TWI645372B - Image calibration system and image calibration method - Google Patents

Abstract

The invention relates to an image correction system and an image correction method, which use a pattern plate to accommodate a target object within the range of the pattern plate. Wherein, the pattern plate is encoded by a virtual random array, and has a plurality of first pattern units and a plurality of second pattern units; the first pattern units are arranged adjacent to each other in a matrix manner, and each of the first pattern units presents a first One of a color and a second color, and the adjacent first pattern units are different in color from each other; the second pattern units are respectively disposed in the first pattern unit. An image of the camera disposed on the target object facing the pattern plate can recognize its orientation relative to the pattern plate to facilitate splicing into a ring image.

Description

Image correction system and image correction method

The invention relates to an image correction system and an image correction method, in particular to a correction system and a correction method for a scene image.

In order to make the driver more aware of the surrounding conditions of the vehicle during driving, today's vehicles have gradually listed the Area View Monitoring (AVM) as standard equipment. The surround image system consists of multiple (for example, four or more) cameras, which are integrated into a bird's-eye view of the bird's eye view by integrating the images captured by the camera for reference by the driver.

However, the positions and angles of the cameras are different. In order to display better image quality, the correction and splicing techniques of the images obtained by each camera are very important. Not only need to be corrected when the new car is installed with the surround image system. If it is later due to an accident collision or a module update, it will also need to be recalibrated. Positioning correction between cameras requires a common coordinate system, such as the World Coordinates System (WCS), so that each image can be integrated, so each camera needs to obtain the calibration system from the camera coordinate system to the world coordinate system. Convert information to identify the position of the lens. If there is an error in the conversion information, the mosaic bird's-eye view image quality will be poor. In order to simplify the calibration procedure and ensure image stitching results, conventional techniques generally configure the cameras in a symmetrical or fixed position with each other, such as cameras located at four positions of the front, rear, left, and right of a general vehicle, and define them. It must be symmetrical or set at a specific position; however, when applied to a target other than a general motor vehicle such as a truck or an airplane, there is inherent difficulty in limiting the camera to be symmetrical only.

A conventional method for correcting a panoramic image, as shown in FIG. 1, is to first arrange a checkerboard 12 around the vehicle 11. After the camera (not shown) on the vehicle 11 captures the image, the image is corrected using the feature points of the checkerboard boundary selected in each image. Considering that the number of cameras cannot be too much, and finally it is necessary to splicing into a panoramic image, the camera usually uses a fisheye lens or a wide-angle lens. As shown in Fig. 2, the image taken by the fisheye lens of one of the cameras 111 is 113. It will be deformed, especially at the boundary. The image correction system needs to convert it into a flat image, and use feature points as the basis for image stitching. In other words, there must be overlapping features between the images 113 captured by two adjacent cameras to provide a basis for converting the two into planar images. Conceivably, the overlapping feature is usually located in the peripheral region of the image 113, but the image 113 converted by the camera 111 using the fisheye lens or the wide-angle lens has a relatively low resolution and a large degree of deformation in the peripheral region, and thus the feature point is judged. And the subsequent mosaic of the surrounding image has certain difficulty.

Therefore, the conventional scene image correction relies on a large number of manual methods for subsequent processing, especially the professional man with considerable experience, and devoted considerable time to image correction and stitching. In particular, in the initial checkerboard layout, accurate measurement is required, so it is usually done manually; and the subsequent image capture, correction alignment and splicing adjustments still require a large number of Manpower and time are judged and parameterized to ensure correction and splicing results. The result of relying heavily on professional input results in problems such as excessive man-hours, labor costs, increased costs, and low efficiency. The quality and experience of manpower will directly affect the quality of calibration and splicing. These factors have all contributed to the popularization of panoramic imaging systems. Obstruction.

In view of this, the development of a fully automatic and accurate panoramic image correction system and calibration method has great demand and development potential in this industry.

One of the main objects of the present invention is to provide an image correction system and an image correction method, which can automatically perform correction and generate a correct scene image, which can greatly reduce manpower input and obtain more accurate and robust results.

