A camera perception range generation method for a target detection task

Abstract

The application relates to a camera perception range generation method for a target detection task, which establishes a pinhole imaging camera model containing camera distortion, a target model and a target field of view environment plane model, and is divided into a standard version and a simple version according to application needs. In the standard version, a gridded target is projected into an image to generate a target bounding box, a target detection minimum bounding box threshold is set, and the perception range of the camera is drawn based on the threshold; in the simple version, the camera model is simplified, and the perception area is generated according to the effective detection range expression of the simple model. The camera effective perception range generation method designed by the application can quantize the effective pixel threshold value of the target detection algorithm into the perception range under the camera field of view plane, select the standard and simple models according to application needs, draw the perception area of the camera according to the camera internal and external parameters, improve the camera installation and arrangement efficiency, and effectively guide personnel to select and install the camera according to the target detection demand.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image analysis, and particularly relates to a camera perception range generation method for a target detection task. BACKGROUND

[0002] Cameras have been widely applied in the fields of artificial intelligence and robots. Through a lens, a perspective view is imaged, and the specific feature is that the farther the target distance is, the smaller the imaging is, and the closer the distance is, the larger the imaging is. At this time, the far target is imaged small in the camera field of view, and may only have a few hundred or tens of pixels. Although the target can be identified by the human eye, it poses a great challenge to the target detection algorithm. In the process of installing a camera on a robot, the camera is often installed through naked eye observation or experience judgment, and there are two problems: on the one hand, this method requires more experience, and the installation test is more tedious; on the other hand, since the target detection ability of the image target detection algorithm is not considered, the effective perception ability of the camera cannot be clearly mastered. The patent document with the patent application publication number CN114742897A discloses a processing method, device and equipment for camera installation information of a road test perception system, comprising: acquiring the maximum effective perception position of the fisheye camera according to the internal parameters of the fisheye camera and the preset installation information, and combining the perception overlap distance of the long-short focus gun machine to calculate the distance corresponding to the lower edge of the short-focus and long-focus gun machine imaging image, so as to guide the installation of the long-short focus gun machine.

[0003] In view of the problem that the effective area of the camera perception for the target detection task cannot be quantified, a camera perception range generation method for the target detection task is proposed, which can quickly acquire the effective perception area of the camera according to the internal and external parameters of the camera, and can quantize the effective pixel value threshold of the target detection algorithm into the perception range under the camera field plane, thereby improving the camera selection and installation efficiency. SUMMARY

[0004] The application aims to overcome the shortcomings of the above-mentioned technology, and provides a camera perception range generation method for a target detection task, which can calculate the effective perception area of the camera according to the internal and external parameters of the camera without manually installing the camera, can quantize the effective pixel value threshold of the target detection algorithm into the perception range under the camera field plane, and form an effective detection range expression of a simple model.

[0005] To achieve the above-mentioned purpose, the following technical solution is adopted: a camera perception range generation method for a target detection task, and the specific steps are as follows:

[0006] Step 1: establishing a target model and a target field environment plane model, and constructing a pinhole imaging standard version camera model and a simple version camera model containing camera distortion;

[0007] Step 2 rasterize the target model, convert it into pixel coordinates using the standard camera model, generate the target bounding box, set the target detection minimum bounding box threshold, and determine the effective detection position based on this threshold;

[0008] Step 3 rasterize the target field of view environment plane model, traverse the discrete area of the plane, and generate the standard camera effective perception area;

[0009] Step 4 approximate estimate the key boundary and position coordinate expression under the simple version model, and generate the simple version camera effective perception area.

