A method for calibrating a Fourier transform-based event camera

Abstract

The application discloses a kind of event camera calibration methods based on Fourier transform, comprising: step S1: event camera calibration system is built;Including event camera, vertically placed and movable and rotatable LCD display screen;LCD display screen flashes according to frequency to trigger event;Step S2: obtaining event stream;Event stream data output by event camera is transformed and handled, and event frame image data is obtained;Step S3: extracting feature points;Two directions wrapped phase diagram is calculated using Fourier transform: φ x0 And φ y0 ;Step S4: LCD display screen is moved before and after or at different angles, at least 3 pose data are obtained, repeat step S1 to step S3, and horizontal and vertical stripe images at different poses are captured respectively;Step S5: according to target image information, the world coordinates corresponding to the feature points are obtained, then according to the corresponding relationship between feature points, the internal and external parameters of event camera are obtained using Zhang calibration method, and the event camera parameter value is optimized using nonlinear optimization algorithm.

Description

Technical Field

[0001] This invention belongs to the field of camera calibration technology. Specifically, this invention relates to an event camera calibration method based on Fourier transform. Background Technology

[0002] An event-based camera is a bio-inspired advanced vision sensor, also known as a dynamic vision sensor (DVS) or DAVIS (Dynamic and Active-PixelVision Sensor). Compared to standard cameras that acquire images at a fixed frame rate, event cameras offer unique advantages such as low latency, high dynamic range, low power consumption, and high temporal resolution. An event camera can be simply understood as a sensor that detects only moving objects. Each pixel has an independent photoelectric sensing module; when the brightness change at that pixel exceeds a set threshold, event data (also called pulse data) is generated and output. Because all pixels operate independently, the data output of an event camera is asynchronous and exhibits sparse spatial characteristics. This is the biggest difference between event cameras and standard cameras, and it is also the core innovation of the camera. The advantage of this imaging method is that it can greatly reduce redundant data, thereby improving the computational efficiency of post-processing algorithms.

[0003] Event cameras are widely used in target tracking, 3D reconstruction, and visual measurement. As part of a vision system, parameter calibration is a crucial prerequisite for ensuring system performance. Since the event stream output by an event camera accumulates into event frames over the time domain, it is insensitive to the spatial distribution of feature points. Traditional event camera calibration methods rely excessively on spatial intensity distribution for feature point extraction, which contradicts the characteristics of event cameras and limits their calibration accuracy. Therefore, this patent proposes an event camera calibration method based on Fourier transform. This method utilizes the frequency domain characteristics of event data to extract feature points from the target, improving the flexibility and accuracy of event camera calibration. Summary of the Invention

[0004] This invention provides an event camera calibration method based on Fourier transform to solve the problems existing in the background art.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: an event camera calibration method based on Fourier transform, specifically including the following steps:

[0006] Step S1: Set up an event camera calibration system; the system includes an event camera, a vertically placed and movable / rotatable LCD screen; the LCD screen flashes at a certain frequency to trigger events;

[0007] Step S2: Acquire the event stream; Place the LCD display screen within the imaging range of the event camera. The circular array pattern displayed on the LCD display screen is regarded as a 2D calibration target. The event stream data output by the event camera is converted and processed. The event stream data is accumulated into two-dimensional event frames within the time window and converted into image information to obtain event frame image data.

[0008] Step S3: Extract feature points; use Fourier transform to calculate the wrap-around phase map in two directions: φ x0 and φ y0 The feature points of the target image are located at φ x0 and φ y0 At the 2π phase value between;

[0009] Step S4: Move the LCD display back and forth or at different angles to obtain data of at least 3 poses. Repeat steps S1 to S3 to capture horizontal and vertical stripe images of different poses respectively.

[0010] Step S5: Based on the target image information, obtain the world coordinates corresponding to the feature points. Then, based on the correspondence between the feature points, use Zhang's calibration method to obtain the intrinsic and extrinsic parameters of the event camera. Use a nonlinear optimization algorithm to optimize the event camera parameter values.

