Vehicle parameter measurement method and system based on three-dimensional vision

Abstract

The invention discloses a vehicle parameter measurement method and system based on three-dimensional vision. According to the method, image data of a target object is obtained based on a neural network, a robot is automatically controlled to move so that a camera can be located at the optimal measurement position, and vehicle data are fully automatically measured; the target object comprises a vehicle tail part, an oil tank outer cover, an outer cover maximum surface, an outer cover opening, an inner cover bottom surface and an inner cover knob; the vehicle data comprises the distance from the maximum surface center to the tail of the vehicle, the shape and depth data of a target object and the actual size; the system comprises N measurement subsystems, a server, a GPU server, network equipment, a system switch and a system power supply, wherein each measurement subsystem comprises an intelligent analyzer, a switch, input equipment, a camera, a fixed part, a robot, a display and a power supply; complete vehicle parameters are provided for automatic refueling, the method can be widely used for measuring the vehicle parameters and automatic refueling, and good economic benefits and market prospects are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of measuring vehicle parameters and fuel tank caps of fuel vehicles, and specifically relates to a vehicle parameter measurement method and system based on three-dimensional vision. Background Art

[0002] Chinese invention patent CN115205495A proposes a method for identifying a fuel tank cap by using 2D rough positioning and then 3D precise positioning; the fuel tank cap is roughly positioned by calculating the surface normal of the fuel tank cap point cloud, and then projecting the point cloud into a 2D image based on the normal, and then the rough positioning point is used to accurately position the outer cover; the shortcomings of the above method are: (1) This method uses image segmentation technology to segment the fuel tank cap, but this processing method has poor consistency in processing images with large noise, and the point cloud image is an image formed by reflection imaging. When the shape of the vehicle fuel tank cap is different, the surface of the fuel tank cap at different angles has large differences in the reflection surface and reflection angle of the point cloud, which may cause incomplete point cloud data, low accuracy and poor consistency of the collected image, thereby affecting the measurement accuracy of the fuel tank cap; (2 ) Although the template matching method is simple, its effect is highly dependent on the quality and quantity of the templates; when the number of templates is small, it may not be possible to match all types of fuel tank caps, especially when dealing with multi-curved fuel tank covers, the matching difficulty and complexity are high, which may make the matching unable to be completed effectively; (3) This method fails to identify the shape and depth data of the surface with the largest area among the multi-curved fuel tank covers; (4) This method does not measure the actual size of the largest surface, the outer cover opening, the inner cover bottom surface and the inner cover knob; (5) This method does not measure the distance between the fuel tank cap and the rear of the vehicle and the height of the fuel tank cap from the ground, which are all key data that the automatic refueling system must have, which may lead to the inability to implement automatic refueling and make it impossible to effectively apply this method in actual refueling.

[0003] In summary, although the existing technology has achieved the recognition and measurement of the fuel tank cap to a certain extent, due to the many problems and shortcomings mentioned above, the method described is difficult to be applied in the automatic refueling system. Therefore, it is necessary to develop a more accurate, stable and adaptable vehicle parameter measurement method and system to meet the actual needs of the automatic refueling system. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a vehicle parameter measurement method and system based on three-dimensional vision, which is used to accurately identify the rear of the vehicle, the fuel tank cover, the maximum surface, the fuel tank cover opening, the bottom surface of the inner cover and the inner cover knob, and automatically measure the vehicle parameters of different vehicles.

[0005] The purpose of the present invention is achieved through the following technical solutions: A vehicle parameter measurement method based on three-dimensional vision, the method is implemented by a vehicle parameter measurement system based on three-dimensional vision, and the implementation steps of the method are as follows: Step A, generating a neural network; the neural network is a neural network that can identify a target object; Step B, measurement processing; The steps of the measurement process are as follows: Keep the vehicle within the measurement range; Acquire image data of the target object and measure the actual distance from the center of the largest surface to the rear of the vehicle; Acquire image data of the target object, measure the shape and depth data of the largest surface and the actual size; Acquire the image data of the target object and measure the fuel tank cap data; Upload the measured vehicle data to the server.

[0006] The steps of placing the vehicle within the measurement range of the present invention are as follows: Step 2.1: If the vehicle meets the first preset condition, go to step 2.3; otherwise go to step 2.2; Step 2.2, issuing an instruction for adjusting the vehicle position; Step 2.3: End the processing of this step.

[0007] The steps of obtaining the image data of the target object are as follows: The intelligent analyzer captures a frame of image and a frame of depth image from the video image output by the camera; the image refers to a color image, wherein the width of the image and the depth image are both W and the height are both H; Compressing the image into a compressed image; the compressed image is an image with a width of w and a height of h; the w and h are the width and height of the image processed by the neural network; Using the neural network to identify each target object from the compressed image, and obtaining the minimum bounding rectangle of each target object in the image based on the minimum bounding rectangle that frames each target object in the compressed image; Using an image processing algorithm, image data of each target object is obtained from the area of ​​the smallest circumscribed rectangle of each target object in the image; the image data of the target object includes the outline of the target object, the area of ​​the target object, the center point of the target object area, and the depth value of each pixel point in the target object area; the targets are target T1, target T2, target T3, target T4, target T5, and target T6; wherein, target T1 is the rear of the vehicle, target T2 is the outer cover, target T3 is the maximum surface, and The target object T4 refers to the outer cover opening, the target object T5 refers to the inner cover bottom surface, and the target object T6 refers to the inner cover knob; the vehicle tail portion refers to the outermost contour portion of the vehicle tail portion; the outer cover refers to the outer cover of the fuel tank and is in a fully closed state; the maximum surface refers to the largest and smoothest surface of the outer cover; the outer cover opening refers to the outer cover opening of the fuel tank and is in a fully open state; the inner cover bottom surface refers to the bottom surface of the inner cover of the fuel tank; the inner cover knob refers to the knob portion on the bottom surface of the inner cover.

[0008] The steps of obtaining the image data of the target object and measuring the actual distance from the center of the largest surface to the rear of the vehicle are as follows: Acquire image data of target T1, target T2 and target T3, so that the camera satisfies the first pose condition; the first pose condition means that the image contains target T1, target T2 and target T3, and the distance from the camera to the vehicle is within a first preset range; the first preset range is the range within which the neural network can identify target T1, target T2 and target T3; The image data of the target T1 and the target T3 are acquired, and the actual distance from the center of the largest surface to the rear of the vehicle is measured.

[0009] The steps of obtaining the image data of the target object and measuring the shape and depth data of the maximum surface and the actual size are as follows: Acquire image data of target T2 and target T3 so that the camera satisfies a second posture condition; the second posture condition means that the image contains target T2 and target T3, and the distance from the camera to the vehicle is within a second preset range; the second preset range is a range within which the neural network can identify target T2 and target T3; The image data of the target object T3 is acquired, and the depth data of the largest surface and the actual size of the largest surface are measured.

[0010] The steps of obtaining the image data of the target object and measuring the fuel tank cap data are as follows: Obtain the image data of the target object T5, and make the camera satisfy the third pose condition; the third pose condition means that the image includes the target object T4, the target object T5, and the target object T6, and the distance from the camera to the vehicle is within a second preset range; the second preset range is the range within which the neural network can recognize the target object T4, the target object T5, and the target object T6. Measure the height from the center point of the bottom surface of the inner cover to the ground. Obtain the image data of the target object T5, and make the camera satisfy the fourth pose condition; the fourth pose condition means the third pose condition and the camera optical axis is perpendicular to the bottom surface of the inner cover. Obtain the image data of the target object T4, measure the depth data of the outer cover opening and measure the actual size of the outer cover opening. Obtain the image data of the target object T6, measure the rotation angle of the inner cover knob and measure the actual size of the inner cover knob. Obtain the image data of the target object T5, measure the depth data of the bottom surface of the inner cover and measure the actual size of the bottom surface of the inner cover.

[0011] The steps of generating the neural network are as follows: Step 7.1: Set the number of captured pictures ImgsN, and set the atlas set and the processing set to be empty sets. Step 7.2: Obtain the atlas; the atlas is an atlas containing the target objects. Step 7.3: Obtain the processing set using the atlas. Step 7.4: Split the processing set into a training set and a test set. Step 7.5: Train the neural network model using the training set. Step 7.6: Test the neural network model using the test set, and count the number of pictures correctly tested. Step 7.7: Calculate the test accuracy rate Av; Av = (the number of pictures correctly tested / the number of pictures in the test set) × 100%. Step 7.8: If Av < SetAv, go to Step 7.9; otherwise, go to Step 7.10; where 80% ≤ SetAv ≤ 100%. Step 7.9: Increase the value of ImgsN, and after increasing, go to Step 7.2; increasing the value of ImgsN means that after increasing, ImgsN = the previous ImgsN × AddScale; where 1.2 ≤ AddScale ≤ 2.0. Step 7.10: Compile the neural network model into a neural network; the neural network is a neural network that can recognize the target objects. Step 7.11: Write the neural network into each intelligent analyzer in the system.

[0012] The steps of obtaining the atlas are as follows: The intelligent analyzer and the camera take pictures in atlas P1, atlas P2, atlas P3, and atlas P4 at different time periods within a day, and the number of pictures in the atlas must satisfy |P1|≥ImgsN∧|P2|≥ImgsN∧|P3|≥ImgsN∧|P4|≥ImgsN; wherein, the atlas P1 is a collection of pictures taken when the camera satisfies the first posture condition; the atlas P2 is a collection of pictures taken when the camera satisfies the second posture condition; the atlas P3 is a collection of pictures taken when the camera satisfies the third posture condition; the atlas P4 is a collection of pictures taken when the camera satisfies the fourth posture condition; The intelligent analyzer uploads all the pictures that have been taken but not uploaded to the server at a preset time, and the server adds each picture received from all the uploaded pictures to the picture collection.

[0013] The steps of obtaining the processing set using the atlas are as follows: Annotation processing: Annotation processing refers to using automatic annotation software to automatically annotate each unannotated picture in the atlas to obtain an annotation file of the picture; Correction processing; Correction processing refers to correcting each incorrect annotation file in the atlas into a correct annotation file by using modification annotation software; Compression processing: Compression processing refers to compressing each uncompressed image in the atlas to obtain a compressed image of the image, and obtaining a labeling file of the compressed image based on the labeling file of the image.

[0014] The present invention provides a vehicle parameter measurement system based on three-dimensional vision, wherein the system is used to implement any one of the vehicle parameter measurement methods based on three-dimensional vision, and the system includes N measurement subsystems, a server, a GPU server, a network device, a system switch and a system power supply; wherein N is a positive integer.

[0015] The measurement subsystem includes an intelligent analyzer, a switch, an input device, a camera, a fixture, a robot, a display and a power supply.

[0016] The measurement subsystem also includes an intercom device, which is used to issue a voice message for adjusting the vehicle position; the power supply is connected to the intercom device and provides the required DC power for the intercom device.

[0017] Compared with the prior art, the significant progress and technical features of the vehicle parameter measurement method and system based on three-dimensional vision in the present invention are as follows: the technical route of the present invention is: A. Generate a neural network to identify the target object, the target object includes the rear of the vehicle, the outer cover of the fuel tank, the maximum surface, the outer cover opening of the fuel tank, the bottom surface of the inner cover and the inner cover knob; B. Use the following steps to measure the vehicle parameters: (1) Place the vehicle within the measurement range; (2) Obtain image data of the target object, automatically control the robot to make the camera meet the first posture condition, and measure the distance from the center of the maximum surface to the rear of the vehicle; (3) Obtain image data of the target object, automatically control the robot to make the camera meet the second posture condition, and measure the shape and depth data of the maximum surface and the actual size; (4) Obtain image data of the target object, automatically control the robot to make the camera meet the third posture condition, and measure the height from the center of the bottom surface of the inner cover to the ground; (5) Obtain image data of the target object, automatically control the robot to make the camera meet the fourth posture condition, and measure the shape and depth data of the outer cover opening, the inner cover knob and the bottom surface of the inner cover The invention has the beneficial effects of automatically placing the camera in the best measuring position, accurately identifying the vehicle tail, the fuel tank cover, the maximum surface, the fuel tank cover opening, the inner cover bottom surface and the inner cover knob, and then measuring the following data of various vehicles: the distance from the center of the maximum surface to the vehicle tail, the height from the center of the maximum surface to the ground, the shape and depth data of the maximum surface and the actual size, the height from the center of the inner cover bottom surface to the ground, the angle between the inner cover bottom surface and the ground, the diameter of the inner cover bottom surface, the rotation angle between the inner cover knob and the ground, the width, height and diameter of the inner cover knob; providing the parking position of the vehicle and automatically opening the fuel tank cover and the inner cover for automatic refueling, providing relatively complete data; the invention can be widely used in occasions of measuring vehicle parameters, fuel tank covers and automatic refueling, and has good economic benefits and broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of a vehicle parameter measurement system based on three-dimensional vision in an embodiment of the present invention; Figure 2 is a schematic diagram of a measurement subsystem in an embodiment of the present invention; Figure 3 A schematic diagram of a process for measuring vehicle parameters in an embodiment of the present invention; Figure 4 A schematic diagram of a structure for measuring vehicle parameters in an embodiment of the present invention; Figure 5 A schematic diagram of a (enlarged) view of a structure for measuring vehicle parameters in an embodiment of the present invention; Figure 6Schematic diagram of the measurement range in an embodiment of the present invention; Figure 7 A schematic diagram of using Euler angles to describe the rotation of the robot's end center point RCP in an embodiment of the present invention; Figure 8 A schematic diagram of measuring the distance from the center of the largest surface to the rear of a vehicle in an embodiment of the present invention; Fig. 9 It is a schematic diagram of measuring the maximum surface shape and depth data and the actual size in an embodiment of the present invention; Fig.10 This is a schematic diagram of measuring the angle between the bottom surface of the inner cover and the ground in an embodiment of the present invention; Fig.11 A schematic diagram of making the optical axis of the camera perpendicular to the bottom surface of the inner cover by rotating and moving the robot in an embodiment of the present invention; Fig.12 It is a schematic diagram of measuring the height of the inner cover knob in an embodiment of the present invention; Fig.13 It is a schematic diagram of measuring the outer cover opening depth data and actual size, the inner cover bottom surface actual diameter, the inner cover knob width and rotation angle in an embodiment of the present invention.

[0019] In the figure: 1. Vehicle; 2. Rear of vehicle; 3. Robot; 4. Refueling island; 5. Ground; 6. Robot base; 7. Robot end axis; 8. Fixing plate; 9. Depth camera; 10. Vehicle fuel tank cover; 11. Cover opening; 12. Screws for fixing the camera to the fixing bar; 13. Fixing bar; 14. Screws for fixing the fixing bar to the fixing plate; 15. Camera optical axis; 16. Maximum surface; 17. Minimum circumscribed rectangle of the maximum surface; 18. Inner cover knob; 19. Inner cover bottom; 20. Minimum circumscribed rectangle of the outer cover opening. DETAILED DESCRIPTION

[0020] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided to make the present invention more comprehensive and complete and to fully convey the concepts of example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar structures, and their repeated description will be omitted.

[0021] The words expressing positions and directions described in the present invention are all explained with reference to the accompanying drawings as examples, but they can be changed as needed, and all such changes are included in the protection scope of the present invention.

[0022] Before introducing the embodiment scheme, the parameters and terms used in the embodiment are first introduced: Set: A set is a set composed of some elements. The number of elements in the set Set is represented by |Set|. b∈Set means that b belongs to the set Set, or that b is an element in the set Set, and ∈ means belongs to; the empty set Set means that there are no elements in the set Set, and |Set|=0.