[0010] Further, the establishment of the target model and the target field of view environment plane model in step 1, the construction of the pinhole imaging standard camera model with camera distortion, specifically:

[0011] ① Establish the camera model:

[0012] The camera rotation matrix R, translation vector t and internal parameter matrix are respectively:

[0013] t = (0 H c 0) T ,

[0014] Where f x = f / rw, f y = f / rh; (pw, ph) is the image resolution, (rw, rh) is the size of the pixel width and length, the lens focal length is f, the camera installation height is H c , the pitch angle is α x , and the camera distortion parameter is kc;

[0015] ② Establish the target model; object = (x w ,y w ,z w ,l, w, h), where (x w ,y w ,z w ) represents the position of the right front point of the target in the ground coordinate system, and (l, w, h) represents the length, width and height of the target;

[0016] ③ Establish the target field of view environment plane model: area = (A 11 ,A 12 ,A 21 ,A 22 ), take the camera position as the origin, and establish the perception area:

[0017]

[0018] Where W, H represent the length and width of the perception area;

[0019] The target model and the target field of view environment plane model satisfy:

[0020] h = 2z.

[0021] Further, the step 1 establishes a simple camera model; specifically:

[0022] ①The camera model is simplified as a pinhole imaging model ignoring distortion factors, and the pitch angle a x = π / 2, at this time the camera axis is parallel to the ground; for the target model, only the cuboid model is considered, and the 8 vertices of the target are respectively denoted as {(x ck ,y ck ,z ck )|1≤k≤8};

[0023] ②The horizontal and vertical boundaries of the camera need to meet

[0024]

[0025] Therefore, the horizontal field of view angle fov_w and the vertical field of view angle fov_h of the camera satisfy the following relationship:

[0026]

[0027] Where f x = f / rw, f y = f / rh.

[0028] Further, the step 2 rasterizes the target model, converts it into pixel coordinates by using the standard camera model, generates a target bounding box, sets a target detection minimum bounding box threshold, and determines the effective detection position based on the threshold, specifically:

[0029] ①The target model object w = (x w ,y w ,z w ,l,w,h) is rasterized to generate the raster point set of the target At this time, the number of raster points is N 3 ;

[0030] ②The target raster points in the ground coordinate system are converted to the camera coordinate system, that is,

[0031] object c = R·object w +t

[0032] The target raster points in the camera coordinate system are transformed into normalized image coordinates, and the target normalized supplementary distortion coordinates are generated based on the distortion model:

[0033]

[0034]

[0035] where, r 2 = x 2 +y 2 , represents the distance from the image point to the camera optical axis, dx represents the tangential distortion parameter, kc = [kc(1) kc(2) kc(3) kc(4) kc(5)] T represents the radial and tangential multi-order distortion coefficients;

[0036] Convert the target normalized supplementary distortion coordinates into pixel coordinates:

[0037]

[0038] At this time, the projection point set of the target grid point on the image is generated object p = {(x ip ,y ip )|1≤i≤N 3};

[0039] ③The effective point set of the target projection point is {(x ip ,y ip )|0≤x ip ≤pw-1,0≤y ip ≤ph-1,1≤i≤N3}, the number of set elements is M, at this time, the size of the target bounding box is object w ,object h , that is

[0040]

[0041] ④Set the size of the target box on the image to be not less than Num p pixels, and the target is completely in the camera field of view to be an effective detection target, when object

[0042] object w ×object h ≥Num p , N 3 =M

[0043] (x w ,y w ) is considered to be an effective detection position in the target field of view environment plane, otherwise it is not an effective detection position.

[0044] Further, the target field of view environment plane model is rasterized in step 3, the discrete area of the plane is traversed, and the standard version camera effective sensing area is generated, specifically:

[0045] ①Discretize the target field of view environment into N W ×N H areas, i.e.,

[0046]

[0047] ②Traverse the discrete areas of the plane, place the target, and repeat step 2 to generate the effective perception area.

