual inspection system based on angle similarity invariance includes a camera bracket 1, a camera 2, a connecting piece 3, a line laser 4 and a two-dimensional reference plate 5.

[0067] The camera bracket 1 is a tripod bracket with adjustable height, which is placed on the ground. The camera 2 is a wide-angle industrial camera equipped with a narrow-band filter. The camera 2 is fixedly connected to the bolt thread on the top of the camera bracket 1 through the threaded hole at the bottom. The two-dimensional reference plate 5 is a part made of a rectangular steel plate. The surface of the two-dimensional reference plate 5 is pasted with regular geometric figures of intersecting straight line families and parallel straight line families that can generate angle features. The two-dimensional reference plate 5 is processed with through holes. The connecting piece 3 is a part welded by a U-shaped steel plate and a steel pipe. The side of the U-shaped steel plate is processed with through holes, and the side of the steel pipe is processed with screws. The two-dimensional reference plate 5 is placed in the slender groove of the U-shaped steel plate of the connecting piece 3, the bolts are inserted into the through holes of the connecting piece 3 and the two-dimensional reference plate 5 and are fixedly connected with the nut thread, the line laser 4 is inserted into the inner hole of the steel pipe of the laser connecting piece 3, the set screw is screwed into the threaded hole on the side of the steel pipe of the connecting piece 3, and the end of the set screw is in contact with the cylindrical surface of the line laser 4 and fits tightly. The line laser 4 is a cylindrical part that can emit line lasers. The laser wavelength emitted by the line laser 4 is consistent with the bandpass wavelength of the narrowband filter of the camera 2. The laser line emitted by the line laser 4 is coplanar with the plane where the two-dimensional reference plate 5 pattern is located;

[0068] See also Figures 7 to 10 ,The vehicle shape visual inspection method based on angle similarity invariance can be divided into the following three steps:

[0069] Step 1: Image acquisition based on angle-similarity-invariant car shape visual inspection:

[0070] The camera 2 is fixed on the camera bracket 1, and the camera bracket 1 is placed on the ground. The line laser 4 is inserted into the inner hole of the steel pipe of the connecting piece 3, and the line laser 4 is turned on and rotated so that the laser line projected by it is coplanar with the two-dimensional reference plate 5. The line laser 4 is fixed by adjusting the screws on the steel pipe of the connecting piece 3. The camera 2 collects an image, and the image includes the two-dimensional reference plate 5 and the projected laser point where the line laser 4 intersects with the vehicle surface;

[0071] Step 2: Obtain the transformation matrix from the projection space to the affine space based on multiple pairs of parallel lines:

[0072] According to the image captured by camera 2, the image coordinates l of the i-th (i=1, 2...m, n≥2) pair of parallel lines are extracted using Hough transform. i，1 , l i，2 , the image of the i-th point at infinity is

[0073] x i = l i，1 × i，2

[0074] The image of a point at infinity is drawn on the straight line at infinity l ∞ =(0 0 1) T Like

[0075]

[0076] in The image of the infinite line can be obtained by SVD decomposition. Then we can get the transformation matrix from projection space to affine space And use H F Multiply the image point coordinates to transform the image from the projective space to the affine space;

[0077] Step 3: Obtain the transformation matrix from the projection space to the similarity space based on the transformation matrix from the projection space to the affine space and multiple pairs of orthogonal lines:

[0078] Extract p (p = 1, 2...n, n ≥ 2) pairs of orthogonal lines l from the image after affine correction. q and m q ,but Can be expanded to

[0079]

[0080] in for l q and m q The image coordinates of is the image coordinate of the quadratic curve dual to the imaginary circle point, and KK can be obtained by SVD decomposition T , and then use Cholesky decomposition KK T Get the upper triangular matrix K, let

[0081] According to the second step, we get the transformation matrix H that corrects the image to the similar space. F

[0082] V=H F H A

[0083] Step 4: The laser projection point where the laser line intersects the body surface is reconstructed in the Euclidean coordinate system:

[0084] Get any line segment L from the two-dimensional reference plate 5 AB Euclidean coordinates of the two endpoints A and B Extract its image coordinates from the image obtained in the first step And the image coordinates x of the laser projection point on the vehicle surface j =(x j y j 1) T, the Euclidean coordinates of the laser projection point on the vehicle surface to be determined are

[0085] According to the transformation matrix V obtained in the second step, the feature points are corrected from the projection space to the similarity space

[0086] A s =V -1 A 1

[0087] B s =V -1 B 1

[0088]

[0089] In similar space, line segment L AB And from point A s and B s and The coordinates of the straight lines are

[0090] l AB =V -1 L AB

[0091]

[0092]

[0093] straight line With straight line The angle is θ A ,straight line With straight line The angle is θ B ,straight line With straight line The angle is θ X , for the jth group of straight lines, the angles are

[0094]

[0095]

[0096]

[0097] According to the angle invariance of similar space, the coordinates of known points A and B in Euclidean space and the sine theorem of triangle, we can get

[0098]

[0099] Substitute the coordinate elements to get

[0100]

[0101]

[0102] make Substituting into the above formula, we get

[0103]

[0104]

[0105] Get the horizontal coordinate of the Euclidean coordinate of the feature point on the body surface and the vertical coordinate for