[0023] Pixel point: Pixel point p(x,y) is a pixel point in the image area PicWH; the value range of the X direction of the image area PicWH is 0 to W-1, and the value range of the Y direction is 0 to H-1; p(x,y)∈PicWH means that the pixel point p(x,y) is every pixel point in the image area PicWH.

[0024] Image point: An image point is an image point with an image value G(x,y) on a pixel p(x,y); for example, a grayscale image point is an image point with a grayscale value Gray(x,y) on a pixel p(x,y), 0≤Gray(x,y)≤V1; a color image point is an image point with a color value Color(x,y) formed by an image point with a red value Red(x,y), an image point with a green value Green(x,y) and an image point with a blue value Blue(x,y) on a pixel p(x,y), 0≤Red(x,y)≤V2, 0≤Green(x,y)≤V2, 0≤Blue(x,y)≤V2; contour image point It is an image point with an outline value OutlineP(x,y) on the pixel point p(x,y), 0≤OutlineP(x,y)≤V3; the depth image point is an image point with a depth value d(x,y) on the pixel point p(x,y), DeepMinV≤d(x,y)≤DeepMaxV; wherein, V1≥255, V2≥255, V3≥255, 30mm≤DeepMinV≤200mm, 300mm≤DeepMaxV≤3000mm; in this embodiment, V1=255, V2=255, V3=255; DeepMinV=200mm, DeepMaxV=2000mm.

[0025] Image: An image is an image composed of each image point G(x,y). p(x,y)∈PicWH; For example, a grayscale image is a grayscale image composed of image points with grayscale values ​​Gray(x,y). p(x,y)∈PicWH; a color image is a color image consisting of image points with color values ​​Color(x,y). p(x,y)∈PicWH; the contour image is the contour image composed of each image point with contour value OutlineP(x,y), p(x,y)∈PicWH; the depth image is a depth image composed of each image point with a depth value d(x,y), p(x,y)∈PicWH.

[0026] Image coordinate system: The image coordinate system is a coordinate system with the upper left corner of the image as the origin Op of the image coordinate system, the upper edge of the image as the Xp axis and the positive Xp axis from point Op to the right, the left edge of the image as the Yp axis and the positive Yp axis from point Op to the bottom.

[0027] Video image: Video image refers to an image that is output continuously at a certain rate (such as 25 frames / s).

[0028] Depth value: The depth value d(x, y) is the actual distance value from the camera 9 to the point on the object corresponding to the image point d(x, y).

[0029] Mean depth of a region: the average of all depth values ​​that are not 0 in the region.

[0030] The outline of an object: the set of each outline point of an object in the image; OutlineSet={P(xt,yt)|P(xt,yt) is the outline point of the object, t=1,2,…,T}; where t is the sequence of the outline points formed according to a set rule (such as in a clockwise direction according to the chain code), T is the number of outline points, T≥5; outline points refer to points on the boundary of an object in the image.

[0031] Actual pixel points of an object: The actual pixel points of an object refer to the number of pixel points / length unit on the object; in this embodiment, the length unit is 100 mm.

[0032] Neighborhood: The area consisting of the points whose distance from point p(x,y) is ≤R pixels, where R is the number of pixels in the image.

[0033] 8 Neighborhood: The area consisting of points whose distance from point p(x,y) is less than 2 pixels.

[0034] The calculation formula of the geometric center point AreaC (AreaCx, AreaCy) of the area Area is as follows:

[0035] In the formula, p(x,y)∈Area means taking each pixel point p(x,y) in the area Area, It is the x coordinate value of each pixel in the accumulated area Area. It is the y coordinate value of each pixel in the accumulated area Area, |Area| is the number of pixels in the area Area; int[] is an integer operation, and 0.5 is added to int[] to minimize the error of the operation value.

[0036] Camera optical axis 15: the center line of the lens of the camera 9 and the extension line of the center line.

[0037] Depth camera 9: A camera that can output color video images and depth video images.

[0038] The pose of the camera 9: The pose of the camera 9 refers to the position and rotation of the center point OP (OPx, OPy, OPz) of the target surface of the camera 9 in space. These two quantities are determined by the pose point RCP (x, y, z, a, b, c) of the robot 3; wherein, OPx=x+dx, OPy=y+dy, OPz=z+dz, in this embodiment, dx=0, dy=0, dz=100mm, so, OPx=x, OPy=y, OPz=z+100mm, and furthermore, OP(OPx, OPy, OPz)=OP(x, y, OPz); because the camera 9 is It is installed in parallel at the end of the robot 3, so when x or y or z in point RCP (x, y, z, a, b, c) translates, point OP (x, y, OPz) also translates; when a or b or c in the Euler angle of point RCP (x, y, z, a, b, c) rotates, point OP (x, y, OPz) rotates at the same angle, but the center point of rotation is point RCP, and the installation of camera 9 also satisfies that when a=b=c=0, the horizontal direction of the image collected by camera 9 is parallel to the XOY plane of the base coordinate system of robot 3, and the camera optical axis 15 is parallel to the X-axis of the base coordinate system of robot 3.

[0039] Robot 3: Robot 3 is composed of a robot body with an integrated controller, a robot teaching pendant, a power adapter and cables, and a network interface; the power adapter and cables provide the power required by robot 3; the network interface is the interface for robot 3 to communicate with external devices over the network; the robot 3 teaching pendant is a controller used to manually control robot 3; the main technical indicators of robot 3 are explosion-proof level, number of axes E, maximum load, maximum working radius, minimum parallel movement in the X, Y and Z directions, minimum rotation of Euler angles a, b, c, and minimum rotation of each axis; the posture of robot 3 refers to the posture of the center point RCP (x, y, z, a, b, c) of the end of robot 3 in the base coordinate system of robot 3; among them, x, y, z are the coordinate values ​​of point RCP in the X, Y and Z directions in the base coordinate system of robot 3, respectively, and a, b, c are the Euler angles rotating in space with point RCP as the center; changing the posture of robot 3 refers to intelligent analysis. The analyzer is connected to the robot 3 network port through a network cable to control the robot 3 to change the posture of the robot 3, wherein the change of posture refers to a change in at least one of a change in position and a change in rotation; the base coordinate system of the robot 3 refers to a coordinate system with the center point BasCP of the bottom surface of the base 6 of the robot 3 as the origin O, a line from the origin O to the left side of the refueling island 4 and parallel to the long side of the refueling island 4 as the Y positive axis, a line from the origin O to the front of the refueling island 4 as the X positive axis and the X axis is perpendicular to the Y axis, and the XOY plane is parallel to the ground 5, and a coordinate system from the origin O to the vertical upward as the Z positive axis and the Z axis is perpendicular to the XOY plane; in this embodiment, the main technical indicators of the robot are: explosion-proof grade ExdIIBT4, number of axes E=6, maximum load of 12Kg, maximum working radius of 1.3m, minimum parallel movement in the X direction, Y direction and Z direction ≤1mm, minimum rotation of Euler angles a, b, c ≤0.01°, and minimum rotation of each axis ≤0.01°.

[0040] Attached Figure 6In the figure, the maximum measurement range from the center point BasCP of the base 6 of the robot 3 to the left is LeftD=D1+D2; where D1=WR×Cosλ ×CosLa+dy; WR is the maximum working radius of the robot 3; λ is the angle between the point RCP and the XOY plane, λ=45°, La is the angle between the point RCPxy and the positive Y axis, La=45°, the point RCP is the center point of the end of the robot 3, and the point RCPxy (x, y, 0) is the vertical projection point of the point RCP (x, y, z, a, b, c) in the XOY plane; D2=DeepMaxV×tgβ; DeepMaxV is the maximum distance of the camera 9 to measure the depth, β=LA / 2, α=LA / 2, LA is the horizontal field of view of the camera 9; the point OP (x, y, OPz) is the center point of the target surface of the camera 9; dx, dy, dz are the X-direction deviation value, Y-direction deviation value, and Z-direction deviation value of the point OP and the point RCP in the base coordinate system of the robot 3, respectively. In the embodiment of the present invention, dx=0, dy=0, dz=100mm; BasCP is the center point of the bottom surface of the base 6 of the robot 3; MFH is the height from the center point MFC of the maximum surface 16 to the ground 5; when the position of the vehicle 1 meets the first preset condition, the point RCP is RCP (x0, y0, z0, a0, b0, c0), and the point OP is OP (x0, y0, OPz0).

[0041] Attached Figure 7 In the XRYRZR coordinate system, the origin is OR, and the XR axis, YR axis, and ZR axis are parallel to the X axis, Y axis, and Z axis of the robot 3 base coordinate system respectively; the origin OR is point RCP; the XvYvZv coordinate system is the coordinate system formed by rotating the XRYRZR coordinate system in space with the origin OR as the center, and the intersection line of the plane XRORYR in the XRYRZR coordinate system and the XvOvYv plane in the XvYvZv coordinate system is InterLine, and a is the intersection line between the XR axis and the intersection line Int where a1=b1=c1=0.

[0042] Attached Figure 8In the above equation, TMw is the actual width from the contour point TLC in the rear part 2 of the vehicle to the center point MFC of the maximum surface 16; TMh is the actual height from the point MFC to the point TLC; FTDistance is the distance from the center point MFC of the maximum surface 16 to the point TLC of the rear part 2 of the vehicle calculated by TMw and TMh; DeepMF is the mean depth from the camera 9 to the center area MFCA of the maximum surface 16; DeepTL is the mean depth from the camera 9 to the area TLCA; DeepCar is the mean depth from the camera 9 to the area CarCA of the vehicle 1; where point TLC is the outermost contour point of the rear part 2 of the vehicle The y value in is equal to the contour point in the center point MFC of the maximum surface 16; the point MidP is the middle point between the center point MFC of the maximum surface 16 and the point TLC; CarCA is the area centered on the intersection CarC of the camera optical axis 15 and the vehicle body; when the point MidP is located at the center point ImgC of the image, the point RCP (x1, y1, z1, a1, b1, c1) changes to RCP (x2, y2, z2, a2, b2, c2), and the point OP (x1, y1, OPz1) changes to OP (x2, y2, OPz2); wherein a2=a1, b2=b1, c2=c1.

[0043] Attached Fig. 9, when the robot 3 is translated so that the center point MFC of the maximum surface 16 is located at the center point ImgC of the image, the point RCP(x2, y2, z2, a2, b2, c2) changes to RCP(x3, y3, z3, a3, b3, c3), and the point OP(x2, y2, OPz2) changes to OP(x3, y3, OPz3); wherein a3=a2, b3=b2, c3=c2; DeepMF1 is the mean depth from the camera 9 to the center area MFCA of the maximum surface 16 when the point RCP is RCP(x3, y3, z3, a3, b3, c3); MFRect is the minimum circumscribed rectangle 17 of the maximum surface 16; MFRealW and MFRealH are the actual width and actual height of MFRect, respectively; RSM1, RSM 2. RSM3 and RSM4 are areas for measuring the maximum surface depth of 16; among them, RSM1=MaxFA∩RS1, RSM2=MaxFA∩RS2, RSM3=MaxFA∩RS3, RSM4=MaxFA∩RS4; the MaxFA is the area of ​​the maximum surface; ∩ is the intersection operation; the RS1, RS2, RS3, RS4 are four small rectangles for measuring the maximum surface depth of 16 and are located at the four corners of MFRect respectively, and the width and height of RS1, RS2, RS3, RS4 are both rw1 and rh1; Line1, Line2, Line3, Line4 are the upper edge line, lower edge line, left edge line, and right edge line of the rectangle MFRect respectively; the upper, lower, left, and right refer to the upper, lower, left, and right in the figure.

[0044] Attached Fig.10 In the figure, when the robot 3 is translated so that the center point InCP of the inner cover bottom surface 19 is located at the center point ImgC of the image, the point RCP(x3, y3, z3, a3, b3, c3) changes to RCP(x4, y4, z4, a4, b4, c4), and the point OP(x3, y3, OPz3) changes to OP(x4, y4, OPz4); wherein a4=a3, b4=b3, c4=c3; UpDeepV1 is the mean depth from the camera 9 to the area UpA; DnDeepV1 is the depth from the camera 9 to the area DnA Depth mean; AveDeepV1=(UpDeepV1+DnDeepV1) / 2; UDh is the height from area UpA to area DnA, UDw=UpDeepV1-DnDeepV1; point UpAC and point DnAC are the center points of area UpA and DnA respectively; σ is the angle between the bottom surface 19 of the inner cover and the Z axis of the coordinate system of the base of the robot 3; ω is the angle between the bottom surface 19 of the inner cover and the ground 5; VA is the vertical field of view of the camera 9; InCH is the height from the center point InCP of the bottom surface 19 of the inner cover to the ground 5.

[0045] Attached Fig.11 , when the posture of camera 9 satisfies the fourth posture condition, point RCP (x4, y4, z4, a4, b4, c4) changes to RCP (x5, y5, z5, a5, b5, c5), and point OP (x4, y4, OPz4) changes to OP (x5, y5, OPz5); wherein, x5 = x4 + KWx, y5 = y4, z5 = z4 + KWz, a5 = a4, b5 = b4 + σ, c5 = c4; KWx = int [0.5 + Kx - Wx]; KWz = int [0.5 + Kz + Wz]; W x and Wz are the X-direction offset and Z-direction offset of point OP (x4, y4, OPz4) when b4 in point RCP (x4, y4, z4, a4, b4, c4) rotates to b5; Kx and Kz are the X-direction movement and Z-direction movement of point RCP when it moves from point RCP (x4, y4, z4, a4, b4, c4) to point RCP (x5, y5, z5, a5, b5, c5); Definition and appendix of AveDeepV1, UDh, UDw, UpAC, DnAC, σ, ω, InCH Fig.10 The same; DeepV5 is the average depth from camera 9 to RS5; DeepV8 is the average depth from camera 9 to RS8; UpDeepV2 and DnDeepV2 are the average depth from camera 9 to area UpA and the average depth from camera 9 to area DnA when point OP is OP(x5,y5,OPz5); AveDeepV2=(UpDeepV2+nDeepV2) / 2.

[0046] Attached Fig.12 In the figure, KCenterDv is the average depth from camera 9 to the center area KCenterA of the inner cover knob 18, KEdgeDv is the average depth from camera 9 to the edge area KEdgeA of the inner cover knob 18, and AveDeepV2 is defined in the same way as in the attached figure. Fig.11 KnobH is the actual height from the inner cover knob 18 to the inner cover bottom surface 19; the measured values ​​in this figure are all measured when point OP is OP (x5, y5, OPz5).

[0047] Attached Fig.13 In the figure, OCRect is the minimum circumscribed rectangle 20 of the outer cover opening 11; OCRealW and OCRealH are respectively The actual width and actual height of OCRect; RealDiam is the actual diameter of the inner cover bottom surface 19; RS5, RS6, RS7, RS8 are areas used to measure the surface depth of the outer cover opening 11, and the widths of RS5, RS6, RS7, RS8 are all rw2 and the heights are rh2; Line5, Line6, Line7, Line8 are the four sides of the inner cover knob 18 respectively; KnobSlope is the average of the slopes of the two long sides Line5 and Line6 of the inner cover knob 18 relative to the horizontal direction of the image captured by the camera 9, and θ is the angle between the inner cover knob 18 and the ground 5; UpA is the neighborhood of the contour points of the inner cover bottom surface 19 with point UpAC as the center point; DnA is a neighborhood with point DnAC as the center point and is a contour point of the inner cover bottom surface 19; LtA is a neighborhood with point LtAC as the center point and is a contour point of the inner cover bottom surface 19; RtA is a neighborhood with point RtAC as the center point and is a contour point of the inner cover bottom surface 19; point KC (KCx, KCy) is the center point of the inner cover knob 18; KCMDv is the distance from the center point KC of the inner cover knob 18 to point MinDp, and point MinDp is the point with the shortest distance from point KC to all contour points of the inner cover knob 18; KnobW is the actual width of the inner cover knob 18.