[0048] Further, the approximate estimation of the key boundary and position coordinate expression in step 4 is a simple version of the model, which generates a simple version of the camera effective perception area, specifically:

[0049] ①For the critical position of the near blind area, the boundary here is mainly to ensure that the target is within the camera field of view range and is not truncated, so for the target, the right front point of the target is located at the lower left corner of the camera, i.e. the right front point of the target satisfies the horizontal and vertical field of view boundary conditions, which satisfies:

[0050]

[0051] Simplify:

[0052]

[0053] At this time, the shortest distance The nearest field of view width is

[0054] ②For the critical position of the next near, the right lower edge of the target can be determined according to the projection position of the left front point of the target, and the left upper edge of the target can be determined according to the projection position of the right rear point of the target:

[0055]

[0056]

[0057] In order to ensure that the target pixel quantity is not less than Num p , i.e. satisfies

[0058] S = w max · h max = Num p

[0059] After simplifying:

[0060] z c2 4 + 2wz c2 3 + (w 2 -A2lh)z c2 2 -A2(wlHc +x c2 wh)z c2 -A2x c2 H c w 2 =0

[0061]

[0062] For convenience of calculation, let x2=0, and simplify:

[0063] z c2 3 +2wz c2 2 +(w 2 -A2lh)z c2 -A2wlH c =0

[0064] At this time, the second farthest distance

[0065] ③For the farthest critical position, the target right front point here meets the horizontal field of view boundary condition, and the image target lower edge can be determined according to the projection position of the target left front point, and the image target upper edge is determined according to the projection position of the target left rear point, that is, it meets:

[0066]

[0067]

[0068]

[0069] In order to ensure that the target pixel quantity is not lower than Num p , that is, it meets

[0070] S=w max ·h max =Num p

[0071] After simplification:

[0072] y1 3 +2wy1 2 +(w 2 -Ah)y1-AwH c =0

[0073]

[0074] At this time, the farthest distance The nearest field of view width is w3=2|x1|

[0075] ④For positions 2, 4, and 6, the analysis and calculation methods are similar to positions 1, 3, and 5, respectively.

[0076] Beneficial effects: the application can quickly draw the effective sensing area of the camera according to the internal and external parameters of the camera, improve the efficiency of camera installation and arrangement, and guide personnel to select and install the camera according to the target detection demand. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 is a simulation flowchart of the effective monitoring and detection area of the camera;

[0078] Figure 2 is a target vertex number diagram;

[0079] Figure 3 is a simulation diagram of the camera sensing field of view (without distortion);

[0080] Figure 4 is a simulation diagram of the camera sensing field of view (with distortion);

[0081] Figure 5 is a key point calculation schematic diagram of the effective sensing area based on the pinhole model;

[0082] Figure 6 is a comparison diagram of the relationship and expression derivation of the camera focal length parameter and l1, l2, and l3;

[0083] Figure 7 is a comparison diagram of the relationship and expression derivation of the camera focal length parameter and the horizontal field of view angle fov_w;

[0084] Figure 8 is a comparison diagram of the relationship and expression derivation of the camera focal length and l1, l2 at the focal length of 27.22-33.78 mm;

[0085] Figure 9 is a comparison diagram of the relationship and expression derivation of the camera focal length and l3 at the focal length of 27.22-33.78 mm;

[0086] Figure 10 is a comparison diagram of the relationship and expression derivation of the camera height parameter and l1, l2, and l3;

[0087] Figure 11 is a comparison diagram of the relationship and expression derivation of the camera height parameter and the horizontal field of view angle fov_w;

[0088] Figure 12 is a comparison diagram of the relationship and expression derivation of the camera pixel size parameter and l1, l2, and l3;

[0089] Figure 13 is a comparison diagram of the relationship and expression derivation of the camera pixel size parameter and the horizontal field of view angle fov_w;

[0090] Figure 14is a comparison chart of camera resolution parameter and l1, l2, l3 relationship and expression derivation;

[0091] Figure 15 is a comparison chart of camera resolution parameter and horizontal field of view angle fov_w relationship and expression derivation;

[0092] Figure 16 is a chart of camera pitch angle change and l1, l2, l3 relationship;

[0093] Figure 17 is a chart of camera pitch angle change and field of view angle fov_w relationship. DETAILED DESCRIPTION

[0094] In order to enable the above-mentioned objects, features and advantages of the present application to be more clearly understood, the following will describe the present application in detail with reference to the drawings and specific embodiments. It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other without conflict. In the following description, a large number of specific details are set forth in order to facilitate a full understanding of the present application. The described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application.