The present invention adopts a Pseudo Random Array (PRA) coded pattern board having a considerable number of accurate correction features thereon, including a plurality of first pattern elements in a checkerboard pattern as a background, and respectively set The plurality of second pattern units in the first pattern unit as feature points can ensure that the partial patterns captured by the camera of any orientation or posture are unique, and can not only determine the correction data of the camera, but also provide A plurality of confirmed evidences can obtain more accurate and robust results when the spatial conversion information of the image is obtained (for example, calculating the Homography matrix), so that the calibration result of the camera is more correct.

Because the pattern plate of the present invention has a virtual random array (PRA) code, the image captured by each lens has its uniqueness, so the position and angle of the object placed on the pattern plate are not limited, and the display of the camera is not It is necessary to limit its symmetrical configuration or specific position, and its position and orientation can be recognized from the image, and the images taken by adjacent cameras do not need to have overlapping regions. Of course, it is no longer necessary to correct the feature points of the overlapping regions. . In addition, the technology of the present invention is not limited to the planar images exemplified above, and more lenses may be disposed on the front, back, left, right, up, and down of the target, and finally the spherical panoramic image may be spliced. Used in aircraft or drones to provide obstacle avoidance capabilities.

The calibration system and the calibration method of the present invention can automatically and accurately obtain the corrected feature points by using the central high-resolution area of the image captured by each camera, and thereby perform image mosaic to generate a panoramic image.

In the correction processing program, the position of the target object relative to the pattern plate is not limited, and the orientation of the upper lens is not limited, as long as the image is captured substantially toward the pattern plate, the width, length, and height of the object. There is no impact, the system can automatically automatically push back the position of the camera and the corresponding vehicle size and category by correcting the known size of the pattern.

In order to achieve the foregoing objective, the present invention provides an image correction system for acquiring a surrounding image of a target object, the target object having at least one lens, the image correction system comprising: a pattern plate, a transmission module, A correction module and an operation module. The pattern plate can accommodate the object within the range of the pattern plate, the pattern plate has a plurality of first pattern units and a plurality of second pattern units; the first pattern units are arranged adjacent to each other in a matrix manner, each The first pattern unit presents one of a first color and a second color, and the adjacent first pattern units are different in color from each other; the second pattern unit is respectively disposed on the first pattern unit in. The transmission module receives a partial image of the lens toward the pattern plate; the correction module is coupled to the transmission module to capture and correct a matrix image from the partial image; the computing module is Recognizing a plurality of the first pattern units and the plurality of second pattern units on a matrix image, and calculating the lens relative to the pattern plate according to the first pattern unit and the second pattern unit One orientation.

The orientation includes a positional parameter and a viewing angle parameter of the lens relative to the pattern plate. The second pattern unit of the pattern board is arranged by a Pseudo Random Array (PRA) code, and preferably, the second pattern unit has a shape feature and a color feature, wherein the color feature Is one of the first color and the second color. The calibration module captures the central portion of the partial image and performs leveling correction to obtain the matrix image.

The lens system includes a plurality of lenses, and is preferably a fisheye lens or a wide-angle lens. The computing module splices the partial images of each lens into the ring around the target according to the orientation of each lens. Scene image.

The present invention also provides an image correction method for obtaining a scene image around a target object, the image correction method comprising the steps of: providing a pattern plate made by a virtual random array code; setting the object object to the pattern a plurality of lenses are disposed on the object; a partial image is obtained through each lens; a matrix image is obtained and corrected from the partial image; and the lens is calculated according to the matrix image One orientation on the pattern plate.

The pattern plate has a plurality of first pattern units and a plurality of second pattern units, the first pattern units are arranged adjacent to each other in a matrix manner, and each of the first pattern units presents a first color and a second One of the colors, and the adjacent first pattern units are different in color from each other, and the second pattern units are respectively disposed in the first pattern unit; wherein, the lens is calculated relative to the The step of one orientation of the pattern plate is to identify a plurality of the first pattern unit and the plurality of the second pattern units on the matrix image, and according to the first pattern unit and the second pattern unit Calculating a positional parameter and a viewing angle parameter of each of the lenses relative to the pattern plate.