[0048] Embodiments 1 to 5 are a vehicle parameter measurement method based on three-dimensional vision, wherein the method is implemented by a vehicle parameter measurement system based on three-dimensional vision, and the implementation steps of the method are as follows: Step A, generating a neural network; the neural network is a neural network that can identify a target object; In Example 1 and Example 2, a neural network is generated; the neural network is a Yolov8 neural network capable of identifying a target object; In Example 2, the completion time of generating the neural network is 15:37 on May 2, 2024; Step B, measurement processing; Embodiment 3 to Embodiment 5, measurement process (the process of measurement process is as shown in the attached Figure 3 shown); In this embodiment, this step starts at 08:12 on May 3, 2024; The steps of the measurement process are as follows: The vehicle 1 is placed within the measuring range; Acquire image data of the target object and measure the actual distance from the center of the largest surface to the rear of the vehicle; Acquire image data of the target object, measure the shape and depth data of the largest surface and the actual size; Acquire image data of the target object and measure the fuel tank cap data; the fuel tank cap data includes parameters of the outer cover opening 11, parameters of the inner cover bottom surface 19 and parameters of the inner cover knob 18; the parameters of the outer cover opening 11 include the shape and depth data of the outer cover opening 11 and the actual size of the outer cover opening 11; the parameters of the inner cover bottom surface 19 include the height from the center point of the inner cover bottom surface 19 to the ground 5, the angle ω between the inner cover bottom surface 19 and the ground 5, the shape and depth data of the inner cover bottom surface 19 and the actual size of the inner cover bottom surface 19; the parameters of the inner cover knob 18 include the shape and depth data of the inner cover knob 18, the rotation angle θ of the inner cover knob 18 and the actual size of the inner cover knob 18; The measured vehicle data is uploaded to the server; the vehicle data includes the actual distance from the center of the maximum surface 16 to the rear of the vehicle 2, the shape and depth data and actual size of the maximum surface 16, and the fuel tank cap data.

[0049] In Embodiment 3 to Embodiment 5, the steps of placing the vehicle 1 within the measurement range are as follows: Step 2.1: If the vehicle meets the first preset condition, go to step 2.3; otherwise go to step 2.2; Embodiment 3: Vehicle 1 does not meet the first preset condition, and the process goes to step 2.2; In Embodiment 4 and Embodiment 5, if the vehicle 1 meets the first preset condition, the process goes to step 2.3; The first preset condition includes the following three conditions: (1) the image captured by the camera 9 includes the outer cover 10 and the rear part 2 of the vehicle, (2) DeepMinV≤the distance from the camera 9 to the vehicle 1≤DeepMaxV, (3) LeftD≤OCLeft∧CarRight≤RightD; wherein ∧ is an AND operation; DeepMinV and DeepMaxV are the minimum value and the maximum value of the depth measured by the camera 9, respectively; OCLeft is the left boundary of the outer cover 10, and CarRight is the right boundary of the rear part 2 of the vehicle; LeftD and RightD are calculated by formula (1) and formula (2), respectively. LeftD=WR×Cosλ×CosLa+DeepMaxV×tgβ+dy (1) RightD=WR×Cosλ×CosRa+DeepMaxV×tgα+dy (2) In equations (1) and (2), WR is the maximum working radius of robot 3; λ is the angle between the line connecting point RCP to the origin O and the XOY plane, λ=45°; La is the angle between the line connecting point RCPxy to the origin O and the positive Y axis when point RCPxy is located to the left of point BasCP, La=45°; Ra is the angle between the line connecting point RCPxy to the origin O and the negative Y axis when point RCPxy is located to the right of point BasCP, Ra=45°; β=LA / 2, α=LA / 2, LA is the horizontal field of view of camera 9; RCP(x,y,z,a,b,c) is the center point of the end of robot 3 and also the position in the base coordinate system of robot 3; point RCPxy(x,y,0) is the vertical projection of point RCP to XOY plane; dy is the Y-direction deviation value between point OP and point RCP; the XOY plane refers to the XOY plane in the base coordinate system of the robot 3; the Y axis refers to the Y axis in the base coordinate system of the robot 3; in this embodiment, WR=1.3m, DeepMinV=0.2m, DeepMaxV=2.0m, LA=64°, dy=0, when the position of the vehicle 1 meets the first preset condition, RCP is RCP(x0,y0,z0,a0,b0,c0), point RCPxy is RCPxy(x0,y0,0), and point OP is OP(x0,y0,OPz0); wherein a0, b0, c0 are all set initial values; in this embodiment, a0=0, b0=0, c0=0; as shown in the attached Figure 6 As shown; Step 2.2, issuing an instruction for adjusting the vehicle position; Embodiment 3, using the intercom device to issue a voice message indicating the position adjustment of the vehicle 1; Step 2.3: End the processing of this step.

[0050] Embodiment 3 to Embodiment 5, end the processing of this step.

[0051] In Embodiment 4 and Embodiment 5, the steps of obtaining the image data of the target object are as follows: The intelligent analyzer intercepts a frame of image Img and a frame of depth image DeepImg from the video image output by the camera 9; the image Img refers to a color image, wherein the width of the image Img and the height of the depth image DeepImg are both W and H; in this embodiment, W=1280, H=800; Compress the image Img into a compressed image CompImg; the compressed image CompImg is an image with a width of w and a height of h; w and h are the width and height of the image processed by the neural network; in this embodiment, the neural network is a Yolov8 neural network, w=608, h=608; Using the neural network to identify each target object from the compressed image, and obtaining the minimum bounding rectangle of each target object in the image based on the minimum bounding rectangle that frames each target object in the compressed image; In this embodiment, the Yolov8 neural network is used to identify each target object from the compressed image CompImg, and the minimum bounding rectangle (ImgRect(1), ImgRect(2), ImgRect(3)) of each target object in the image Img is obtained according to the minimum bounding rectangle (CompIRect(1), CompIRect(2), CompIRect(3)) that frames each target object in the compressed image CompImg; wherein CompIRect(n) is the minimum bounding rectangle that frames the target object (n), and n is in increasing order of the center point of the rectangle CompIRect(n). The numbers formed are n=1, 2, 3; the numbers formed in increasing order of the points refer to the numbers formed according to the y coordinate values ​​of the points in the image coordinate system from small to large, and then according to the x coordinate values ​​from small to large; the image coordinate system refers to a coordinate system with the upper left corner of the image as the origin Op, the upper edge of the image as the Xp axis and the positive Xp axis pointing from the origin Op to the right, and the left edge of the image as the Yp axis and the positive Yp axis pointing from the origin Op to the bottom; in this embodiment, the image coordinate system refers to the coordinate system of the compressed image CompImg; the following is a calculation example of obtaining ImgRect(1) in the image Img based on CompIRect(1) in the compressed image CompImg, IRSx(1)=Min(w-1,int[0.5+CIsx(1)×W / w]), IREx(1)=Min(w-1,int[0.5+CIex(1)×W / w]) IRSy(1)=Min(h-1,int[0.5+CIsy(1)×H / h]), IREy(1)=Min(h-1,int[0.5+CIey(1)×H / h]) Where, Min() is the minimum value operation; CIsx(1), CIex(1), CIsy(1), CIey(1) are the left boundary value, right boundary value, upper boundary value, and lower boundary value of CompIRect(1); IRSx(1), IREx(1), IRSy(1), IREy(1) are the left boundary value, right boundary value, upper boundary value, and lower boundary value of ImgRect(1); W=1280, w=608, H=800, h=608; Using an image processing algorithm to obtain image data of each target object from the area of ​​the smallest circumscribed rectangle of each target object in the image; the image data of the target object includes the outline of the target object, the area of ​​the target object, the center point of the target object area, and the depth value of each pixel point in the target object area; the target objects are target T1, target T2, target T3, target T4, target T5, and target T6; wherein the target T1 is the rear part 2 of the vehicle, the target T2 is the outer cover 10, the target T3 is the maximum surface 16, and the target T4 is The outer cover opening 11, the target object T5 refers to the inner cover bottom surface 19, and the target object T6 refers to the inner cover knob 18; the vehicle tail 2 refers to the outermost contour part of the vehicle tail; the outer cover 10 refers to the outer cover of the fuel tank and is the outer cover in a fully closed state; the maximum surface 16 refers to the surface with the largest area and the smoothest surface among the surfaces of the outer cover 10; the outer cover opening 11 refers to the outer cover opening of the fuel tank and is the outer cover opening when the outer cover is in a fully open state; the inner cover bottom surface 19 refers to the bottom part of the inner cover of the fuel tank; the inner cover knob 18 refers to the knob part on the inner cover bottom surface 19.

[0052] In this embodiment, the maximum contour algorithm and the filling algorithm in the image processing algorithm are used to obtain the image data of each target object Tk from the area of ​​the minimum circumscribed rectangle of each target object Tk in the image Img. k∈Z[1,6]; the image data of the target object Tk includes a set of contour points of the target object Tk, an area of ​​the target object Tk, a center point of the area of ​​the target object, and a set of depth values ​​d(x,y) of each pixel point p(x,y) in the area of ​​the target object Tk; wherein Z[n1,n2] is a set from integer n1 to integer n2, n1≤n2; The steps of processing the target object Tk by the maximum contour algorithm are as follows: (1) graying the color image Img to obtain a grayscale image GrayImg; (2) performing noise reduction filtering on the grayscale image GrayImg using a filtering algorithm (such as a Gaussian filtering algorithm); (3) obtaining edge points in the area ImgRect(n) of the minimum circumscribed rectangle of the target object Tk in the grayscale image GrayImg using an edge detection algorithm (such as a Canny edge detection algorithm); edge points refer to edge points in the outline image OutlineImg; (4) in the area ImgRect(n), using a filtering algorithm (such as a Gaussian filtering algorithm) to delete isolated edge points in the outline image OutlineImg; isolated edge points refer to edge points that have no edge points in the 8-neighborhood of the point; (5) in the area ImgRect(n), using a convex hull algorithm (such as a Convex Hull algorithm) to connect the edge points in the outline image OutlineImg to obtain the contour of the target object Tk, k∈Z[1,6]; The filling algorithm refers to using a contour filling algorithm to fill the contour of the target object Tk to obtain the area enclosed by the contour of the target object Tk and the geometric center point of the area of ​​the target object Tk, k∈Z[1,6]; in this embodiment, the contour filling algorithm is a seed filling algorithm; In this embodiment, when measuring the actual distance from the center of the maximum surface 16 to the rear end 2 of the vehicle, the maximum contour algorithm and the filling algorithm are used to obtain the image data of the target object T1 from the area of ​​the minimum circumscribed rectangle of the rear end 2 of the target object T1 in the image Img; similar processing is performed to obtain the image data of the maximum surface 16 of the target object T3; the image data of the target object T3 includes a set MFOutline of contour points of the maximum surface 16 of the target object T3, MFOutline={G1(xt,yt)|G1(xt,yt) is the contour point of the maximum surface 16, t=1,2 ,...,T(3)}; where T(3) is the number of contour points of the maximum surface 16, and T(3) is a positive integer; the set MaxFA of the area of ​​the target object T3, MaxFA={p(x,y)|p(x,y) is in the area surrounded by MFOutline}; the center point MFC(MFCx,MFCy) of the area of ​​the target object T3; the set MFDeepA of the depth value of each pixel in the target area, MFDeepA={d(x,y)|p(x,y)∈MaxFA}; where d(x,y) is the depth value at the pixel point p(x,y).

[0053] In Embodiment 4 and Embodiment 5, the steps of acquiring the image data of the target object and measuring the actual distance from the center of the largest surface to the rear of the vehicle are as follows: Step 4.1, obtain the image data of the target objects T1, T2 and T3, so that the camera 9 satisfies the first pose condition; the first pose condition means that the image Img contains the target objects T1, T2 and T3, and the distance Distance1 from the camera 9 to the vehicle 1 is within the first preset range Rang1; the first preset range Rang1 is the range within which the neural network can recognize the target objects T1, T2 and T3; wherein, MinRang1≤Rang1≤MaxRang1; MinRang1 and MaxRang1 are the minimum and maximum values ​​of Rang1 respectively; the neural network can recognize the target within the range Rang1 means RealPnX1≥RegPnX∧R ealPnY1≥RegPnY; RealPnX1 and RealPnY1 are respectively the actual pixel points in the X direction and the actual pixel points in the Y direction of the smallest object in the image recognized by the neural network; RegPnX and RegPnY are respectively the pixel points in the X direction and the pixel points in the Y direction of the smallest object recognized by the neural network; in this embodiment, MinRang1=500mm, MaxRang1=2000mm; RealPnX1=RealPnY1≥16, RegPnX=RegPnY=16 (for example, the P4 output layer of the Yolov8 neural network can recognize at least an object with 16 pixels in the X direction and 16 pixels in the Y direction, the same below); the smallest object is the largest surface 16 with the smallest size (because the size of the rear part 2 of the vehicle and the outer cover 10 are both larger than the largest surface 16); In this embodiment, when Distance1=2000mm, the measurement range and RealPnX1 and RealPnY1 are calculated as follows: TotalW1=2×MaxRang1×tg(LA / 2)=4000mm×tg(32.0°)=2499.48mm TotalH1=2×MaxRang1×tg(VA / 2)=4000mm×tg(20.5°)=1495.54mm RealPnX1=ThingW1×w / TotalW1=100mm×608 / 2499.48mm=24.33>RegPnX=16 RealPnY1=ThingH1×h / TotalH1=100mm×608 / 1495.54mm=40.65>RegPnY=16 In the above formula, VA is the vertical field of view of the camera 9, VA=41°; TotalW1 (=2499.48mm) and TotalH1 (=1495.54mm) are respectively the actual total width and total height of the object corresponding to the image Img when Distance1=2000mm; in this embodiment, ThingW1=ThingH1=100mm is the size of the smallest object (the largest surface 16 with the smallest size) at the narrowest point; RealPnX1 and RealPnY1 are the actual pixel points when Distance1=2000mm; because the actual pixel points are at the maximum distance , so it is only necessary to calculate the actual pixel point when the maximum distance Distance1=2000mm (the same below); according to the above calculation, it can be known that in this embodiment, within the first preset range Rang1, the actual pixel point can meet the requirements of the Yolov8 neural network to identify the target object T1, the target object T2, and the target object T3, and because the distance between the boundary of the outer cover 10 of most vehicles and the rear part 2 of the vehicle is ≤2499mm; therefore, within the range Rang1, the system can meet the requirements of identifying the target object and measuring the distance between the rear part 2 of the vehicle and the center point of the maximum surface 16; In this embodiment, the steps of acquiring the image data of the target objects T1, T2 and T3 so that the camera 9 satisfies the first pose condition are as follows: Step 4.1.1, make the camera optical axis 15 / / X axis and make the horizontal direction of the image Img / / XOY plane; / / indicates parallel; the X axis refers to the X axis in the base coordinate system of the robot 3; In this embodiment, the steps of making the camera optical axis 15 / / X axis and making the horizontal direction of the image Img / / XOY plane are as follows: If the camera optical axis 15 is not parallel to the X-axis or the horizontal direction of the image Img is not parallel to the XOY plane, the intelligent analyzer is used to control the robot 3 to rotate the Euler angles a, b, and c in the point RCP, so that the Xv axis, Yv axis, and Zv axis in the XvYvZv coordinate system are respectively parallel to the X axis, Y axis, and Z axis in the base coordinate system of the robot 3; the XvYvZv coordinate system is a coordinate system formed by rotating the XRYRZR coordinate system in space with the origin OR as the center of the circle, and the XRYRZR coordinate system is a coordinate system with the origin OR as the center and the XR axis, YR axis, and Z axis as the center. , ZR axes are respectively parallel to the X axis, Y axis, and Z axis of the robot 3 base coordinate system, and the origin OR is the point RCP; after the rotation, the camera optical axis 15 / / X axis and the horizontal direction of the image Img / / XOY plane; the rotation changes the point RCP (x0, y0, z0, a0, b0, c0) to RCP (x1, y1, z1, a1, b1, c1), and the point OP (x0, y0, OPz0) changes to OP (x1, y1, OPz1); wherein a1=0, b1=0, c1=0; as shown in the attached figure Figure 7 As shown; Step 4.1.2, measure the depth mean DeepCar from the camera 9 to the area CarCA of the vehicle 1; the area CarCA is an area centered on the intersection CarC of the camera optical axis 15 and the body of the vehicle 1; wherein, the depth mean from the camera 9 to the area refers to the average value of all depth values ​​in the area that are not 0, and the area refers to a neighborhood consisting of ≤R pixels from the center point of the area; the depth value is the distance value from the camera 9 to the point on the object corresponding to the image point; int[H / 50]≤R≤int[H / 20]; H is the height of the image Img; in this embodiment, R=in[H / 20]=int[800 / 20]=40; In this embodiment, the depth mean DeepCar is calculated by the following formula:

[0054] Wherein, the depth value d(x, y)' is the depth value d(x, y) that is not 0 in the area CarCA; the area CarCA' is the area where the depth value d(x, y) is not 0 in the area CarCA; the area CarCA is the neighborhood formed by the point CarC≤R=40 pixels; the following depth mean values ​​are all calculated using a similar method; Step 4.1.3: If DeepCar≠0, go to step 4.1.4; otherwise, go to step 4.1.7; Step 4.1.4: If MinRang1≤DeepCar∧DeepCar≤MaxRang1, go to step 4.1.6; otherwise go to step 4.1.5; Step 4.1.5, the intelligent analyzer controls the robot 3 to move so that the x coordinate value of point RCP changes, and the x coordinate value of point OP also changes accordingly, thereby satisfying the condition of MinRang1≤DeepCar∧DeepCar≤MaxRang1; For example, when DeepCar<MinRang1, the intelligent analyzer controls the robot 3 to move so that the x coordinate value of the point RCP moves in the negative X direction by MinRang1-DeepCar; after the movement, DeepCar=MinRang1; Step 4.1.6, acquiring image data of target object T1, target object T2 and target object T3, so that point MidP in the image Img is located at the center point ImgC of the image; wherein point MidP is the middle point between the center point MFC (MFCx, MFCy) of the maximum surface 16 and point TLC (TLCx, TLCy); point TLC (TLCx, TLCy) is the contour point of the rear part 2 of the vehicle in the image Img and is the point where TLCy=MFCy; In this embodiment, the steps of acquiring the image data of the target objects T1, T2 and T3 so that the point MidP in the image Img is located at the center point ImgC of the image are as follows: Acquire image data of the target object T2 and the target object T3, and obtain the regional center point MFC (MFCx, MFCy) of the target object T3 from the image data; Acquire image data of the target object T1, obtain a set of contour points of the target object T1 from the image data, and then find the outermost contour point TLC (TLCx, TLCy) of the target object T1 from the set of contour points; wherein TLCy=MFCy; The midpoint MidP (MidPx, MidPy) between point MFC and point TLC is calculated using the following formula: MidPx=int[0.5+(MFCx+TLCx) / 2] MidPy=MFCy The intelligent analyzer controls the robot 3 to move the point RCP, and the camera 9 moves accordingly, so that the point MidP (MidPx, MidPy) in the image Img captured by the camera 9 moves to the center point ImgC (ImgCx, ImgCy) of the image Img; The specific steps to move point MidP (MidPx, MidPy) to point ImgC (ImgCx, ImgCy) are as follows: When the distance from the camera 9 to the vehicle 1 is DeepCar, the point MidP (MidPx, MidPy) is moved to the point ImgC (ImgCx, ImgCy), and the movement amount DCy of the point RCP in the Y direction and the movement amount DCz in the Z direction are calculated using the following formula: DCy=DeepCar×tg(LA / 2)×(MidPx-ImgCx) / W DCz=DeepCar×tg(VA / 2)×(MidPy-ImgCy) / H Wherein, the intelligent analyzer controls the robot 3 to make the point RCP move DCy in the Y direction and DCz in the Z direction, and if DCy≥0, the Y direction moves to the positive Y direction, otherwise it moves to the negative Y direction; if DCz≥0, the Z direction moves to the positive Z direction, otherwise it moves to the negative Z direction; after the movement is completed, the point MidP moves to the point ImgC; the movement of the following points is realized by a similar method; When the point MidP is moved so that it is located at the center point ImgC of the image, the point RCP(x1,y1,z1,a1,b1,c1) changes to RCP(x2,y2,z2,a2,b2,c2), and the point OP(x1,y1,OPz1) changes to OP(x2,y2,OPz2); where a2=a1, b2=b1, c2=c1; as shown in the attached figure Figure 8 As shown; Step 4.1.7, finish the process of acquiring the image data of the target objects T1, T2 and T3, so that the camera 9 satisfies the first pose condition; Step 4.2, acquiring image data of target object T1 and target object T3, and measuring the actual distance from the center of the largest surface to the rear of the vehicle; In this embodiment, the steps of acquiring image data of the target object T1 and the target object T3 and measuring the actual distance from the center of the largest surface to the rear of the vehicle are as follows: Step 4.2.1, obtain the image data of the target object T1 and the target object T3, measure the depth mean DeepMF from the camera to the area MFCA, the depth mean DeepTL from the camera to the area TLCA, and the depth mean DeepCar from the camera to the area CarCA; where MFCA is the area centered on the center point MFC (MFCx, MFCy) of the maximum surface; TLCA is the area centered on the outermost contour point TLC (TLCx, TLCy) of the rear of the vehicle; CarCA is the area centered on the intersection point CarA of the camera optical axis and the vehicle; In this embodiment, image data of the target object T1 and the target object T3 are obtained, and the depth mean DeepMF from the camera 9 to the area MFCA, the depth mean DeepTL from the camera 9 to the area TLCA, and the depth mean DeepCar from the camera 9 to the area CarCA are measured; Step 4.2.2: If DeepMF=0∨DeepTL=0∨DeepCar=0, go to step 4.2.3; otherwise go to step 4.2.4; Example 4, DeepMF=0, go to step 4.2.3; Example 5, DeepMF≠0∧DeepTL≠0∧DeepCar≠0, go to step 4.2.4; Step 4.2.3: The device fails. After this step is completed, go to step 4.2.5; Embodiment 4: Display device failure using a display connected to the intelligent analyzer. After this step is completed, proceed to step 4.2.5; Step 4.2.4, calculate the actual distance from the center of the largest surface to the rear of the vehicle; Embodiment 5, the steps of calculating the actual distance from the center of the largest surface 16 to the rear of the vehicle 2 are as follows: The intelligent analyzer uses formula (3) to calculate the total width TotalCA of the actual size of the image Img corresponding to the depth DeepCar, TotalCA=2×DeepCar×tg(LA / 2) (3) The intelligent analyzer uses formula (4) to calculate the actual width TMw from point TLC to point MFC, TMw=TotalCA×(TLCx-MFCx) / W (4) The intelligent analyzer uses formula (5) to calculate the actual height TMh from point TLC to point MFC. TMh=DeepMF-DeepTL (5) The intelligent analyzer calculates the actual distance FTDistance from the center of the largest surface 16 to the rear of the vehicle 2 using formula (6): FTDistance=int[0.5+(TMw 2 +TMh 2 ) 1 / 2 ] (6) Embodiment 5, the actual distance FTDistance from the center of the maximum surface 16 to the rear of the vehicle 2 is calculated using formula (6); Step 4.2.5, end the process of acquiring the image data of the target object T1 and the target object T3, and measuring the actual distance from the center of the largest surface to the rear of the vehicle.

[0055] In the fourth embodiment and the fifth embodiment, the process of acquiring the image data of the target object T1 and the target object T3 and measuring the actual distance from the center of the largest surface to the rear of the vehicle is completed.

[0056] In Example 5, the steps of obtaining the image data of the target object and measuring the shape and depth data of the maximum surface and the actual size are as follows: Step 5.1, obtain the image data of the target object T2 and the target object T3, so that the camera 9 satisfies the second posture condition; the second posture condition means that the image Img contains the target object T2 and the target object T3, and the distance Distance2 from the camera 9 to the vehicle 1 is within the second preset range Rang2; the second preset range Rang2 is the range within which the neural network can recognize the target object T2 and the target object T3; wherein MinRang2≤Rang2≤MaxRang2; MinRang2 and MaxRang2 are Rang 2; the neural network can identify the target object within the range Rang2 means RealPnX2≥RegPnX∧RealPnY2≥RegPnY; RealPnX2 and RealPnY2 are respectively the actual pixel points in the X direction and the actual pixel points in the Y direction of the smallest object in the image recognized by the neural network; in this embodiment, MinRang2=200mm, MaxRang2=500mm; RealPnX2=RealPnY2≥16, RegPnX=RegPnY=16; the smallest object is the largest surface with the smallest size 16; In this embodiment, when Distance2=500mm, the measurement range and the calculation of RealPnX2 and RealPnY2 are as follows: TotalW2=2×MinRang2×tg(LA / 2)=1000mm×tg(32.0°)=624.87mm TotalH2=2×MinRang2×tg(VA / 2)=1000mm×tg(20.5°)=373.89mm RealPnX2=ThingW1×w / TotalW2=100mm×608 / 624.87mm=97.30>RegPnX=16 RealPnY2=ThingH1×h / TotalH2=100mm×608 / 373.89mm=162.62>RegPnY=16 Wherein, TotalW2 (=624.87mm) and TotalH2 (=373.89mm) are respectively the total width and total height of the actual object corresponding to the image Img when Distance2=500mm; ThingW1=ThingH1=100mm is the size of the narrowest part of the smallest object (the largest surface 16 with the smallest size) to be identified; According to the above calculation, in this embodiment, within the second preset range Rang2, the Yolov8 neural network can correctly identify the target T2 and the target T3, and the width of most of the fuel tank covers 10 is ≤620mm and the height is ≤370mm, so the second preset range Rang2 can meet the demand of measuring the maximum surface 16; In this embodiment, the steps of acquiring the image data of the target object T2 and the target object T3 so that the camera 9 satisfies the second posture condition are as follows: Step 5.1.1, make the camera optical axis 15 / / the X axis and make the horizontal direction of the image Img / / the XOY plane; In this embodiment, because the camera optical axis 15 has been / / the X axis and the horizontal direction of the image Img has been / / the XOY plane, this step is not processed; Step 5.1.2, measure the depth mean DeepMF1 from camera 9 to the regional MFCA; Step 5.1.3: If MinRang2≤DeepMF1∧DeepMF1≤MaxRang2, go to step 5.1.5; otherwise go to step 5.1.4; Step 5.1.4, the intelligent analyzer controls the robot 3 to move so that the x coordinate value of point RCP changes, and the x coordinate value of point OP also changes accordingly, thereby satisfying the condition of MinRang2≤DeepMF1∧DeepMF1≤MaxRang2; Step 5.1.5, acquiring image data of the target object T2 and the target object T3, so that the point MFC is located at the center point ImgC of the image; The steps of acquiring the image data of the target object T2 and the target object T3 so that the point MFC is located at the center point ImgC of the image are as follows: Acquire image data of the target object T2 and the target object T3, and obtain the center point MFC (MFCx, MFCy) of the maximum surface 16 area of ​​the target object T3 from the image data; The robot 3 is controlled by the intelligent analyzer to change the posture of the robot 3, and the camera 9 moves accordingly, so that the point MFC (MFCx, MFCy) in the image Img moves to the center point ImgC (ImgCx, ImgCy) of the image Img; when the robot 3 is translated, when the center point MFC of the maximum surface 16 is located at the center point ImgC of the image, the point RCP (x2, y2, z2, a2, b2, c2) changes to RCP (x3, y3, z3, a3, b3, c3), and the point OP (x2, y2, OPz2) changes to OP (x3, y3, OPz3); wherein a3=a2, b3=b2, c3=c2; as shown in the attached figure Fig. 9 As shown; Step 5.2, acquiring image data of the target object T3, measuring the depth data of the maximum surface and measuring the actual size of the maximum surface; the actual size of the maximum surface 16 includes the height MFH from the center point of the maximum surface to the ground, the actual width MFRealW and the actual height MFRealH of the minimum circumscribed rectangle of the maximum surface; The steps of acquiring the image data of the target object T3, measuring the depth data of the largest surface and measuring the actual size of the largest surface are as follows: Step 5.2.1, acquiring image data of the target object T3, and measuring depth data of the maximum surface 16; The steps of acquiring image data of the target object T3 and measuring the depth data of the maximum surface 16 are as follows: Acquire image data of the target object T3, obtain a region MaxFA of the maximum surface 16 from the image data, and obtain a minimum circumscribed rectangle MFRect of the maximum surface 16 according to a boundary value of the region MaxFA; Four measurement areas RS1, RS2, RS3, and RS4 are obtained according to the minimum circumscribed rectangle MFRect; wherein RS1, RS2, RS3, and RS4 are areas with a width of rw1 and a height of rh1, and are areas of the minimum circumscribed rectangle MFRect parallel to the maximum surface 16, and the upper left corner of RS1 is the upper left corner of MFRect, the upper right corner of RS2 is the upper right corner of MFRect, the lower right corner of RS3 is the lower right corner of MFRect, and the lower left corner of RS4 is the lower left corner of MFRect; wherein rw1=int[MFRectW / M1], rh1=int[MFRectH / M1]; MFRectW and MFRectH are respectively the width and height of MFRect in the image Img, 4≤M1≤6; in this embodiment, M1=6; Through the following calculation, four actual measurement areas RSM1, RSM2, RSM3, and RSM4 are obtained: RSM1=MaxFA∩RS1, RSM2=MaxFA∩RS2, RSM3=MaxFA∩RS3, and RSM4=MaxFA∩RS4; Measure the mean depth value DeepV1 from camera 9 to area RSM1, the mean depth value DeepV2 from camera 9 to area RSM2, the mean depth value DeepV3 from camera 9 to area RSM3, and the mean depth value DeepV4 from camera 9 to area RSM4; Step 5.2.2, acquiring image data of the target object T3, and measuring the actual size of the maximum surface 16; The steps of acquiring image data of the target object T3 and measuring the actual size of the maximum surface 16 are as follows: The height MFH from the center point MFC of the maximum surface 16 to the ground 5 is calculated using formula (7): MFH=int[0.5+StrH+BasH+dz+z] (7) (7) Wherein, StrH is the height of point BasCP from the ground 5, where the ground 5 is the ground where the tires of the vehicle are located; BasH is the height from point RCP to point BasCP when the coordinate z in the base coordinate system of robot 3 is 0; dz is the Z-direction deviation value between point OP and point RCP, and z is the height value of point RCP in space; point BasCP is the origin of the base coordinate system of robot 3; point RCP is the center point of the end of robot 3 in the base coordinate system of robot 3, and the position and rotation amount of point RCP are respectively the spatial position of the base coordinate system of robot 3 and the rotation amount described by Euler angles; in this embodiment, z=z3; According to the DeepMF1, the total width TotalWMF1 and the total height TotalHMF1 of the actual object corresponding to the image Img captured by the camera 9 at the depth DeepMF1 are calculated using formula (8). TotalWMF1=2×DeepMF1×tg(LA / 2) TotalHMF1=2×DeepMF1×tg(VA / 2) (8) According to the TotalWMF1 and TotalHMF1, the MFRealW and MFRealH are calculated using formula (9). MFRealW=int[0.5+TotalWMF1×MFRectW / W] MFRealH=int[0.5+TotalHMF1×MFRectH / H] (9) (9) In formula, W and H are the width and height of the image Img respectively.