[0095] In the embodiments of the present application, in order to facilitate the description without limiting the present application, the term "connection" used in the patent application specification and claims of the present application is not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "below", "left", "right", etc. are only used to indicate relative positional relationship, which changes accordingly when the absolute position of the described object changes.

[0096] As shown in the drawings, the embodiment provides a camera perception range generation method for a target detection task, which includes a camera distortion pinhole imaging camera model, a target model and a target field of view environment plane model, and is divided into a standard version and a simple version according to application needs. In the standard version, the rasterized target is projected into the image to generate a target bounding box, a target detection minimum bounding box threshold is set, and the perception range of the camera is drawn based on the threshold. In the simple version, the camera model is simplified, and the perception area is generated according to the effective detection range expression of the simple model.

[0097] The specific steps are as follows:

[0098] Step 1): Establish target model and target field of view environment plane model, construct small aperture imaging standard version camera model and simple version camera model containing camera distortion, specifically:

[0099] S21: Establish a small aperture imaging standard version camera model containing camera distortion, (pw, ph) is the image resolution, (rw, rh) is the size of the pixel width and length, the lens focal length is f, the camera installation height is H c , the pitch angle is α x , and the camera distortion parameter is kc;

[0100] Therefore, the camera rotation matrix R, the translation vector t and the internal parameter matrix are respectively:

[0101] t = (0 H c 0) T ,

[0102] Where f x = f / rw, f y = f / rh;

[0103] ② Establish the target model; object = (x w ,y w ,z w ,l,w,h), where (x w ,y w ,z w ) represents the position of the target right front point in the ground coordinate system, and (l,w,h) represents the length, width and height of the target.

[0104] S23: Establish a target field of view environment plane model: area = (A 11 ,A 12 ,A 21 ,A 22 ), establish a perception area with the camera position as the origin:

[0105]

[0106] Where W, H represent the length and width of the perception area

[0107] S24: It should be noted that since the target is on the ground, the above target model and target field of view environment plane model satisfy:

[0108] h = 2z

[0109] S25: If the camera axis is approximately parallel to the ground in the application scenario, the lens distortion is small, the simple version can be used.

[0110] S26: Simplify the camera model to a pinhole imaging model which ignores distortion factors, and the pitch angle a x = π / 2, at this time the camera axis is parallel to the ground; for the target model, only consider the cuboid model, see the attached Figure 2 , the 8 vertices of the target are respectively denoted as {(x ck ,y ck ,z ck )|1≤k≤8};

[0111] S27: The horizontal and vertical boundaries of the camera need to meet

[0112]

[0113] Therefore, the horizontal and vertical field angles of the camera fov_w and fov_h satisfy the following relationship:

[0114]

[0115] Where f x = f / rw, f y = f / rh.

[0116] Step 2): Rasterize the target model, convert it to pixel coordinates using the standard camera model, generate the target bounding box, set the target detection minimum bounding box threshold, and determine the effective detection position based on this threshold, which is:

[0117] S31: Rasterize the target model object w = (x w ,y w ,z w ,l, w, h) to generate the target grid point set At this time, the number of rasterized points is N 3 ;

[0118] S32: Convert the target grid points in the ground coordinate system to the camera coordinate system, that is

[0119] object c = R·object w +t

[0120] Transform the target grid points in the camera coordinate system to normalized image coordinates, and generate target normalized supplementary distortion coordinates based on the distortion model:

[0121]

[0122]

[0123] Where r 2 = x 2 +y2 , where d represents the distance from the image point to the camera optical axis, dx represents the tangential distortion parameter, and kc = [kc(1) kc(2) kc(3) kc(4) kc(5)] T , where d represents the distance from the image point to the camera optical axis, dx represents the tangential distortion parameter, and kc = [kc(1) kc(2) kc(3) kc(4) kc(5)]

[0124] Finally, the target normalized supplementary distortion coordinate is converted into a pixel coordinate:

[0125]

[0126] At this time, the projection point set object of the target grid point on the image is generated p = {(x ip ,y ip )|1≤i≤N 3}.