The step of extracting and correcting a matrix image from the partial image captures a central portion of the partial image and performs leveling correction to obtain the matrix image.

The second pattern unit has at least one shape feature and one color feature. Wherein the color feature is one of the first color and the second color.

The calibration method of this embodiment further includes a step of splicing each of the partial images of each of the lenses into the panoramic image around the target according to the orientation of each of the lenses.

The above objects, technical features and advantages will be more apparent from the following description.

In the following, examples will be provided to explain in detail embodiments of the invention. The advantages and effects of the present invention will be more apparent by the disclosure of the present invention. The drawings appended hereto are simplified and are used for illustration. The number, shape and size of the components shown in the drawings can be modified as the case may be, and the configuration of the components may be more complicated. Other variations and modifications can be made without departing from the spirit and scope of the invention as defined in the invention.

Please refer to FIG. 3 and FIG. 4 together. FIG. 3 is a block diagram of the image correction system of the present invention, and FIG. 4 is a schematic diagram of a pattern plate of the image correction system of the present invention. The image correction system 3 of the present invention comprises a pattern board 30, a transmission module 31, a correction module 33 and an operation module 35. As shown in FIG. 4, a target object 50 (the present invention is exemplified by a vehicle) is firstly randomly placed on a pattern plate 30 having a Pseudo Random Array (PRA) code, and the pattern plate 30 is arbitrarily disposed. The size is sufficient to accommodate the target 50 within the range of the pattern sheet 30, and the target 50 is provided with at least one lens 51 (not shown in FIG. 4). The lens 51 captures a partial image in a field of view toward the pattern board 30 in its own orientation or posture, and transmits it to the transmission module 31, and the correction module 33 is connected to the transmission module 31 to The image is captured and corrected, and then the computing module 35 calculates an orientation of the lens 51 relative to the pattern plate 30 based on the corrected image, as described in detail below.

The following is an illustration of identifying the position of a 5 x 5 bit array in a virtual random array (PRA), which is based on the orientation of the camera in four directions and the bit array (even columns) at coordinates (13, 18). For example, starting with 18, please refer to Figure 5.

Since the even columns (0, 2, ..., 60) of the virtual random array (PRA) are the same as the original virtual random number column (PRS), this feature is used to find out the bit array interlaced or interlaced, and its 5 bits. The same direction (that is, an even column) is used to represent the horizontal direction of the virtual random array (PRA) (ie, the i-axis direction), and the original virtual random number column (PRS) is determined according to the camera origin position (01111). And 01111 is at the 13th position (a) of the original virtual random number sequence (PRS) 10000100101100111110001101110101, so the i coordinate is (i = a = 13). The (13, 18) bit array seen by the front, rear, left, and right lenses is shown in Figure 5.

Next, the complementary virtual random number sequence (PRS) of the odd-numbered columns of the partition wall is (10110), and it is at the fourth position of the complementary virtual random number column 0111101101001100000111001000101, so the number of times the complementary sequence is shifted (b=4). When the first column is the original virtual random number sequence (PRS), the j-axis coordinate is j = 2 x (a-b) mod 31 = 2 x (13 - 4) mod 31 = 2 x 9 = 18. Combining the above two, the position of the 5 x 5-bit array in the entire virtual random array (PRA) (i, j) = (13, 18) can be determined.

In addition, the bit array (starting with the odd-numbered column 31) coordinates (8, 31) is taken as an example, and the sixth figure is shown.

The even columns (0, 2, ..., 60) of the virtual random array (PRA) are the same as the original virtual random number column (PRS). Using this feature, first find the bit array interlaced or lattice, and its 5 bits. The same direction (that is, an even column) is used to represent the horizontal direction of the virtual random array (PRA) (ie, the i-axis direction), and the number of the original virtual random number (PRS) is determined according to the camera origin position ( 10110), and 10110 is at the 8th position (a) of the original virtual random number sequence 10000100101100111110001101110101, so the i coordinate is (i = a = 8). The (8, 31) bit array seen by the front, rear, left, and right lenses is shown in Figure 6.