[0057] In Example 5, the steps of obtaining the image data of the target object and measuring the fuel tank cover data are as follows: Step 6.1, obtain the image data of the target object T5, so that the camera 9 satisfies the third posture condition; the third posture condition means that the image Img contains the target objects T4, T5 and T6, and the distance Distance2 from the camera 9 to the vehicle 1 is within the second preset range Rang2; the second preset range Rang2 is the range within which the neural network can identify the target objects T4, T5 and T6; the neural network can identify the target within the range Rang2 means RealPnX3≥RegPnX∧RealPnY3≥RegPnY; RealPnX3 and RealPnY3 are respectively the actual pixel points in the X direction and the actual pixel points in the Y direction of the smallest object in the image recognized by the neural network; in this embodiment, MinRang2=200mm, MaxRang2=500mm; RealPnX3=RealPnY3≥16, RegPnX=RegPnY=16; the smallest object is the inner cover knob 18 with the smallest size; In this embodiment, when Distance2=200mm, the measurement range and the calculation of RealPnX3 and RealPnY3 are as follows: TotalW3=2×MinRang2×tg(LA / 2)=400mm×tg(32.0°)=249.95mm TotalH3=2×MinRang2×tg(VA / 2)=400mm×tg(20.5°)=149.55mm RealPnX3=ThingW3×w / TotalW3=7mm×608 / 249.95mm=17.03>RegPnX=16 RealPnY3=ThingH3×h / TotalH3=7mm×608 / 149.55mm=28.46>RegPnY=16 In the above formula, TotalW3 (=249.95mm) and TotalH3 (=149.55mm) are respectively the actual total width and total height of the object corresponding to the image Img when Distance2=200mm; According to the above calculation, in this embodiment, when Distance2=200mm, the smallest object that can be identified (the width of the smallest inner cover knob 18) has the narrowest size of ThingW3=ThingH3=7mm; In this embodiment, when Distance2=500mm, RealPnX3 and RealPnY3 are calculated as follows: RealPnX3=ThingW2×w / TotalW2=17mm×608 / 624.87mm=16.54>RegPnX=16 RealPnY3=ThingH2×h / TotalH2=17mm×608 / 373.89mm=27.65>RegPnY=16 According to the above calculation, in this embodiment, when Distance2=500mm, the smallest object that can be identified (the width of the inner cover knob 18 with the smallest size) has a size of ThingW2=ThingH2=17mm at its narrowest point; therefore, within the second preset range Rang2, the Yolov8 neural network can identify a target object whose narrowest point is ≥7mm, and when the narrowest point of the object is ≥17mm, it can identify a target object whose width is ≤624mm and height is ≤373mm; the target object includes the outer cover opening 11, the inner cover bottom surface 19, and the inner cover knob 18; therefore, the range Rang2 meets the requirements for measuring the fuel tank cap of a vehicle; In this embodiment, the steps of acquiring the image data of the target object T5 so that the camera 9 satisfies the third posture condition are as follows: Step 6.1.1, make the camera optical axis 15 / / the X axis and make the horizontal direction of the image Img / / the XOY plane; In this embodiment, because the camera optical axis 15 has been / / the X axis and the horizontal direction of the image Img has been / / the XOY plane, this step does not change the posture of the RCP. Step 6.1.2, acquiring image data of the target object T5, so that the camera 9 satisfies the third posture condition; In this embodiment, because the distance Distance2 from the camera 9 to the vehicle 1 is already within the second preset range Rang2, the robot 3 is not moved in this step. Step 6.2, measure the height from the center point of the bottom surface of the inner cover to the ground; The steps to measure the height from the center point of the bottom surface of the inner cover to the ground are as follows: Step 6.2.1, acquiring image data of the target object T5, so that the center point of the inner cover bottom surface 19 is located at the center point ImgC of the image; In this embodiment, the steps of acquiring the image data of the target object T5 and making the center point InCP of the inner cover bottom surface 19 located at the center point ImgC of the image Img are as follows: Acquire image data of the target object T5, and obtain the center point InCP (InCPx, InCPy) of the bottom surface 19 of the inner cover of the target object T5 from the image data; The intelligent analyzer controls the robot 3 to change the posture of the robot 3, and the camera 9 moves accordingly, so that the center point InCP (InCPx, InCPy) of the inner cover bottom surface 19 in the image Img captured by the camera 9 moves to the center point ImgC (ImgCx, ImgCy) of the image Img; when the robot 3 is translated, when the center point InCP (InCPx, InCPy) of the inner cover bottom surface 19 is located at the center point ImgC, the point RCP (x3, y3, z3, a3, b3, c3) changes to RCP (x4, y4, z4, a4, b4, c4), and the point OP (x3, y3, OPz3) changes to OP (x4, y4, OPz4); wherein, a4=a3, b4=b3, c4=c3; as shown in the attached figure Fig.10 As shown; Step 6.2.2, use formula (10) to calculate the height InCH from the center point InCP of the inner cover bottom surface 19 to the ground 5, InCH=int[0.5+StrH+BasH+dz+z] (10) (10) In the formula, z is the Z-direction coordinate value z4 of point RCP when point OP is OP(x4,y4,OPz4); Step 6.3, acquiring image data of the target object T5, so that the camera satisfies the fourth posture condition; the fourth posture condition refers to the third posture condition and the camera optical axis is perpendicular to the bottom surface of the inner cover; In this embodiment, the steps of acquiring the image data of the target object T5 so that the camera satisfies the fourth posture condition are as follows: Step 6.3.1, obtain the image data of the target object T5, and calculate the angle ω between the bottom surface 19 of the inner cover and the ground 5; The steps of acquiring the image data of the target object T5 and calculating the angle ω between the bottom surface 19 of the inner cover and the ground 5 are as follows: Acquire image data of the target object T5, and obtain a set ICOutline of contour points of the inner cover bottom surface 19 of the target object T5 from the image data; Find the point UpAC (UpACx, UpACy) from the set ICOutline, where UpACx=InCPx and is the contour point with the smallest y value in the set ICOutline; According to the set ICOutline and the point UpAC, the area UpA is obtained; UpA={G5(x,y)|G5(x,y)∈ICOutline∧G5(x,y)∈MinYA} Wherein, MinYA is a neighborhood consisting of pixels ≤Ry from the point UpAC. In this embodiment, Ry=int[Rx×H / W]; Find the point DnAC (DnACx, DnACy) from the set ICOutline, where DnACx=InCPx and is the contour point with the largest y value in the set ICOutline; According to the set ICOutline and the point DnAC, the region DnA is obtained; DnA={G6(x,y)|G6(x,y)∈ICOutline∧G6(x,y)∈MaxYA} Where MaxYA is the neighborhood consisting of pixels ≤Ry from the point DnAC; Measuring the mean depth UpDeepV1 from the camera 9 to the area UpA, and measuring the mean depth DnDeepV1 from the camera 9 to the area DnA; Formula (11) is used to calculate the average depth AveDeepV1 from the camera 9 to the center point InCP of the inner cover bottom surface 19, AveDeepV1=(UpDeepV1+DnDeepV1) / 2 (11) (11) In the formula, UpDeepV1 and DnDeepV1 are the measured values ​​when the point OP is OP(x4,y4,OPz4); Formula (12) is used to calculate the actual total height TotalHIC of the object corresponding to the image Img at the depth mean AveDeepV1, TotalHIC=2×AveDeepV1×tg(VA / 2) (12) Use formula (13) to calculate the Y-axis average value AveUpAy of area UpA: (13) Use formula (14) to calculate the Y-direction average value AveDnAy of area DnA: (14) The height UDh from the Y-direction average value AveDnAy of the region DnA to the Y-direction average value AveUpAy of the region UpA in the image Img is calculated using formula (15). UDh=TotalHIC×(AveDnAy-AveUpAy) / H (15) The depth UDw from the area DnA to the area UpA in the image Img is calculated using formula (16): UDw=UpDeepV1-DnDeepV1 (16) Use formula (17) to calculate the rotation amount σ, σ = tg -1 (UDw / UDh) (17) The angle ω between the bottom surface 19 of the inner cover and the ground 5 is calculated using formula (18): ω=90°-σ (18) Step 6.3.2, controlling the robot to rotate and move according to the angle between the bottom surface of the inner cover and the ground, so that the optical axis of the camera is perpendicular to the bottom surface of the inner cover; The steps of controlling the robot to rotate and move according to the angle between the bottom surface of the inner cover and the ground so that the optical axis of the camera is perpendicular to the bottom surface of the inner cover are as follows: The intelligent analyzer controls the robot 3 to rotate the Euler angle b in the posture of the robot 3 counterclockwise by σ; the counterclockwise refers to the counterclockwise rotation facing the positive Z axis; Use formula (19) to calculate the X-direction offset Wx and Z-direction offset Wz of point OP caused by the rotation of point RCP. Wx=dz×Sinσ Wz=dz×(1-Cosσ) (19) Use formula (20) to calculate the movement distance Kxz of point RCP when it moves from point RCP (x4, y4, z4, a4, b4, c4) to RCP (x5, y5, z5, a5, b5, c5), Kxz = AveDeepV1 × Sinσ (20) Use formula (21) to calculate the corresponding movement amount Kx in the X direction and the movement amount Kz in the Z direction. Kx=Kxz×Sinσ Kz=Kxz×Cosσ (21) The intelligent analyzer controls the robot 3 to move the point RCP in the positive X-axis direction by KWx and in the positive Z-axis direction by KWz; KWx=int[0.5+Kx-Wx], KWz=int[0.5+Kz+Wz]; After the robot 3 rotates and moves, the camera optical axis 15 is perpendicular to the inner cover bottom surface 19, and the center point InCP of the inner cover bottom surface 19 is located at the center point ImgC of the image Img; after the robot 3 rotates and translates, the point RCP (x4, y4, z4, a4, b4, c4) changes to RCP (x5, y5, z5, a5, b5, c5), and the point OP (x4, y4, OPz5) changes to OP (x5, y5, OPz5); wherein, x5=x4+KWx, y5=y4, z5=z4+KWz, a5=a4, b5=b4+σ, c5=c4; in this embodiment, because before this step, the camera 9 has satisfied the third posture condition, and this step makes the camera optical axis 15 perpendicular to the inner cover bottom surface 19, therefore, after this step is executed, the camera 9 satisfies the fourth posture condition; Step 6.4, acquiring image data of the target object T4, measuring the depth data of the outer cover opening 11 and measuring the actual size of the outer cover opening 11; The specific steps of acquiring the image data of the target object T4, measuring the depth data of the outer cover opening 11 and measuring the actual size of the outer cover opening 11 are as follows: Acquire image data of the target object T4, and obtain the area OCArea of ​​the outer cover opening 11 of the target object T4 from the image data; Obtain the minimum circumscribed rectangle OCRect of the area OCArea according to the boundary value of the area OCArea; According to the minimum circumscribed rectangle OCRect, four measurement areas RS5, RS6, RS7, and RS8 are obtained; wherein RS5, RS6, RS7, and RS8 are all areas with a width of rw2 and a height of rh2, and are all areas of the minimum circumscribed rectangle OCRect parallel to the outer cover opening 11, and the lower left corner of RS5 is the upper left corner of OCRect, the lower right corner of RS6 is the upper right corner of OCRect, the upper right corner of RS7 is the lower right corner of OCRect, and the upper left corner of RS8 is the lower left corner of OCRect; wherein rw2=int[OCRectW / M2], rh2=int[OCRectH / M2], OCRectW and OCRectH are the width and height of the rectangle OCRect, respectively, 4≤M2≤6; in this embodiment, M2=6; as shown in the attached Fig.13 As shown; Measure the mean depth value DeepV5 from camera 9 to area RS5, the mean depth value DeepV6 from camera 9 to area RS6, the mean depth value DeepV7 from camera 9 to area RS7, and the mean depth value DeepV8 from camera 9 to area RS8; Measure the actual size of the outer cover opening 11; the actual size of the outer cover opening 11 refers to the actual width OCRealW and the actual height OCRealH of the outer cover opening 11; The steps for measuring the actual size of the outer cover opening 11 are as follows: The average depth AveDeepOCV from the camera 9 to the outer cover opening 11 is calculated using formula (22): AveDeepOCV=(DeepV5+DeepV6+DeepV7+DeepV8) / 4 (22) Formula (23) is used to calculate the actual total width TotalWOCV and total height TotalHOCV of the object corresponding to the image Img when the depth is AveDeepOCV. TotalWOCV = 2 × AveDeepOCV × tg (LA / 2) TotalHOCV=2×AveDeepOCV×tg(VA / 2) (23) The actual width OCRealW and actual height OCRealH of the outer cover opening 11 are calculated using formula (24): OCRealW=int[0.5+TotalWOCV×OCRectW / W] OCRealH=int[0.5+TotalHOCV×OCRectH / H] (24) Step 6.5, acquiring image data of the target object T6, measuring the rotation angle of the inner cover knob and measuring the actual size of the inner cover knob; The steps for acquiring the image data of the target object T6, measuring the rotation angle of the inner cover knob, and measuring the actual size of the inner cover knob are as follows: Step 6.5.1, acquiring image data of the target object T6, and measuring the rotation angle KnobSlope of the inner cover knob 18; The steps of acquiring the image data of the target object T6 and measuring the rotation angle KnobSlope of the inner cover knob 18 are as follows: Acquire image data of the target object T6, and obtain a set of contour points KnobOutline of the inner cover knob 18 of the target object T6 from the image data; Obtain contour points on two long sides from the set KnobOutline; in this embodiment, the two long sides are Line5 and Line6; The slopes Slope1 and Slope2 of the contour points on the two long sides relative to the horizontal direction of the image Img are calculated respectively by the least square method; in this embodiment, Slope1 and Slope2 are the calculated slopes of Line5 and Line6 respectively; The rotation angle KnobSlope is calculated by formula (25): KnobSlope=(Slope1+Slope2) / 2 (25) (25) In the formula, the rotation angle KnobSlope is the slope relative to the horizontal direction of the image Img. Since the horizontal direction of the image Img is parallel to the XOY plane and the XOY plane is parallel to the ground 5, KnobSlope=tgθ, where θ is the angle between the inner cover knob 18 and the ground 5. Fig.13 As shown; Step 6.5.2, obtain the image data of the target object T6, and measure the actual size of the inner cover knob; The steps to obtain the image data of target T6 and measure the actual size of the inner cover knob are as follows: Acquire image data of the target object T6, and obtain the area KnobA of the inner cover knob 18 of the target object T6 and the center point KC (KCx, KCy) of the inner cover knob 18 of the target object T6 from the image data, as well as the set KnobOutline of the outline points of the inner cover knob 18 of the target object T6; Find the point MinDp (Xdp, Ydp) closest to the point KC (KCx, KCy) from the set KnobOutline; Use the following formula to calculate the distance KCMDv from point KC (KCx, KCy) to point MinDp (Xdp, Ydp), KCMDv=int[0.5+((KCx-Xdp) 2 +(KCy-Ydp) 2 ) 1 / 2 ] According to the area KnobA and the distance KCMDv, the edge area KEdgeA of the inner cover knob 18 is obtained. KEdgeA={G2(x,y)|G2(x,y)∈KnobA∧G2(x,y)∈EdgeA} Wherein, the region EdgeA is the region formed by the center point KC (KCx, KCy) of the inner cover knob 18 as the center and the points KCMDv as the radius; Measure the mean depth KEdgeDv from camera 9 to area KEdgeA; The point KC (KCx, KCy) is used to form the center area KCenterA of the inner cover knob 18; KCenterA={G3(x,y)|G3(x,y)∈KnobA∧G3(x,y)∈ARc} Wherein, ARc is a neighborhood consisting of Rc pixels whose distance from the center point KC (KCx, KCy) of the inner cover knob 18 is ≤, int[H / 66]≤Rc≤int[H / 40]; in this embodiment, Rc=int[H / 66]=int[800 / 66]=12; Measure the mean depth KCenterDv from camera 9 to area KCenterA; Use formula (26) to calculate the actual width KnobW of the inner cover knob 18: KnobW=int[0.5+2×(KEdgeDv 2 -KCenterDv 2 ) 1 / 2 ] (26) Use formula (27) to calculate the actual height KnobH of the inner cover knob 18: KnobH=int[0.5+AveDeepV2-KCenterDv] (27) Step 6.6: Acquire the image data of the target object T5, and measure the depth data of the bottom surface of the inner cover and the actual size of the bottom surface of the inner cover.