[0127] S33: The valid point set of the target projection point is {(x ip ,y ip )|0≤x ip ≤pw-1,0≤y ip ≤ph-1,1≤i≤N 3}, and the number of set elements is denoted as M. At this time, the size of the target bounding box is object w , object h , that is

[0128]

[0129] S34: The size of the target box on the image is set to be not less than Num p pixels, and the target is completely in the camera field of view to be a valid detection target. When object

[0130] object w × object h ≥ Num p , N 3 = M

[0131] It is considered that the (x w ,y w ) position in the target field of view environment plane is a valid detection position, otherwise it is a non-valid detection position.

[0132] Step 3): The target field of view environment plane model is rasterized, and the standard version of the camera effective sensing area is generated by traversing the discrete area of the plane;

[0133] For details, see the attached Figure 3 and the attached Figure 4 , specifically:

[0134] S41: rasterize the target field of view environment plane, discretize the continuous plane into N W ×N H regions, that is

[0135]

[0136] S42: traverse the plane discrete region, place the target, repeat step 3, generate an effective perception region;

[0137] Step 4): approximate estimation of key boundary and position coordinate expression under simple version model, generate simple version camera effective perception region;

[0138] See the attached Figure 5 , specifically:

[0139] S51: for the critical position of the near blind area, that is, position 1, the boundary here is mainly to ensure that the target is within the camera field of view range and is not truncated, so for the target, point 1 is located at the lower left corner of the camera, that is, point 1 satisfies the horizontal and vertical field of view boundary conditions, which satisfy:

[0140]

[0141] Simplify to get:

[0142]

[0143] At this time, the shortest distance The nearest field of view width is

[0144] S52: for the critical position of the next far distance, that is, position 5, the right lower edge of the image target can be determined according to the projection position of point 2, and the left upper edge of the image target can be determined according to the projection position of point 8:

[0145]

[0146]

[0147] In order to ensure that the target pixel amount is not less than Num p , that is, it satisfies

[0148] S = w max ·h max = Num p

[0149] After simplifying:

[0150] z c2 4 +2wz c2 3 +(w 2 -A2lh)zc2 2 -A2(wlH c +x c2 wh)z c2 -A2x c2 H c w 2 =0

[0151]

[0152] For convenience of calculation, let x2=0, and simplify:

[0153] z c2 3 +2wz c2 2 +(w 2 -A2lh)z c2 -A2wlH c =0

[0154] At this time, the next far distance

[0155] S53: For the farthest critical position, that is, position 3, the 1st point here satisfies the horizontal field of view boundary condition, and the image target lower edge can be determined according to the projection position of the 2nd point, and the image target upper edge can be determined according to the projection position of the 7th point, that is, it satisfies:

[0156]

[0157]

[0158]

[0159] In order to ensure that the target pixel quantity is not lower than Num p , that is, it satisfies

[0160] S = w max ·h max = Num p

[0161] After simplification:

[0162] y1 3 +2wy1 2 +(w 2 -Ah)y1-AwH c =0

[0163]

[0164] At this time, the farthest distance The nearest field of view width is w3=2|x1|

[0165] S54: For positions 2, 4, and 6, the analysis and calculation methods are similar to those for positions 1, 3, and 5, respectively.