The complementary virtual random number sequence (PRS) of the odd-numbered column is (10001), and it is at the 24th position of the complementary virtual random number column 0111101101001100000111001000101, so the number of times the complementary sequence is shifted (b=24). When the first column is not the original virtual random number column (PRS), the j-axis coordinate is j = 2 x (ab) mod 31+1 = 2 x (8 - 24) mod 31+1 = 2 x (-16) mod 31+1 = 2 x 15 + 1 = 31. By combining the two, the position of the 5 x 5-bit array in the entire virtual random array (PRA) (i, j) = (8, 31) can be determined.

Through the above properties, a virtual random array (PRA) can be designed as a positioning pattern. Then, as long as the positioning pattern is captured by the camera and an array of m x m bits can be identified from the image, the camera can be positioned. The most straightforward way to design a positioning pattern is to convert a bit of "0" in the virtual random array (PRA) to black, and the bit of "1" to white, and match the direction of the X and Y axes of the world coordinate system. The original virtual random array (PRA) is arranged from the upper left to the lower right, and the positioning pattern is generated from the lower left to the upper right.

In accordance with the foregoing theory, it will be further explained below that it is applied to a pattern plate 30 having a virtual random array (PRA) code. Figure 7 shows a portion of the pattern plate 30. For clarity of illustration, the defined checkerboard background is comprised of a plurality of first pattern elements 301, 302. The first pattern unit 301, 302 is arranged adjacent to each other in a matrix manner, and each of the first pattern units 301, 302 presents one of a first color and a second color. In this embodiment, the first The color and the second color respectively refer to black and white, and the colors of the adjacent first pattern units 301 and 302 are different from each other, thereby forming a black and white checkerboard background pattern, but understandably, The first color and the second color are not limited. An important feature of the present invention is that the pattern plate 30 further includes a plurality of second pattern units 303, 304 respectively disposed in the first pattern units 301, 302, and preferably, the second pattern unit 301, The 302 series is arranged in a virtual random array (PRA) code and has at least one shape feature and one color feature.

In other words, it can be considered that each of the square cells on the pattern plate 30 is composed of a first pattern unit 301, 302 containing a second pattern unit 303, 304. The first pattern units 301, 302 of this embodiment are black or white squares, and the second pattern units 303, 304 are diamond-shaped (in practical applications, any other shape, such as a circle), and black or white. Therefore, there are 4 possibilities after the combination, please refer to Figures 8A to 8D. In detail, the checker cell 30a shown in FIG. 8A is composed of a black first pattern unit and a black second pattern unit, so that the visual upper cell unit 30a is all black, and the eighth cell shown in FIG. 30b is composed of a black first pattern unit and a white second pattern unit, and the box unit 30c shown in FIG. 8C is composed of a white first pattern unit and a black second pattern unit, and the 8D image is formed. The displayed cell unit 30d is composed of a white first pattern unit and a white second pattern unit, so that the visual upper cell unit 30d is completely white.

Figure 9 is a diagram showing the arrangement ratio of the first pattern units 301, 302 and the second pattern units 303, 304 in each of the square cells in a preferred embodiment. Taking the grid cell 30c of FIG. 8C as an example, the first pattern unit 301, 302 has a base rectangle having a length (1) and a width (w), and the second pattern unit 303, 304 is a suitable diamond pattern (a practical application). It may also be in any other shape, such as a circular shape, located at the center of the first pattern unit 301, 302, that is, the center of the first pattern unit 301, 302 overlaps with the center of the second pattern unit 303, 304, as Subsequent location identification. As shown in FIG. 9, the size and position of the second pattern units 303, 304, the second pattern units 303, 304 are centered on the basis and extend l/4 in the length (l) direction and the width (w) direction, respectively. And w/4.