[0058] The steps of acquiring the image data of the target object T5 and measuring the depth data of the bottom surface of the inner cover and the actual size of the bottom surface of the inner cover are as follows: When the camera optical axis 15 is perpendicular to the bottom surface 19 of the inner cover, the average depth UpDeepV2 from the camera 9 to the area UpA and the average depth DnDeepV2 from the camera 9 to the area DnA are measured; as shown in the attached Fig.11 As shown; Formula (28) is used to calculate the depth value AveDeepV2 from the camera 9 to the center point InCP of the inner cover bottom surface 19, AveDeepV2=(UpDeepV2+DnDeepV2) / 2 (28) Acquire image data of the target object T5, and obtain a set ICOutline of contour points of the inner cover bottom surface 19 of the target object T5 from the image data; Find the point LtAC (LtACx, LtACy) from the set ICOutline; the point LtAC is the contour point with LtACy=InCPy and the smallest x value in the set ICOutline; According to the set ICOutline and the point LtAC, the area LtA is obtained; LtA={G7(x,y)|G7(x,y)∈ICOutline∧G7(x,y)∈LeftA} Among them, the area LeftA is a neighborhood consisting of points whose distance from point LtAC is ≤Rx pixels. int[H / 66]≤Rx≤int[H / 40]; in this embodiment, Rx=int[H / 66]=12; Find the point RtAC (RtACx, RtACy) from the set ICOutline; the point RtAC is the contour point with RtACy=InCPy and has the largest x value in the set ICOutline; According to the set ICOutline and the point RtAC, the region RtA is obtained; RtA={G8(x,y)|G8(x,y)∈ICOutline∧G8(x,y)∈RightA} Among them, the area RightA is a neighborhood consisting of points whose distance from the point RtAC is ≤Rx pixels; The left boundary value LeftPx and the right boundary value RightPx of the inner cover bottom surface 19 are calculated using formula (29). (29) According to the LeftPx and RightPx, RealDiam is calculated using formula (30): RealDiam=int[0.5+AveDeepV2×tg(LA / 2)×(RightPx-LeftPx) / W] (30) In formula (30), RealDiam is the actual diameter of the bottom surface 19 of the inner cover.

[0059] The working principle of the above technical solution is: when measuring different vehicle parameters, the intelligent analyzer identifies the target object based on the neural network, obtains the image data of the target object, and automatically controls the robot 3 to move to the optimal measurement position according to the image data of the target object to measure the vehicle parameters. The steps of identifying the target based on the neural network and obtaining the image data of the target are as follows: (1) intercepting a frame (color) image Img and a depth image DeepImg from the video image output by the camera 9; (2) compressing the image Img into a compressed image CompImg; (3) identifying each target from the compressed image CompImg based on the Yolov8 neural network (such as identifying target T1, target T2, target T3, or identifying target T4, target T5, target T6), and obtaining the minimum bounding rectangle that frames each target in the compressed image CompImg; (4) obtaining the minimum bounding rectangle of each target in the image Img by expanding the minimum bounding rectangle of each target in the compressed image; (5) obtaining the image data of each target from the area of ​​the minimum bounding rectangle of each target in the image Img using the maximum contour algorithm and the filling algorithm in image processing; the image data of the target includes the contour of the target, the area of ​​the target, the center point of the target area, and the depth value of each pixel in the target area; When measuring the distance between the maximum surface 16 and the rear of the vehicle 2, the intelligent analyzer calculates the midpoint MidP of the two points according to the center point MFC in the image data of the target object T3 and the point TLC of the target object T1, and automatically controls the robot 3 to move so that the point MidP is located at the center point ImgC of the image; after the movement, the intelligent analyzer measures the depth mean DeepMF from the camera 9 to the area MFCA, the depth mean DeepTL from the camera 9 to the area TLCA, and the depth mean DeepCar from the camera 9 to the area CarCA; if any of the measured depth mean values ​​is 0, The intelligent analyzer will display the equipment failure information on the display connected to it and prompt to check; if the measured depth mean is normal, the intelligent analyzer uses DeepMF and DeepTL to calculate the height TMh from the center point MFC of the maximum surface 16 to the point TLC in the rear part 2 of the vehicle, and calculates the width TMw from the point MFC to the point TLC based on the X-direction difference between the point MFC and the point TLC in the image and the depth mean DeepCar, and then calculates the actual distance FTDistance from the center of the maximum surface 16 to the rear part 2 of the vehicle based on TMw and TMh.

[0060] When measuring the data of the maximum surface 16, the intelligent analyzer automatically controls the movement of the robot 3 based on the center point of the maximum surface 16 so that the center point MFC of the maximum surface 16 is located at the center point ImgC of the image. After the movement, the depth mean DeepMF1 from the camera 9 to the area MFCA is measured, and DeepMF1 is used to calculate the total width TotalWMF1 and the total height TotalHMF1 of the actual object corresponding to the image Img captured by the camera 9. TotalWMF1 and TotalHMF1 are then used to calculate the actual width MFRealW and the actual height MFRect of the minimum circumscribed rectangle MFRect of the maximum surface 16. According to the installation data StrH and BasH of the robot 3, the posture data z of the robot 3, and the Z-direction deviation value dz between the point OP and the point RCP, the height MFH from the center of the maximum surface 16 to the ground 5 is calculated.

[0061] When measuring the fuel tank cap, the intelligent analyzer automatically controls the movement of the robot 3 based on the image data of the inner cover bottom surface 19 obtained so that the center point InCP of the inner cover bottom surface 19 is located at the center point ImgC of the image Img; then, the depth value UDw and the height value UDh from the camera 9 to the area UpA and the area DnA of the inner cover bottom surface 19 are calculated, and the angle ω between the inner cover bottom surface 19 and the ground 5 is calculated using UDw and UDh, and the actual height InCH from the center point of the inner cover bottom surface 19 to the ground 5 is calculated.

[0062] The intelligent analyzer controls the robot 3 to rotate and translate the posture of the robot 3 according to the angle ω, so that the camera optical axis 15 is perpendicular to the bottom surface 19 of the inner cover, and after the camera optical axis 15 is perpendicular to the bottom surface 19 of the inner cover, measures the depth average UpDeepV2 from the camera 9 to the area UpA and the depth average DnDeepV2 from the camera 9 to the area DnA, and measures and calculates the following data: the depth data DeepV5~DeepV8 of the outer cover opening 11, the actual width OCRealW and height OCRealH of the outer cover opening, the shape and rotation angle KnobSlope, actual depth KnobH, actual width KnobW of the inner cover knob 18, and the shape and actual diameter RealDiam of the inner cover bottom surface 19.

[0063] The beneficial effects of the above technical solution are: using the Yolov8 neural network to identify the target object (such as identifying the rear of the vehicle 2, the outer cover 10, the maximum surface 16, or identifying the outer cover opening 11, the bottom surface of the inner cover 19, and the inner cover knob 18), and obtaining the image data of the target object, automatically moving the camera 9 to the best measurement position based on the image data of the target object, and then accurately measuring the relevant parameters of the vehicle 1, so that the measurement has high accuracy and reliability. The intelligent analyzer can automatically capture images, identify targets, calculate parameters, and measure vehicle data, and can automatically control the movement and rotation of the robot 3 to make the camera 9 at a suitable measurement distance and the best measurement posture, and measure the vehicle fully automatically. After the measurement is completed, the measured vehicle data is uploaded to the server, which greatly improves the efficiency of the measurement; when measuring the distance between the rear of the vehicle 2 and the maximum surface 16, by changing the position and angle of the camera 9, the clarity and integrity of the target in the image can be ensured, and the distance from the maximum target 16 to the rear of the vehicle can be accurately measured; when measuring the maximum surface 1 6 data, the intelligent analyzer automatically controls the robot 3 to move so that the center point MFC of the maximum surface 16 is located at the center point ImgC of the image and the camera 9 is at a suitable distance from the vehicle 1, and then measures the data of the maximum surface 16, so that the collected data has a higher resolution and the measurement has a higher accuracy; when measuring the vehicle fuel tank cap, by automatically changing the position of the camera 9, the camera optical axis 15 is perpendicular to the bottom surface 19 of the inner cover and the center point of the bottom surface 19 of the inner cover is located at the center point of the image, ensuring that the measured outer cover opening 11, the bottom surface 19 of the inner cover, and the inner cover knob 18 have high accuracy and consistency; By using the preset posture conditions, it can be ensured that the camera 9 can collect as many pixels as possible and collect a complete, clear and consistent image of the target object when collecting the required target object, and can measure the shape and depth data and actual size of the target object at the best position; the technical solution can completely measure the following data: the distance from the center of the maximum surface 16 to the rear of the vehicle 2 and the height from the center of the maximum surface 16 to the ground 5, the shape and depth data and actual size of the maximum surface 16, the shape and depth data and actual size of the outer cover opening 11, the height from the center of the inner cover bottom surface 19 to the ground 5 and The angle between the bottom surface 19 of the inner cover and the ground 5, the actual diameter of the bottom surface 19 of the inner cover, the shape and depth data of the inner cover knob 18 and the actual diameter provide relatively complete data for automatic refueling. On the other hand, the neural network has a high recognition ability for different vehicles and fuel tank caps, and by measuring the shape and depth data of the target object as well as parameters such as the actual size, and saving the vehicle parameters required for automatic refueling, the system can directly extract these parameters when refueling the vehicle of the same model later, thus omitting many repeated measurements, improving the efficiency of refueling, and making the technical solution more convenient and practical.

[0064] In Example 1 and Example 2, the steps of generating the neural network are as follows: Step 7.1, set the number of taken pictures ImgsN, and set the atlas and processing sets to empty sets; where ImgsN is a positive integer ≥ 1000; Example 1: Set ImgsN=5000, and set the atlas PicSet and the processing set ProcSet to empty sets; the execution time of this step is 08:00 on January 6, 2024; Step 7.2, obtain the atlas; In this embodiment, a picture set PicSet is obtained; PicSet={P(j)|j=1,2,...,J}; wherein P(j) is a captured picture; j is a serial number of the captured picture in the order of time when the captured picture is added to the picture set PicSet; J is the total number of the captured pictures, and J is a positive integer; Example 1: This step starts at 08:00 on January 6, 2024. When this step is completed, J=21530. Example 2: This step starts at 08:29 on March 29, 2024. When this step is completed, J=26490. Step 7.3, using the atlas to obtain a processing set; In Example 1 and Example 2, the processing set ProcSet is obtained using the picture set PicSet; wherein, ProcSet={CompP(cj),CompPFile(cj)|cj=1,2,...,CJ}; CompP(cj) is the compressed image of image P(j), with width w and height h; CompPFile(cj) is the annotation file of the compressed image CompP(cj); CJ is the number of images in the processing set ProcSet, CJ=J; Step 7.4, splitting the processing set into a training set and a test set; In Example 1 and Example 2, the processing set ProcSet is split into a training set TrainSet and a test set TestSet; wherein TrainSet={CompP(q), CompPFile(q)|q=1,2,...,Q}, Q=0.6×CJ; TestSet={CompP(s),CompPFile(s)|s=1,2,...,S}, S=0.4×CJ; Step 7.5, training the neural network model; in this embodiment, the neural network model is the Yolov8 neural network model; The specific steps for training a neural network model are as follows: Step 7.5.1, set the initial value of q to 1, set TrainPn to the number of images in the training set TrainSet; q is the number of images and annotation files in a set of images and annotation files in the training set TrainSet; Example 1: Set q=1, and set TrainPn=12918; Example 2: Set q=1, and set TrainPn=15894; Step 7.5.2, the training neural network model program in the GPU server uses the picture CompP(q) and the annotation file CompPFile(q) in TrainSet to train the Yolov8 neural network model; Step 7.5.3, add 1 to the q; Step 7.5.4: If q>TrainPn, go to step 7.5.5; otherwise go to step 7.5.2; Step 7.5.5, end the process of training the neural network model; Step 7.6, test the neural network model; The specific steps for testing the neural network model are as follows: Step 7.6.1, set the initial value of s to 1, set TestPn to the number of images in the test set TestSet, and set the initial value of TestCn to 0; s is the number of images and annotation files in a set of images and annotation files in the test set TestSet; TestCn is the number of correct images tested; Example 1, set s=1, set TestPn=8612, set TestCn=0; Example 2, set s=1, set TestPn=10596, set TestCn=0; Step 7.6.2, the test neural network model program in the GPU server uses the image CompP(s) and the annotation file CompPFile(s) in the TestSet to test the Yolov8 neural network model, and records whether the image CompP(s) test is correct; the correct test of the image CompP(s) means that the Yolov8 neural network model's recognition of the image CompP(s) in the test satisfies the following five conditions at the same time: ① If there are Mc(s) targets in the image CompP(s), then these Mc(s) targets and only these Mc(s) targets are recognized, Mc(s)=2∨Mc(s)=3 (such as s=1 ② When there are two targets in the image CompP(1), these two targets and only these two targets are identified, Mc(1)=2); ② The rectangle RegRect(m) of the identified target (m) position satisfies the condition [Pn(RegRect(m)∩CompRect(m)) / Pn(CompRect(m))]×100%≥MinSv, 85%≤MinSv≤96%; in this embodiment, MinSv=95%; where Pn(AR) is the number of pixels in the area AR; CompRect(m) is the minimum bounding rectangle of the target (m) in the annotation file; m is the target according to the target (m) The number formed by the center points of the minimum enclosing rectangle RegRect(m) in increasing order; Mc(s) is the number of targets in CompPFile(s); (e.g. RegRect(m) satisfies the condition [Pn(RegRect(m)∩CompRect(m)) / Pn(CompRect(m))]×100%≥95%, m=1,2; target (1) is the outer cover 10 of target T2, target (2) is the maximum surface 16 of target T3); ③ The identifier of each target (m) in the image CompP(s) is the correct marker of the target (m), m=1,2,...,Mc(s); the identifier of the target is the marker that identifies the target (e.g., the identifiers of target T1, target T2, target T3, target T4, target T5, and target T6 are the correct markers Tail, OutCover, MaxFace, OutCoverOpening, InCover, and Knob, respectively); ④ accurately identify the shape of each target (m), m=1,2,...,Mc(s) (e.g., accurately identify that the shape of the outer cover 10 of target (1) in the image CompP(1) is circular, and the shape of the maximum surface 16 of target (2) is circular); ⑤ correctly identify the number of surfaces of each target (m), m=1,2,..., Mc(s) (For example, when correctly identifying the number of surfaces of the outer cover 10 of the target object (1) in the picture CompP(1) is 2, and the number of surfaces of the largest surface 16 of the target object (2) is 1); Recording whether the picture CompP(s) test is correct means that if the picture CompP(s) test is correct, then add 1 to the said TestCn;. Step 7.6.3, increment s by 1; Step 7.6.4, if s > TestPn, then go to Step 7.6.5; otherwise go to Step 7.6.2; Step 7.6.5, end the processing of testing the neural network model; Step 7.7, calculate the test accuracy rate Av; Av = (the number of correctly tested pictures / the number of pictures in the test set) × 100%; In Example 1, calculate the recognition accuracy rate Av = (TestCn / TestPn) × 100% = 95%; In Example 2, calculate the recognition accuracy rate Av = (TestCn / TestPn) × 100% = 99%; Step 7.8, if Av < SetAv, then go to Step 7.9; otherwise go to Step 7.10; where, 80% ≤ SetAv ≤ 100%; in this embodiment, SetAv = 98%; In Example 1, Av = 95% < SetAv = 98%, go to Step 7.9; In Example 2, Av = 99% ≥ SetAv = 98%, go to Step 7.10; Step 7.9, increase the value of ImgsN, and after increasing, go to Step 7.2; Increasing the value of ImgsN means that after increasing, ImgsN = the previous ImgsN × AddScale; where, 1.2 ≤ AddScale ≤ 2.0; In Example 1, increase the value of ImgsN, and after increasing, go to Step 7.2; After increasing, ImgsN = 6500; in this embodiment, AddScale = 1.3; The execution time of this step is 17:08 on March 28, 2024; Step 7.10, compile the neural network model into a neural network; The said neural network refers to a neural network that can identify the said target object; In Example 2, the GPU server uses compilation software to compile the Yolov8 neural network model into a Yolov8 neural network; The said Yolov8 neural network refers to a Yolov8 neural network that can identify the said target object; Step 7.11, write the said neural network into each intelligent analyzer in the said system.