[0106]

[0107]

Claims

1. A reconstruction method of an automobile shape visual inspection system based on angle similarity invariance, the automobile shape visual inspection system based on angle similarity invariance comprising a camera bracket (1), a camera (2), a connecting piece (3), a line laser (4) and a two-dimensional reference plate (5); The camera bracket (1) is placed on the ground, the camera (2) is threadedly fixedly connected to the bolt on the top of the camera bracket (1) through the threaded hole at the bottom, the two-dimensional reference plate (5) is placed in the slender groove of the U-shaped steel plate of the connecting piece (3), the bolts are inserted into the through holes of the connecting piece (3) and the two-dimensional reference plate (5) and are threadedly fixedly connected with the nuts, the line laser (4) is inserted into the inner hole of the steel pipe of the laser connecting piece (3), the set screw is screwed into the threaded hole on the side of the steel pipe of the connecting piece (3), the end of the set screw is in contact with the cylindrical surface of the line laser (4) and is tightly matched, the laser line emitted by the line laser (4) is coplanar with the plane where the two-dimensional reference plate (5) pattern is located, It is characterized in that The specific steps are as follows: Step 1: Image acquisition based on angle-similarity-invariant car shape visual inspection: The camera (2) is fixed on the camera bracket (1), and the camera bracket (1) is placed on the ground. The line laser (4) is inserted into the inner hole of the steel pipe of the connecting piece (3), the line laser (4) is turned on, and rotated so that the laser line projected by it is coplanar with the two-dimensional reference plate (5), and the line laser (4) is fixed by adjusting the screws on the steel pipe of the connecting piece (3). The camera (2) collects an image, and the image includes the two-dimensional reference plate (5) and the projected laser point where the line laser (4) intersects with the vehicle surface; Step 2: Obtain the transformation matrix from the projection space to the similarity space based on multiple pairs of parallel lines and orthogonal lines: Based on the image captured by the camera (2), the image coordinates l of the i-th (i=1, 2...m, m≥2) pair of parallel straight lines are extracted by using Hough transform. i，1 , l i，2 , the image of the i-th point at infinity is x i ＝l i，1 ×l i，2 The image of a point at infinity is drawn on the straight line at infinity l ∞ =(0 0 1) T Like in The image of the infinite line can be obtained by SVD decomposition. Then we can get the transformation matrix from projection space to affine space And use H F Multiply the image point coordinates to transform the image from the projective space to the affine space; Extract p (p = 1, 2...n, n ≥ 2) pairs of orthogonal lines l from the image after affine correction. q and m q ,but in for l q and m q The image coordinates of KK can be obtained by SVD decomposition. T , and then use Cholesky decomposition KK T Get the upper triangular matrix K, let Get the matrix that corrects the image to a similar space V=H F H A Step 3: Reconstruction of laser projection points where the laser line intersects the body surface: Get any line segment L from the two-dimensional reference plate (5) AB Euclidean coordinates of the two endpoints A and B Extract its image coordinates from the image obtained in the first step And the image coordinates x of the laser projection point on the vehicle surface j =(x j y j 1) T , the Euclidean coordinates of the laser projection point on the vehicle surface to be determined are According to the transformation matrix V obtained in the second step, the feature points are corrected from the projection space to the similarity space And S =V -1 And I B S =V -1 B I In similar space, line segment L AB And from point A S and B S and The coordinates of the straight lines are l AB =V -1 L AB straight line With straight line The angle is θ A ,straight line With straight line The angle is θ B ,straight line With straight line The angle is θ x , for the jth group of straight lines, the angles are According to the angle invariance of similar space, the coordinates of known points A and B in Euclidean space and the sine theorem of triangle, we can get Substitute the coordinate elements to get make Substituting into the above formula, we get Get the horizontal coordinate of the European coordinate of the feature point on the body surface and the vertical coordinate for 2. The reconstruction method of the automobile shape visual inspection system based on angle similarity invariance according to claim 1, Features The camera bracket (1) is a tripod bracket with adjustable height.

3. The reconstruction method of the automobile shape visual inspection system based on angle similarity invariance according to claim 1, Features The camera (2) is a wide-angle industrial camera equipped with a narrow-band filter.

4. The reconstruction method of the automobile shape visual inspection system based on angle similarity invariance according to claim 1, Features The connecting piece (3) is a part formed by welding a U-shaped steel plate and a steel pipe. A through hole is processed on the side of the U-shaped steel plate, and a threaded hole is processed on the side of the steel pipe.

5. The reconstruction method of the automobile shape visual inspection system based on angle similarity invariance according to claim 1, Features The line laser (4) is a cylindrical part capable of emitting line laser light. The wavelength of the laser light emitted by the line laser (4) is consistent with the bandpass wavelength of the narrow-band filter of the camera (2).

6. The reconstruction method of the automobile shape visual inspection system based on angle similarity invariance according to claim 1, Features The two-dimensional reference plate (5) is a part made of a rectangular steel plate. Regular geometric figures of intersecting straight line families and parallel straight line families that can generate angle features are pasted on the surface of the two-dimensional reference plate (5). The two-dimensional reference plate (5) is processed with through holes.

Images