It should be noted that the second pattern units 303, 304 may be of any shape, such as a circle or a diamond shape, wherein the higher the resolution of the lens, the more complicated the shape can be used. When the second pattern elements 303, 304 have more features in shape, more information can be implied, and there will be more application possibilities in subsequent image recognition or other orientation recognition. Therefore, the shape and color of the second pattern units 303, 304 are not limited in the present invention.

Next, the process of image capture and correction will be explained. First, considering the number of lenses to be set, the aforementioned lens 51 is a plurality of wide-angle lenses or fisheye lenses. As shown in FIG. 10, when the lens 51 is a fisheye lens, it is photographed toward the pattern plate 30 to obtain a partial image 61. As shown in the figure, the closer to the surrounding portion, the lower the resolution and the distortion of the image (distortion) The more serious it is. The lens 51 is imaged toward the pattern plate 30 to obtain a partial image 61, and then transmitted to the image correction system 3 of the present invention. The transmission module 31 receives the partial image 61 captured by the lens 51. The correction module 33 is connected to the transmission module 31 to extract and correct a matrix image 63 from the partial image 61, as shown in FIG. More specifically, the correction module 33 captures the central portion of the partial image 61 with high resolution and less distortion, and performs leveling correction to obtain the matrix image 63. The matrix image 63 is transmitted to the computing module 35. The computing module 35 identifies a plurality of the first pattern units 301 and 302 and the plurality of second pattern units 303 and 304 on the matrix image 63. And determining, according to the first pattern unit 301, 302 and the second pattern unit 303, 304, an orientation of the lens 51 relative to the pattern plate 30.

Since the second pattern units 303 and 304 of the pattern plate 30 are arranged in a virtual random array (PRA) code, the corrected matrix image 63 is returned to the pattern plate 30 for comparison. A unique result, so the computing module 35 can calculate an orientation of the lens 51 relative to the pattern plate 30, preferably including a positional parameter and a viewing angle of the lens 51 relative to the pattern plate 30. parameter. According to the orientation information such as the positional parameters and the viewing angle parameters, the image correction system 3 of the present invention can accurately and quickly splicing the partial images 61 of the respective lenses 51 into a panoramic image around the target 50.

One of the technical points of the present invention is to use a pattern plate 30 containing a virtual random array (PRA) code. The following is a configuration for the pattern plate 30, and four lenses 51 are arranged on the target object 50 to face the pattern plate. 30 for explanation.

The pattern plate of the present invention is combined into a size based on a one-dimensional m-order Pseudo Random Sequences (PSS). The virtual random array (PRA) two-dimensional pattern is such that any one of the mxm partial two-dimensional images observed by any of the lenses 51 at any position is unique. Also, since each lens 51 sees a unique mxm partial two-dimensional image, the orientation of the image on the entire virtual random array (PRA) two-dimensional pattern (ie, the pattern plate 30) can also be discriminated. Therefore, camera correction can be performed by mxm local 2D images, and can also be converted into its orientation on the overall virtual random array (PRA).

To explain the design procedure of the pattern plate 30, the construction of the pattern plate 30 having the virtual random array (PRA) code will be described below by taking a virtual random number (PRS) of m = 5 as an example. Primitive polynomial through a 5th-order virtual random number sequence: While a solution of the initial series _{(A i + 4, A i} + 3, A i + 2, A i + 1, A i) 00001, so to obtain pseudorandom number _{as: a = a 0 a 1 a} 2 ... a 30 = 10000100101100111110001101110101, however, no other series of length 5 has the same properties.

Then, through the above-mentioned virtual random number sequence (PRS) of order 5 and length n=31, a virtual random array (PRA) of size nx 2n = 31 x 62 can be constructed as follows: = = . Among them, all even columns c _2k (0≦k≦30) are identical to the original virtual random number sequence (a= a ₀ a ₁ a ₂ ... a _n-1 ). c ₁ complementary to a number of columns (complement series): . Next, c _{1 is made} to the right circular shift right, and a total of 31 virtual random number sequences (PRS) of csr _k (0≦k≦30) are obtained. They are discharged into the PRA _{5, 31} at a time to obtain all the odd _columns as c _2k+1 = csr _k to obtain a corresponding array of virtual random arrays (PRAs).