[0065] In Example 2, the server takes out the Yolov8 neural network from the GPU server through the system switch, and writes the taken out Yolov8 neural network into each intelligent analyzer in the system through the system switch and the switch in each of the 24 measurement subsystems; the completion time of this step is 15:37 on May 2, 2024.

[0066] In Example 1 and Example 2, the steps of obtaining the atlas are as follows: The intelligent analyzer and the camera take pictures in atlas P1, atlas P2, atlas P3, and atlas P4 at different time periods within a day, and the number of pictures in the atlas must satisfy |P1|≥ImgsN∧|P2|≥ImgsN∧|P3|≥ImgsN∧|P4|≥ImgsN; wherein, the atlas P1 is a collection of pictures taken when the camera satisfies the first posture condition; the atlas P2 is a collection of pictures taken when the camera satisfies the second posture condition; the atlas P3 is a collection of pictures taken when the camera satisfies the third posture condition; the atlas P4 is a collection of pictures taken when the camera satisfies the fourth posture condition; In this embodiment, the intelligent analyzer and the camera take pictures in the atlas P1, the atlas P2, the atlas P3, and the atlas P4 at different time periods within one day, and the number of pictures in the atlas must satisfy |P1|≥ImgsN∧|P2|≥ImgsN∧|P3|≥ImgsN∧|P4|≥ImgsN; The intelligent analyzer uploads all the pictures that have been taken but not uploaded to the server at a preset time, and the server adds each picture received from all the uploaded pictures to the picture collection.

[0067] In this embodiment, the intelligent analyzer uploads all the pictures that have been taken but not uploaded to the server at the preset time of 23:00 every day, and the server adds each picture received from all the uploaded pictures to the atlas PicSet; after the step of completing the processing of obtaining the atlas is completed, the atlas PicSet={P(j)|j=1,2,...,J}; wherein P(j) is the taken picture; j is the number generated in the order of the time when the taken picture P(j) is added to the atlas PicSet; J is the number of pictures in the atlas PicSet, and J is a positive integer; Example 1, J=21530.

[0068] Example 2, J=26490.

[0069] In Example 1 and Example 2, the steps of using the atlas to obtain the processing set are as follows: Step 9.1, labeling processing; in this embodiment, the rectangle marking the position of the target object is a rectangle represented by four boundary values; the markers of the vehicle tail, outer cover, maximum surface, outer cover opening, inner cover bottom surface, and inner cover knob are Tail, OutCover, MaxFace, OutCoverOpening, InCover, and Knob respectively; The steps of the annotation processing are as follows: Step 9.1.1, set the initial value of j to StrJ, set ImgPn to the number of pictures in the atlas; j is the number of unlabeled pictures in the atlas PicSet; StrJ is the starting number of unlabeled pictures in the atlas PicSet; Example 1, set j=StrJ=1, set ImgPn=|PicSet|=21530; Example 2, set j=StrJ=21531, set ImgPn=|PicSet|=26490; Step 9.1.2, automatically annotate the image P(j) in the atlas PicSet using automatic annotation software to obtain the annotation file LabFile(j) of the image P(j); wherein the annotation file LabPFile(j) at least includes the file name, the file save path, the width and height of the image P(j) and the marking value of the color image, the number of identified targets M(j), and the following data of each target (m): a rectangle Rect(m) marking the position of the target (m), a marker Symbol(m) of the target (m), a marker Shape(m) of the shape of the target (m), and a marking value FsN(m) of the surface number of the target (m), m=1,2,...,M(j); wherein the rectangle marking the position of the target (m) is a rectangle represented by four boundary values; m is a number formed in increasing order of the center points of the rectangle Rect(m); In this embodiment, when j=1, the picture P(1) in the atlas PicSet is automatically annotated to obtain the annotation file LabFile(1) of the picture P(1); wherein the annotation file LabPFile(1) at least includes the file name, the file save path, the width and height of the picture P(1) and the marking value of the color picture, the number of identified targets M(1)=2, and the following data of each target (m): the rectangle Rect(m) marking the position of the target (m) (the rectangle Rect(2) marking the position of the target (2) is inaccurate), the marker Symbol(m) of the target (m) (Symbol(1)=OutCover, Symbol(2)=OutCove r (Symbol(2) should be = MaxFace)), the shape marker Shape(m) of the target (m) (Shape(1) = Shape(2) = CirType, CirType indicates that the shape of the target is a circle), the surface number marker value FsN(m) of the target (m) (FsN(1) = 1 (FsN(1) should be = 2), FsN(2) = 1), m = 1, 2; wherein, the rectangle marking the position of the target (m) is a rectangle represented by four boundary values; m is a number formed in increasing order of the center points of the rectangle Rect(m); the target T2 is numbered m = 1, and the target T3 is numbered m = 2; the number is a number formed in increasing order of the center points of the rectangle marking the position of the target; Step 9.1.3, adding the annotation file LabFile(j) to the atlas PicSet; In this embodiment, when j=1, the annotation file LabFile(1) is added to the atlas PicSet; Step 9.1.4, add 1 to the j; Step 9.1.5: If j>ImgPn, go to step 9.1.6; otherwise go to step 9.1.2; Step 9.1.6, end the marking process; In this embodiment, after the step of ending the labeling process is completed, PicSet={P(j), LabFile(j)|j=1,2,...,J}; Step 9.2, correction processing; in this embodiment, the modified labeling software is LabelImg; The specific steps for the correction process are as follows: Step 9.2.1, set the initial value of j to StrJ; j is the number of the uncorrected annotation file in the atlas PicSet; StrJ is the starting number of the uncorrected pictures in the atlas PicSet; In Example 1, set j=StrJ=1; Example 2, set j=StrJ=21531; Step 9.2.2, use the modification annotation software to judge the picture P(j) in the atlas PicSet and the annotation file LabFile(j), if it is judged that the annotation file LabFile(j) is a correct annotation file, go to step 9.2.5; otherwise, go to step 9.2.3; In this embodiment, when j=1, the modified labeling software LabelImg is used to judge the picture P(1) in the atlas PicSet and the labeling file LabFile(1), and it is judged that the labeling file LabFile(1) is an incorrect labeling file, and the process goes to step 9.2.3; In this embodiment, when j=2, the modified labeling software LabelImg is used to judge the picture P(2) in the atlas PicSet and the labeling file LabFile(2), and judge that the labeling file LabFile(2) is the correct labeling file, and then go to step 9.2.5; Step 9.2.3, use the labeling modification software LabelImg to modify the labeling file LabFile(j) to become the correct labeling file LabFile(j); the correct labeling file LabFile(j) means that the labeling of the labeling file LabFile(j) satisfies the following five conditions at the same time: (1) if there are M(j) targets in the image P(j), then label these M(j) targets and only label these M(j) targets, M(j)=2∨M(j)=3; (2) each Each Rect(m) is the smallest enclosing rectangle that encloses the target object (m), m=1,2,...,M(j); (3) Each Symbol(m) is the correct marker of the target object (m), m=1,2,...,M(j); (4) Each Shape(m) is the correct marker of the shape of the target object (m), m=1,2,...,M(j); (5) Each FsN(m) is the correct marker value of the surface quantity of the target object (m), m=1,2,...,M(j); In this embodiment, when j=1, the annotation file LabFile(1) is modified by the annotation modification software LabelImg to be the correct annotation file LabFile(1); the correct annotation file LabFile(1) means that the annotation of the annotation file LabFile(1) satisfies the following five conditions at the same time: (1) if there are two targets in the image P(j), then these two targets and only these two targets are marked, M(1)=2; (2) each Rect(m) is the minimum circumscribed rectangle that encloses the target (m), m=1,2; (3) Symbol(1)= OutCover, Symbol(2)=MaxFace are both correct markers for the target object (m), m=1,2; (4) Shape(1)=Shape(2)= CirType is the correct identifier of the shape of the target object (m), m=1,2; (5) FsN(1)=2, FsN(2)=1 are the correct identifiers of the number of surfaces of the target object (m), m=1,2; Step 9.2.4, saving the corrected annotation file LabFile(j) to the original annotation file LabFile(j) in the atlas PicSet; In this embodiment, when j=1, the corrected annotation file LabFile(1) is saved to the original annotation file LabFile(1) in the atlas PicSet; Step 9.2.5, add 1 to the j; Step 9.2.6: If j>ImgPn, go to step 9.2.7; otherwise go to step 9.2.2; Step 9.2.7, ending the correction process; In this embodiment, after the step of ending the correction process is completed, the atlas PicSet={P(j), LabFile(j)|j=1,2,...,J}; Step 9.3: Compression processing.

[0070] The specific steps of compression processing are as follows: Step 9.3.1, set the initial value of j to StrJ, set the initial value of cj to CStrJ; j is the number of the uncompressed picture in the atlas PicSet; StrJ is the starting number of the uncompressed picture in the atlas PicSet; cj is the number of the picture in the processing set ProcSet; CStrJ is the starting number of the picture in the processing set ProcSet; In Example 1, set j=StrJ=1, and set cj=CStrJ=1; Example 2, set j=StrJ=21531, set cj=CStrJ=21531; Step 9.3.2, compress the image P(j) into a compressed image CompP(cj) with a width of w and a height of h; In this embodiment, when j=1, cj=1, the picture P(1) is compressed into a compressed picture CompP(1) with a width of w and a height of h; Step 9.3.3, obtain the compressed image annotation file CompPFile(cj) according to the annotation file LabFile(j) of the image P(j); In this embodiment, when j=1 and cj=1, a labeling file CompPFile(1) of a compressed image CompP(1) is obtained according to the labeling file LabPFile(1) of the image P(1); wherein the labeling file CompPFile(1) at least includes the file name, the file save path, the width and height of the image CompP(1) and the label value of the color image, the number of objects Mc(1) (=M(1)=2) in the image CompP(1), and the following data of each object (m): the minimum enclosing rectangle CompRect(m) of the object (m), the symbol CompSym(m) (=Symbol(m)) of the object (m), The shape marker of the object (m) CompShape(m) (=Shape(m)); the surface number marker value CompFn(m) (=FsN(m)), m=1,2; wherein the X-boundary values ​​of CompRect(m) are all obtained by compressing the X-boundary values ​​of Rect(m) according to the compression ratio w / W(1), and the Y-boundary values ​​of CompRect(m) are all obtained by compressing the Y-boundary values ​​of Rect(m) according to the compression ratio h / H(1); the X-boundary values ​​refer to the left and right boundary values; the Y-boundary values ​​refer to the upper and lower boundary values; W(1) and H(1) are the width and height of the image P(1) respectively; Step 9.3.4, adding the compressed image CompP(cj) to the processing set ProcSet, and adding the annotation file CompPFile(cj) of the compressed image to the processing set ProcSet; In this embodiment, when cj=1, the compressed image CompP(1) is added to the processing set ProcSet, and the annotation file CompPFile(1) of the compressed image is added to the processing set ProcSet; ProcSet={CompP(1), CompPFile(1)}; Step 9.3.5, add 1 to the j, and add 1 to the cj; Step 9.3.6: If j>ImgPn, go to step 9.3.7; otherwise go to step 9.3.2; Step 9.3.7: End the compression process.

[0071] In this embodiment, after the step of ending the compression process is completed, ProcSet={CompP(cj), CompPFile(cj)|cj=1,2,...,CJ}; wherein CJ=J; Example 1, CJ=21530.

[0072] Example 2, CJ=26490.

[0073] The beneficial effect of the above technical solution is that by taking pictures containing target objects at different time periods and under different shooting conditions, pictures containing target objects in various scenes and environments are obtained. At the same time, some restrictions are imposed on the shooting conditions, such as the direction in which the camera 9 collects images, the direction of the camera optical axis 15, the distance from the camera 9 to the shooting vehicle 1, the target objects contained in the taken pictures, etc. These restrictions can ensure the quality of the taken pictures and the consistency of the shooting; after obtaining enough pictures, these pictures are annotated, corrected and compressed to obtain a processing set containing compressed pictures and corresponding annotation files; then the processing set is split into a training set and a test set; then the compressed pictures and corresponding annotation files in the training set are used to train the neural network model, so that the neural network learns a wider, richer, More detailed and more practical features can improve the generalization and robustness of the neural network. After training, the neural network model is tested with compressed images in the test set and the corresponding annotation files. If the test accuracy does not reach the set accuracy, the number of captured images needs to be increased, and the newly added images need to be annotated, corrected and compressed. The enlarged processing set is then split into a training set and a test set again. The neural network model is then trained again and tested again. If the test accuracy reaches the set accuracy, the neural network model is compiled into a neural network for identifying the target object, and the neural network for identifying the target object is written into each intelligent analyzer in the system. The neural network obtained in this way has a stronger recognition ability and a higher recognition accuracy and precision.