[0166] Embodiments

[0167] A camera perception range generation method for a target detection task is designed. First, a pinhole imaging camera model with camera distortion, a target model, and a target field of view environment plane model are established, and are divided into a standard version and a simple version according to application needs. In the standard version, a gridded target is projected into an image to generate a target bounding box, a target detection minimum bounding box threshold is set, and the perception range of the camera is drawn based on the threshold. In the simple version, the camera model is simplified, and the perception area is generated according to the effective detection range expression of the simple model. The simulation results of the simple version and the standard version are compared to verify the effectiveness of the method. The simulation environment is built on Ubuntu using python, and experiments are carried out.

[0168] I. Model parameters and experimental scenarios

[0169] (1) The simulation analysis parameters are set as follows:

[0170]

[0171] (2) Experimental scenario settings:

[0172] Simulation experiment

[0173] II. Experimental results

[0174] The internal and external parameters of the camera that affect the camera perception range are analyzed, specifically:

[0175] 1) Camera field of view

[0176] According to the parameter settings of (I), the target effective perception range based on the pinhole model is shown in Figure 2 , the farthest perception distance is 57.58 m, and the nearest perception distance is 7.67 m; the target effective perception range of the pinhole model with lens distortion is shown in Figure 3 , the farthest perception distance is 50 m, and the nearest perception distance is 7.76 m. Due to the presence of distortion parameters, the edge of the effective detection area is curved, and the effective detection area is smaller.

[0177] 2) Adjust the focal length of the camera

[0178] According to the parameter settings of (I), set different focal lengths (1-50 mm) of the camera, and observe the influence of focal length changes on l1, l2, l3, and fov_w, as shown in Figure 6 and Figure 7As shown in FIG. 5 and FIG. 6, where the triangular points represent the value of l1, the square points represent the value of l2, the round points represent the value of l3, and the diamond points represent the value of fov_w, it can be seen that as the focal length increases, the effective perception distance increases, the perception range decreases, and the effective field of view angle decreases.

[0179] Figure 5 and Figure 6 As shown in FIG. 5 and FIG. 6, where the triangular points represent the value of l1, the square points represent the value of l2, the round points represent the value of l3, and the diamond points represent the value of fov_w, it can be seen that as the focal length increases, the effective perception distance increases, the perception range decreases, and the effective field of view angle decreases. Figure 6 As shown in FIG. 5 and FIG. 6, where the triangular points represent the value of l1, the square points represent the value of l2, the round points represent the value of l3, and the diamond points represent the value of fov_w, it can be seen that as the focal length increases, the effective perception distance increases, the perception range decreases, and the effective field of view angle decreases. Figure 8 Figure 9 As shown in FIG. 5 and FIG. 6, where the triangular points represent the value of l1, the square points represent the value of l2, the round points represent the value of l3, and the diamond points represent the value of fov_w, it can be seen that as the focal length increases, the effective perception distance increases, the perception range decreases, and the effective field of view angle decreases.

[0180] 3) Adjust the camera installation height,

[0181] According to the parameter setting of (1), set different installation heights (0.5-16m) of the camera, and observe the influence of the installation height change on l1, l2, l3 and fov_w, as shown in FIG. 7 and FIG. 8, where the triangular points represent the value of l1, the square points represent the value of l2, the round points represent the value of l3, and the diamond points represent the value of fov_w, it can be seen that as the height increases, the effective perception distance slowly increases, the perception range decreases, and the effective field of view angle almost remains unchanged. Figure 10 Figure 11

[0182] 4) Adjust the camera pixel size

[0183] According to the parameter setting of (1), set different pixel sizes (0.001-0.008mm) of the camera, and observe the influence of the pixel size change on l1, l2, l3 and fov_w, as shown in FIG. 9 and FIG. 10, where the triangular points represent the value of l1, the square points represent the value of l2, the round points represent the value of l3, and the diamond points represent the value of fov_w, it can be seen that as the pixel size increases, the effective perception distance rapidly decreases, the perception range decreases, and the effective field of view angle increases. Figure 12 Figure 13