Since the complementary sequence of a virtual random number sequence is still a virtual random number sequence (except that 111....1 will be defined as 000....0), all odd columns in the virtual random array (PRA) are virtual random numbers ( PRS). In addition, through the simulation verification of the program, it can be confirmed that the virtual random array (PRA) has window characteristics in four different viewpoint directions: when an mxm window block is in a virtual random array (PRA) of order m. Sliding derived bit arrays, where each mxm bit array, will only appear once.

Returning to the length and width of the aforementioned object 50, the plane area of the target 50 is set to L x W. Considering the size of the general construction field, the changeable amount of the target 50, and the effective observation area on the pattern plate 30, the correction area of the T cm is maintained in the four directions of the object 50. In order to meet the above requirements, the size of the pattern plate 30 can be designed as . For a virtual random array (PRA) of order m, the length and width of each of the grid cells 30a, 30b, 30c, 30d in the pattern plate 30 are: ; .

Since the constructed virtual random array (PRA) may have consecutive bit 1 or bit 0, if the all-black basic rectangle represents bit 1, and the all-white basic rectangle represents bit 0, it will represent PRA. The same color area of the strip is displayed in the correction pattern, and it is difficult to identify the bit. To overcome this problem, the present invention re-encodes the bits of a virtual random array (PRA). First targeted The size of the virtual random array (PRA) creates a black-and-white checkerboard background image consisting of black and white basic rectangular square cells. The basic rectangle is then re-formulated according to the corresponding bit value of the virtual random array (PRA), and combined with the virtual random array (PRA) coded pattern plate 30 of the invention, as shown in FIGS. 4 and 7.

The design of the pattern plate 30 of the present invention has at least the following advantages: (1) The pattern plate 30 comprises two matrixes arranged in a line group which are perpendicular to each other and fixed in pitch, and is relatively easy to be detected in image recognition technology; (2) phase Compared with the traditional four-corner extraction and correction method of artificial or basic rectangle, this method can use the line group to calculate the intersection point with the accuracy of sub-pixels; (3) the number of corners is redundant, so it can be used More than four correction feature points are used to obtain the image conversion matrix to provide more accurate conversion information. (4) Since the virtual random array (PRA) is composed of virtual random number series (PRS), the m-order virtual random array ( The PRA) pattern can be deformed according to the initial sequence and arrangement, increasing the pairing type between the system and the correction pattern, and establishing a basic anti-counterfeiting mechanism.

In order to find a block size of m x m from the captured image for positioning, the feature lines found by Hough transform are first classified. From the positioning pattern used, the feature line is mainly composed of two types of straight lines, including horizontal and vertical straight lines of the first pattern units 301 and 302, and straight lines formed by the four sides of the second pattern unit 303 and 304. Since the edge points of the horizontal and vertical lines in the first pattern unit 301, 302 are more than the edge points formed by the diamond-shaped four sides of the second pattern unit 303, 304, they are first found in the line detection. The line formed by the diamond-shaped four sides of the second pattern unit 303, 304 is later found to intersect the horizontal and vertical lines of the first pattern unit 301, 302, and this characteristic can be used to filter it out. Finally, two sets of feature lines composed of horizontal and vertical lines can be obtained. However, in other embodiments, the second pattern elements 303, 304 may also take other shapes, such as a circle.

Then from these two feature line groups, try to find a block size of m x m. Although each of the grid cells 30a, 30b, 30c, 30d on the pattern plate 30 are actually the same size, if the viewpoint of the fisheye lens or the camera is viewed from an oblique angle, the originally identical grid unit 30a , 30b, 30c, 30d, because of the perspective projection relationship, the pixels of each of the square cells 30a, 30b, 30c, 30d in the image are different. The distortion effect caused by this perspective projection will be skewed with the camera viewpoint, causing the size of the grid cells 30a, 30b, 30c, 30d to appear several times in the image, as shown in Fig. 10.