[0074] Embodiments 1 to 5, a vehicle parameter measurement system based on three-dimensional vision, the system is used to implement any method of the vehicle parameter measurement method based on three-dimensional vision, the system includes N measurement subsystems, a server, a GPU server, a network device, a system switch and a system power supply; the server is connected to the system switch, and is connected to the GPU server and the network device through the system switch, and is connected to the external network through the system switch and the network device, and is connected to the switch in each of the N measurement subsystems through the system switch and is connected to the intelligent analyzer in the measurement subsystem through the switch; the system power supply is connected to the system switch and the network device, and provides the required DC power for the system switch and the network device; wherein N is a positive integer; in this embodiment, N=24; as shown in the attached Figure 1 As shown; The server is a server that meets the use requirements and has a hard disk, an input device, and a display. The hard disk of the server is installed with a Windows operating system, a database MySQL, and a first application software. The server is used to manage each measurement subsystem in the system and process the information and data of each measurement subsystem; the first application software at least includes database management software, automatic marking software, modification marking software, splitting program, and network communication program; The GPU server is a GPU server that meets the use requirements and has a hard disk on which a Linux operating system and a second application software are installed, an input device and a display, and is used for training neural network models, testing neural network models, compiling neural network models, deep learning, and video encoding and decoding. The second application software includes at least the following software and programs: a neural network model training program, a neural network model testing program, a compiling software, and a network communication program; The network device is one of the following devices or a combination of the following devices: a switch, a fiber optic transceiver, a switch and a fiber optic transceiver, a wireless network device, a switch and a wireless network device, and the wireless network device refers to a wireless router, a wireless bridge, and a wireless wifi device; in this embodiment, the network device is a switch and a fiber optic transceiver; the system power supply is connected to the system switch and the network device, and provides the required DC power for the system switch and the network device; The servers, GPU servers, network equipment, system switches, and system power supplies are all placed indoors.

[0075] In Examples 1 to 5, the measurement subsystem includes an intelligent analyzer, a switch, an input device, a camera 9, a fixing, a robot 3, a display and a power supply; the intelligent analyzer is connected to the switch, the input device, the camera 9, and the display, and is connected to the robot 3 and the system switch through the switch, and is connected to the server through the switch and the system switch; the camera 9 is fixed to the robot 3 with the fixing, and the camera 9 is a depth camera; the power supply is connected to the intelligent analyzer, the switch, the camera 9, and the display, and provides the required DC power for the intelligent analyzer, the switch, the camera 9, and the display, and the power cord provided to the camera 9 is an explosion-proof power cord; as shown in the attached Figure 2 As shown; The intelligent analyzer is an intelligent analyzer that meets the use requirements and has a hard disk and a Linux operating system and a third application program installed on the hard disk. The third application program includes at least the following programs: a picture shooting program, an image processing program, a Yolov8 neural network program, a depth measurement program, a robot control 3 program, and a network communication program; the image processing program includes at least an image capture program, an image compression program, an image expansion program, a color image conversion program to a grayscale image program, a gradient algorithm program, an edge point acquisition program, a filtering algorithm program, a maximum contour algorithm program, a target object contour acquisition program, a filling algorithm program, a target object minimum circumscribed rectangle acquisition program, a region center point acquisition program, and a pixel depth value acquisition program in the region; the Yolov8 neural network can at least identify the vehicle rear 2, the fuel tank cover 10, the maximum surface 16, the outer cover opening 11, the inner cover bottom 19, and the inner cover knob 18; The program for controlling the robot 3 is a program for changing the posture of the robot 3 by controlling the robot 3; the posture of the robot 3 refers to the spatial position and rotation amount of the end center point RCP (x, y, z, a, b, c) of the robot 3 in the base coordinate system of the robot 3, the spatial position is the coordinate values ​​x, y, z in the X, Y and Z directions, and the rotation amount is the rotation amount described by the Euler angles a, b, c with the point RCP as the center of the circle; the base coordinate system of the robot 3 refers to the center point BasCP of the bottom surface of the base 6 of the robot 3 as the origin O, the line from the origin O to the left and parallel to the long side of the refueling island 4 is the Y positive axis, the line from the origin O to the front is the X positive axis and the X axis is perpendicular to the Y axis, and the XOY plane is parallel to the ground 5, and the Z positive axis is from the origin O to the vertical upward and the Z axis is perpendicular to the XOY plane; the left side and the front side refer to the center point BasCP of the robot 3 facing the back and located on the left and front of the point BasCP; The camera 9 outputs its color video and depth video to the intelligent analyzer through its USB interface with an explosion-proof USB extension cable. The camera 9 has a color resolution ≥ 920,000 pixels, a depth resolution ≥ 920,000 pixels, a measurement depth range of 200mm-2000mm, and is fixed to the end of the robot 3 with a fixing part. It is a depth camera with an explosion-proof grade of ExdIIBT4. When the depth camera 9 is installed, the deviation (dx, dy, dz) between the point OP (OPx, OPy, OPz) of the camera 9 and the center point RCP (x, y, z, a, b, c) of the end of the robot 3 in the base coordinate system of the robot 3 is determined; in this embodiment, the X-direction deviation value dx=OPx-x=0, and the Y-direction deviation value dy=OP y-y=0, Z-direction deviation value dz=OPz-z=100mm; the fixing part is composed of a fixing plate 8, a fixing bar 13 and three groups of fixing screws, the first group of four screws pass through the through holes on the fixing plate 8 to fix it on the plane of the end of the robot 3, the second group of two screws 14 pass through the through holes on one side of the fixing bar 13 to fix the fixing bar 13 to the side of the fixing plate 8, and the third group of two screws 12 pass through the through holes on the other side of the fixing bar 13 to fix the camera 9 on the other side of the fixing bar 13; in this embodiment, the installation of the camera 9 must also meet the following conditions: when the Euler angle a=b=c=0 in the point RCP(x,y,z,a,b,c), the horizontal direction of the camera 9 collecting the image is parallel to the XOY plane, and the camera optical axis 15 is parallel to the X-axis; as shown in the attached figure Figure 5 As shown; The explosion-proof grade of the robot 3 in the measurement subsystem is ExdIIBT4. The robot 3 is composed of a robot body including an integrated controller, a robot teaching pendant, a power adapter and cables, and a network interface; the power adapter and cables provide the power required by the robot 3; the network interface is the interface for the robot 3 to communicate with other devices over the network; in this embodiment, the intelligent analyzer uses a network cable to connect the network interface of the robot 3 through the network interface thereon to control the robot 3; the robot teaching pendant is a controller for manually controlling the position and posture of the robot 3; the robot 3 is a robot that meets the use requirements and is equipped with a camera 9 at its end; as shown in the attached Figure 4 With attached Figure 5 As shown; The intelligent analyzer, display, input device, switch and power supply are all placed indoors.

[0076] In Embodiments 1 to 5, the measurement subsystem further comprises an intercom device, the explosion-proof grade of the intercom device is ExdIIBT4, the intercom device is used to issue a position adjustment instruction of the vehicle 1, the power supply is connected to the intercom device and provides the required DC power to the intercom device, and the power cord provided to the intercom device is an explosion-proof power cord; as shown in the attached Figure 2 shown.

[0077] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, substitute and modify the above embodiments within the scope of the invention without departing from the principles and purpose of the present invention. However, all such changes should fall within the scope of protection of the claims of the present invention.

Claims

1. A vehicle parameter measurement method based on three-dimensional vision, wherein the method is implemented by a vehicle parameter measurement system based on three-dimensional vision, characterized in that: The implementation steps of the method described are as follows: Step A, generating a neural network; the neural network is a neural network that can identify a target object; Step B, measurement processing; The steps of the measurement process are as follows: Keep the vehicle within the measurement range; Acquire image data of the target object and measure the actual distance from the center of the largest surface to the rear of the vehicle; Acquire image data of the target object, measure the shape and depth data of the largest surface and the actual size; Acquire the image data of the target object and measure the fuel tank cap data; Upload the measured vehicle data to the server.

2. The measuring method according to claim 1, characterized in that: The steps of placing the vehicle within the measurement range are as follows: Step 2.1: If the vehicle meets the first preset condition, go to step 2.3; otherwise, go to step 2.2; Step 2.2, issuing an instruction for adjusting the vehicle position; Step 2.3: End the processing of this step.

3. The measuring method according to claim 1, characterized in that: The steps of obtaining the image data of the target object are as follows: The intelligent analyzer captures a frame of image and a frame of depth image from the video image output by the camera; the image refers to a color image, wherein the width of the image and the depth image are both W and the height are both H; Compressing the image into a compressed image; the compressed image is an image with a width of w and a height of h; the w and h are the width and height of the image processed by the neural network; Using the neural network to identify each target object from the compressed image, and obtaining the minimum bounding rectangle of each target object in the image based on the minimum bounding rectangle that frames each target object in the compressed image; Using an image processing algorithm, image data of each target object is obtained from the area of ​​the smallest circumscribed rectangle of each target object in the image; the image data of the target object includes the outline of the target object, the area of ​​the target object, the center point of the target object area, and the depth value of each pixel point in the target object area; the targets are target T1, target T2, target T3, target T4, target T5, and target T6; wherein, target T1 is the rear of the vehicle, target T2 is the outer cover, target T3 is the maximum surface, and The target object T4 refers to the outer cover opening, the target object T5 refers to the inner cover bottom surface, and the target object T6 refers to the inner cover knob; the vehicle tail portion refers to the outermost contour portion of the vehicle tail portion; the outer cover refers to the outer cover of the fuel tank and is in a fully closed state; the maximum surface refers to the largest and smoothest surface of the outer cover; the outer cover opening refers to the outer cover opening of the fuel tank and is in a fully open state; the inner cover bottom surface refers to the bottom surface of the inner cover of the fuel tank; the inner cover knob refers to the knob portion on the bottom surface of the inner cover.

4. The measuring method according to claim 1, characterized in that: The steps of obtaining the image data of the target object and measuring the actual distance from the center of the largest surface to the rear of the vehicle are as follows: Obtain the image data of target T1, target T2, and target T3, and make the camera satisfy the first pose condition; the first pose condition means that the image contains target T1, target T2, and target T3, and the distance from the camera to the vehicle is within a first preset range; the first preset range is the range within which the neural network can recognize target T1, target T2, and target T3; Obtain the image data of target T1 and target T3, and measure the actual distance from the center of the largest surface to the rear of the vehicle.

5. The measuring method according to claim 1, characterized in that: The steps of obtaining the image data of the target, measuring the shape and depth data of the largest surface and the actual size are as follows: Obtain the image data of target T2 and target T3, and make the camera satisfy the second pose condition; the second pose condition means that the image contains target T2 and target T3, and the distance from the camera to the vehicle is within a second preset range; the second preset range is the range within which the neural network can recognize target T2 and target T3; Obtain the image data of target T3, and measure the depth data of the largest surface and the actual size of the largest surface.

6. The measuring method according to claim 1, characterized in that: The steps of obtaining the image data of the target and measuring the fuel tank cap data are as follows: Obtain the image data of target T5, and make the camera satisfy the third pose condition; the third pose condition means that the image contains target T4, target T5, and target T6, and the distance from the camera to the vehicle is within a second preset range; the second preset range is the range within which the neural network can recognize target T4, target T5, and target T6; Measure the height from the center point of the bottom surface of the inner cover to the ground. Obtain the image data of target T5, and make the camera satisfy the fourth pose condition; the fourth pose condition means the third pose condition and the camera optical axis is perpendicular to the bottom surface of the inner cover; Obtain the image data of target T4, and measure the depth data of the outer cover opening and the actual size of the outer cover opening. Obtain the image data of target T6, and measure the rotation angle of the inner cover knob and the actual size of the inner cover knob. Obtain the image data of target T5, and measure the depth data of the bottom surface of the inner cover and the actual size of the bottom surface of the inner cover.

7. The measuring method according to claim 1, characterized in that:The steps of generating the neural network are as follows: Step 7.1: Set the number of captured pictures ImgsN, and set the atlas set and the processing set to be empty sets; Step 7.2: Obtain the atlas; the atlas is the atlas containing the target; Step 7.3: Obtain the processing set using the atlas; Step 7.4: Split the processing set into a training set and a test set; Step 7.5: Train the neural network model using the training set; Step 7.6: Test the neural network model using the test set, and count the number of pictures correctly tested; Step 7.7: Calculate the test accuracy rate Av; Av = (the number of pictures correctly tested / the number of pictures in the test set) × 100%; Step 7.8: If Av < SetAv, then go to Step 7.9; Otherwise, go to step 7.10; where 80%≤SetAv≤100%; Step 7.9, increase the ImgsN value, and then go to step 7.2; increasing the ImgsN value means that the ImgsN after increase = ImgsN before increase × AddScale; where 1.2≤AddScale≤2.0; Step 7.10, compiling the neural network model into a neural network; the neural network refers to a neural network that can recognize the target object; Step 7.11, write the neural network into each intelligent analyzer in the system.

8. The measuring method according to claim 7, characterized in that: The steps of obtaining the atlas are as follows: The intelligent analyzer and the camera take pictures in atlas P1, atlas P2, atlas P3, and atlas P4 at different time periods within a day, and the number of pictures in the atlas must satisfy |P1|≥ImgsN∧|P2|≥ImgsN∧|P3|≥ImgsN∧|P4|≥ImgsN; wherein, the atlas P1 is a collection of pictures taken when the camera satisfies the first posture condition; the atlas P2 is a collection of pictures taken when the camera satisfies the second posture condition; the atlas P3 is a collection of pictures taken when the camera satisfies the third posture condition; the atlas P4 is a collection of pictures taken when the camera satisfies the fourth posture condition; The intelligent analyzer uploads all the pictures that have been taken but not uploaded to the server at a preset time, and the server adds each picture received from all the uploaded pictures to the picture collection.

9. The measuring method according to claim 7, characterized in that: The steps of obtaining the processing set using the atlas are as follows: Annotation processing: Annotation processing refers to using automatic annotation software to automatically annotate each unannotated picture in the atlas to obtain an annotation file of the picture; Correction processing; Correction processing refers to correcting each incorrect annotation file in the atlas into a correct annotation file by using modification annotation software; Compression processing; The compression process refers to compressing each uncompressed picture in the atlas to obtain a compressed picture of the picture, and obtaining a labeling file of the compressed picture according to the labeling file of the picture.

10. A vehicle parameter measurement system based on three-dimensional vision, the system is used to implement any method of claim 1 to claim 9, characterized in that: The system includes N measurement subsystems, a server, a GPU server, a network device, a system switch and a system power supply; wherein N is a positive integer.

11. The vehicle parameter measurement system according to claim 10, characterized in that: The measurement subsystem includes an intelligent analyzer, a switch, an input device, a camera, a fixture, a robot, a display and a power supply.

12. The vehicle parameter measurement system according to claim 11, characterized in that: The measurement subsystem also includes an intercom device, which is used to issue a voice message for adjusting the vehicle position.

Images