[0184] 5) Adjust the camera resolution

[0185] According to the parameter setting of (1), set different resolutions of the camera, and observe the changes of l1, l2, l3 and fov_w, as shown in FIG. 11 and FIG. 12, where the triangular points represent the value of l1, the square points represent the value of l2, and the round points represent the value of l3, it can be seen that as the resolution increases, the effective perception distance slowly increases, the perception range slowly increases, and the effective field of view angle increases. Figure 14 Figure 15

[0186] 6) Adjust the camera pitch angle​​​​​​​

[0187] According to the parameter setting of (one), different camera pitch angles (0-0.8 rad) are set, and the effects of different pitch angles on l1, l2, l3 and fov_w are observed, as shown in Figure 16 and Figure 17 , where the triangular points represent l1 values, the square points represent l2, the circular points represent l3, and the diamond points represent fov_w. It can be seen that as the pitch angle increases, the camera gradually faces the ground, the effective field of view angle gradually decreases, and the effective perception distance first increases and then decreases after exceeding fov_h2.

[0188] The root mean square error statistics of the expression calculation and the simulation results are shown in the following table. It can be seen that the simple model derived by the present patent can better match the simulation results, and the small amount of error is mainly caused by the size of the target object. Therefore, the effective detection area of the camera can be quickly obtained by using the above expression.

[0189] Root mean square error statistics table of expression calculation and simulation results

[0190]

[0191] The pitch angle expression around x is relatively complex, but the effective perception range of the camera can also be calculated according to the design idea of the present patent.

[0192] The above reference examples are illustrative rather than limiting, and a number of embodiments can be listed according to the defined range, so changes and modifications within the overall concept of the present application shall be within the scope of protection of the present application.

Claims

1. A method for generating a camera perception range for a target detection task, characterized in that: Specific steps As follows: Step 1: Establish a target model and a target field of view environment plane model, and construct a pinhole imaging standard version camera model and a simple version camera model containing camera distortion; the simple version camera model is specifically: x = π / 2, at this time the camera axis is parallel to the ground; for the target model, only consider the cuboid model, the 8 vertices of the target are respectively denoted as {(x ck ,y ck ,z ck )|1≤k≤8};​ ②The horizontal and vertical boundaries of the camera need to meet The horizontal field of view angle fov_w and the vertical field of view angle fov_h of the camera meet the following relationship: where f x = f / rw, f y = f / rh, where (pw, ph) is the image resolution, (rw, rh) is the pixel width and length size, and the lens focal length is f. Step 2: Grid the target model, convert it to pixel coordinates using the standard version camera model, generate a target bounding box, set a target detection minimum bounding box threshold, and determine the effective detection position based on this threshold; Step 3: Grid the target field of view environment plane model, traverse the discrete area of the plane, and generate a standard version camera effective sensing area; Step 4: Approximate estimate the key boundary and position coordinate expression under the simple version model, and generate a simple version camera effective sensing area; the approximate estimate of the key boundary and position coordinate expression under the simple version model, and the generation of the simple version camera effective sensing area, are specifically: ①For the critical position of the near blind area, i.e. position No. 1, the boundary here is mainly to ensure that the target is within the camera field of view range and is not truncated, so for the target, the right front point of the target is located at the lower left corner of the camera, i.e. the right front point of the target meets the horizontal and vertical field of view boundary conditions, which meet: Simplify to get: At this time, the shortest distance The recent field of view width is w is the width of the target, and the camera installation height is H c ; ②For the critical position of the next far distance, i.e. position No. 5, the right lower edge of the image target can be determined according to the projection position of the left front point of the target, and the left upper edge of the image target can be determined according to the projection position of the right rear point of the target: To ensure that the target pixel quantity is not less than Num p i.e. to meet S = w max ·h max = Num p After simplification: z c2 4 +2wz c2 3 +(w 2 -A2lh)z c2 2 -A2(wlH c +x c2 wh)z c2 -A2x c2 H c w 2 =0 where: (l, w, h) represents the length, width and height of the target; For convenience of calculation, let x c2 = 0, after simplification: z c2 3 +2wz c2 2 +(w 2 -A2lh)z c2 -A2wlH c =0 At this time, the secondary long-range ③For the critical position of the farthest distance, i.e. position No. 3, the right front point of the target meets the horizontal field of view boundary condition, the lower edge of the image target can be determined according to the projection position of the left front point of the target, and the upper edge of the image target can be determined according to the projection position of the left rear point of the target, i.e. meet: To ensure that the target pixel quantity is not less than Num p i.e. satisfies S = w max ·h max = Num p After simplification: y1 3 +2wy1 2 +(w 2 -Ah)y1-AwH c =0 At this time, the farthest distance The recent field of view width is w3=2|x1| ④For the key boundary points 2, 4, and 6 that constitute the other side of the effective sensing area, the analysis and calculation methods are similar to those of positions 1, 3, and 5, respectively, where positions 2, 4, and 6 correspond to the critical positions of the other side of the near blind area, the farthest distance, and the next far distance, respectively.