Finally, adjacent m+1 lines are selected from each of the two feature line groups to form a block size of mxm, and then the pixel color is taken out from the center point of each grid in the block, and is 0 according to black. The white-to-one conversion method obtains a bit array and then compares it into a virtual random array (PRA) to know the position of the bit array in the virtual random array (PRA). And the horizontal and vertical directions corresponding to the two feature line groups.

In view of the above, since the pattern plate 30 of the present invention has a virtual random array (PRA) code, the image captured by each lens 62 has its uniqueness, so that as shown in FIG. 12, the object 50 can be infinitely oriented. The present invention can still be implemented by being arbitrarily placed on the pattern plate 30. In addition, the arrangement of each lens 62 does not need to limit its symmetrical arrangement or specific position. As long as it is substantially facing the pattern plate 30, the position and orientation of each lens 62 can be recognized from the image. It is of course also possible to define any point on the pattern plate 30 as the origin, and then calculate the coordinate position and orientation of all the lenses 51 with respect to the origin. In addition, the technology of the present invention is not limited to the above-described planar image, and can also be applied to a stereoscopic image; in detail, for example, at least six lenses are respectively disposed in front, back, left, right, up, and down of the target object, and finally It can be spliced out of the spherical panoramic image, and it can be equipped with perfect obstacle avoidance when applied to airplanes or drones.

In conjunction with the foregoing image correction system, another embodiment of the present invention discloses an image correction method, such as the flowchart shown in FIG. The image correction method comprises the following steps: (S11) providing a pattern plate 30 made by a virtual random array code; (S12) arranging the object 50 on the pattern plate 30, wherein the object 50 is provided with a plurality of a lens 51; (S13) obtaining a partial image 61 via each of the lenses 51; (S14) extracting and correcting a matrix image 63 from the partial image 61; (S15) calculating each of the matrix images 63 based on the matrix image 63 The partial orientation of the lens 51 relative to the pattern plate is (S16). The partial image 61 of each of the lenses 51 is spliced into the panoramic image around the target object 50 according to the orientation of the lens 51.

The step (S14) captures the central portion of the partial image 61 and performs leveling correction to obtain the matrix image 63.

As in the foregoing embodiment, the pattern plate 30 has a plurality of first pattern units 301, 302 and a plurality of second pattern units 303, 304, the first pattern units 301, 302 being arranged adjacent to each other in a matrix manner, each of which The first pattern unit 301, 302 presents one of the first color and the second color, and the adjacent first pattern units 301, 302 are different in color from each other, and the second pattern units 303, 304 are respectively set In the first pattern unit 301, 302. The step of calculating the orientation of each of the lenses 51 relative to the pattern plate 30 is to identify a plurality of the first pattern units 301, 302 and the plurality of second patterns on the matrix image 63. The units 303 and 304 calculate the position parameters and the angle of view parameters of the lens 51 relative to the pattern plate 30 according to the first pattern unit and the second pattern unit 301 and 302. The second pattern sheet 303, 304 has at least one shape feature and one color feature, wherein the color feature is one of the first color and the second color.

In summary, the image correction system and the image correction method provided by the present invention use a pattern plate having a virtual random array code, and the partial patterns captured by the lens of any orientation or posture are unique, so the object is relatively The position of the pattern plate is not limited, and the orientation of the lens is not limited. Each lens only needs to be substantially oriented toward the pattern plate, and there is no need to have feature points in the overlapping area as in the prior art. Thereby, the correction can be accurately performed and the correct panoramic image can be generated, and can be automatically performed to greatly reduce the input of manpower.

The embodiments described above are only intended to illustrate the embodiments of the present invention, and to explain the technical features of the present invention, and are not intended to limit the scope of protection of the present invention. Any change or singularity that can be easily accomplished by those skilled in the art is within the scope of the invention. The scope of the invention is defined by the scope of the patent application.

11‧‧‧ Vehicles

111‧‧‧ camera

113‧‧‧ images

12‧‧‧Plate

3‧‧‧Image Correction System

30‧‧‧pattern board

30a~30d‧‧‧ square unit

301‧‧‧First pattern unit

302‧‧‧First pattern unit

303‧‧‧Second pattern unit

304‧‧‧Second pattern unit

31‧‧‧Transmission module

33‧‧‧ calibration module

35‧‧‧The computing module

50‧‧‧ Subject matter

51‧‧‧ lens

61‧‧‧Partial imagery

63‧‧‧ Matrix Image

S11~S16‧‧‧Steps

1 is a schematic diagram of a conventional image correction system; FIG. 2 is a schematic diagram of an image taken by a lens of a conventional image correction system; FIG. 3 is a block diagram of the image correction system of the present invention; Figure 5 is a schematic view of a correction embodiment; Figure 6 is a schematic view of another correction embodiment; Figure 7 is a partial schematic view of the pattern plate; Figures 8A to 8D are drawings on the pattern plate Figure 9 is a schematic diagram of one of the cells on the pattern board; Figure 10 is a schematic diagram of a partial image captured by the lens; Figure 11 is a schematic diagram of the correction of the partial image to obtain a matrix image Figure 12 is a schematic view of another embodiment of the present invention; and Figure 13 is a flow chart of the image correction method of the present invention.

Claims (9)

An image correction system for obtaining a panoramic image around a target object, the target object having at least one lens, the image correction system comprising: a pattern plate for accommodating the target object within the range of the pattern plate, The pattern plate has a plurality of first pattern units arranged adjacent to each other in a matrix manner, each of the first pattern units exhibiting one of a first color and a second color, and the adjacent first patterns The units are different in color from each other; and the plurality of second pattern units are respectively disposed in the first pattern unit in a Pseudo Random Array (PRA) code arrangement, wherein each of the second pattern units Each has a shape feature and a color feature, and the color feature is one of the first color and the second color; a transmission module receives a partial image of the lens toward the pattern plate; a calibration mode a group, connected to the transmission module, to extract and correct a matrix image from the partial image; an operation module, to identify a plurality of the first image on the matrix image And a plurality of the second pattern units, and calculating an orientation of the lens relative to the pattern board according to the first pattern unit and the second pattern unit. The image system of claim 1, wherein the correction module captures a central portion of the partial image and performs leveling correction to obtain the matrix image. The image system of claim 1, wherein the at least one lens system comprises a plurality of lenses, and the operation module splices the partial images of each of the lenses into the image according to the orientation of each lens. The panoramic image around the subject matter. The image system of claim 3, wherein the lens is a fisheye lens. The image system of claim 1, wherein the orientation comprises a positional parameter and a viewing angle parameter of the lens relative to the pattern plate. An image correction method for obtaining a scene image around a target object, the image correction method comprising the steps of: providing a pattern plate, wherein the pattern plate has a plurality of first pattern units and a plurality of second pattern units, The first pattern units are arranged adjacent to each other in a matrix manner, and each of the first pattern units presents one of a first color and a second color, and the adjacent first pattern units are different in color from each other. The second pattern unit is respectively disposed in the first pattern unit in a Pseudo Random Array (PRA) code arrangement, and each of the second pattern units has a shape feature and a color respectively. a feature, and the color feature is one of the first color and the second color; the object is disposed on the pattern plate, wherein the target object is provided with a plurality of lenses; and a partial image is obtained through each lens Extracting and correcting a matrix image from the partial image; and calculating, according to the matrix image, an orientation of each lens relative to the pattern plate. The image correction method of claim 6, wherein the step of calculating an orientation of each of the lenses relative to the pattern plate identifies a plurality of the first pattern units on the matrix image And a plurality of the second pattern units, and calculating a position parameter and a viewing angle parameter of each of the lenses relative to the pattern plate according to the first pattern unit and the second pattern unit. The image correction method of claim 7, wherein the step of extracting and correcting a matrix image from the partial image is performed by capturing a central portion of the partial image and leveling the correction to obtain The matrix image. The image correction method of claim 6, further comprising the step of: splicing each of the partial images of each of the lenses into the panoramic image around the target according to the orientation of each lens.