2. The method of claim 1, wherein: Step 1: Establish a target model and a target field of view environment plane model, and construct a pinhole imaging standard version camera model, specifically: ①Establish a camera model: The camera rotation matrix R, the translation vector t, and the internal parameter matrix are respectively: wherein f x = f / rw, f y = f / rh; (pw, ph) is the image resolution, (rw, rh) is the pixel width and length size, the lens focal length is f, the camera installation height is H c , the pitch angle is a x , and the camera distortion parameter is kc; ② Establish target model; object = (x w ,y w ,z w ,l,w,h), wherein (x w ,y w ,z w ) represents the position of the target right front point in the ground coordinate system, and (l,w,h) represents the length, width and height of the target; ③ Establish the target field of view environment plane model: area = (A 11 ,A 12 ,A 21 ,A 22 ), with the camera position as the origin, establish the perception area: Where W, H represent the length and width of the sensing area; The target model and the target field of view environment plane model meet: h = 2z.

3. The method of claim 2, wherein: Step 2: Grid the target model, convert it to pixel coordinates using the standard version camera model, generate a target bounding box, set a target detection minimum bounding box threshold, and determine the effective detection position based on this threshold, specifically: w w w 3 ;​​​​​ ②Convert the target grid points in the ground coordinate system to the camera coordinate system, i.e. object c = R - object + t Transform the target grid points in the camera coordinate system to normalized image coordinates, and generate target normalized supplementary distortion coordinates based on the distortion model: where r 2 = x 2 + y 2 , represents the distance of the image point to the camera optical axis, dx represents the tangential distortion parameter, kc = [kc(1) kc(2) kc(3) kc(4) kc(5)] T represents the radial and tangential multi-order distortion coefficients; Convert the target normalized supplementary distortion coordinates to pixel coordinates: At this time, the projection point set object of the target grid point on the image is generated p = {(x ip ,y ip )|1≤i≤N 3} ③ The effective point set of the target projection point is {(x ip ,y ip )|0≤x ip ≤pw-1,0≤y ip ≤ph-1,1≤i≤N 3}, the number of set elements is M, at this time, the size of the target bounding box is object w ,object h , that is (iv) set the size of the target frame on the image to be no less than Num p pixels, and the target is completely within the field of view of the camera for valid detection of the target, when the following conditions are met object w x object h ≥ num p , n 3 = m The position (x w ,y w ) in the target field of view environment plane is considered as an effective detection position, otherwise it is considered as an ineffective detection position.

4. The method of claim 2, wherein: Step 3: Grid the target field of view environment plane model, traverse the discrete area of the plane, and generate a standard version camera effective sensing area, specifically: ①The target field of view environment is gridded, and the continuous plane is discretized into N W ×N H regions, that is, 2. Traverse the discrete region of the plane, place the target, and repeat step 2 to generate the effective perception region.

Images