[0011] Furthermore, in step S1, each pixel of the event camera works independently and asynchronously, and only outputs changes in brightness. The format of an event is a vector, represented as follows:

[0012] e = (x i ,y i ,t i ,p i (1)

[0013] Among them, (x i ,y i ) represents the pixel coordinates of the event, t i p represents the timestamp of the event. i Indicates the polarity of the event, i.e. whether the brightness has increased or decreased compared to the previous sample;

[0014] The event camera detects the logarithmic change in light intensity illuminating each pixel individually. Let the brightness be I and the logarithm of light intensity be denoted as logI. Based on the change in the logarithm of light intensity, it asynchronously outputs a brightening or darkening event. The event output is represented as follows:

[0015] logI(x i ,t i )-logI(x i ,t i -Δt i)=±C (2)

[0016] Where, x i =(x i ,y i ) indicates that at t i The pixel position at time Δt represents a tiny time interval, and C represents a certain threshold. If the brightness information of the current pixel changes by more than the threshold C and increases, a positive event p is triggered. i = +1, if it is a decrease, then a negative event p is triggered. i =-1. Within the entire camera's field of view, as long as a single pixel value changes, an event will be output. All these events occur asynchronously. In the event camera calibration system, events are triggered by controlling the LCD screen to flash a circular array pattern at a certain frequency.

[0017] Furthermore, in step S2, the event camera outputs a series of events, including the timestamp of the event, the pixel coordinates of the event, and the polarity of the event. The event stream can be denoted as S, and is represented as follows:

[0018] S={e(x1,y1,t1,p1),e(x2,y2,t2,p2),...,e(x i ,y i ,t i ,p i )} (3)

[0019] A two-dimensional event frame can be denoted as F(x,y,T), which represents the accumulation of all events occurring at position (x,y) within a time window T;

[0020] Considering polarity information, positive and negative polarity events can be accumulated separately. The accumulation process is represented as follows:

[0021]

[0022] Where t0 is the start time of the event frame, and I(x,y) represents the generated image, which can be generated from the event frame in the following way:

[0023] I(x,y)=(f(F(x,y,T) + ,f(F(x,y,T) - ,f(F(x,y,T))) (5)

[0024]

[0025] Where, F(x,y,T) + and F(x,y,T) -Let F(x,y,T) represent the accumulation of positive and negative events, respectively, and let F(x,y,T) represent the accumulation of the total number of events. f is a transformation function. max It is the maximum event count in the event frame. This mapping normalizes the positive and negative polarity event counts to the range of 0 to 255 and puts them into the red and blue channels respectively, converting the event frame data into image data.

[0026] Furthermore, step S3 employs a frequency domain analysis algorithm to extract the target's feature points from the event frames accumulated over time in the event stream output by the event camera, and uses Fourier transform on the periodic image to extract the frequency domain characteristics of the event data, thereby more accurately determining the parameters of the event camera.

[0027] The array of periodic targets can be viewed as a two-dimensional periodic signal, which can be obtained through Fourier transform as follows:

[0028]

[0029] Among them, H 0 It is a zero-order spectrum. and H represents the k-th order spectrum. * Indicates conjugate operation;

[0030] Step S2 uses a circular array pattern as an example as the 2D calibration target.

[0031] The intensity calculations for Fourier Transform (FT), Inverse Fourier Transform (IFT), and Feature Detection (FD) are as follows:

[0032]

[0033] Where (x,y) represents an image point, (x c ,y c ) represents the center of the feature point, N is the distance between the centers of adjacent circles, which can be used as the period of the circular target, and R is the radius of each circle; Return the largest integer less than the input;

[0034] The circular array pattern is periodic in both the horizontal and vertical directions; when the camera's optical axis is perpendicular to the target plane, the circular target image can be considered as a two-dimensional periodic signal, and its intensity function can be converted into a Fourier series:

[0035] I(x,y)=A0+A x0 cosφ x0 +A y0 cosφ y0 (10)

[0036] Where A0 represents the DC component, A x0 and Ay0 Represents the alternating current component, φ x0 and φ y0 The two wrap-around phase maps represent the fundamental frequency along the x and y axes; using Fourier transform (FT), the Fourier spectrum of the circular target image can be obtained as follows:

[0037]

[0038] Where Q0 is the Fourier spectrum or zeroth-order spectrum of A0; Q x (f x -f x0 ,f y )express The Fourier spectrum or the fundamental spectrum along the x-axis. Q represents x The conjugate of; similarly, Q y (f x ,f y -f y0 )express The Fourier spectrum or the fundamental spectrum along the y-axis; Q represents y Conjugate;

[0039] Separate the Fourier spectrum Q by selecting an appropriate window. x and Q y And perform the inverse Fourier transform (IFT). and Two wrapper phase maps φ from 0 to 2π are obtained. x0 and φ y0 And the center of the circle is found to be at φ x0 and φ y0 Each circle has a 2π phase value, and the center of the circle coincides with the 2π phase point. Therefore, the center of the circle can be located by extracting the 2π phase point.

[0040] Furthermore, the event camera (1) calibration system adopts a pinhole imaging model. The imaging position of any point in space in the image can be represented by the pinhole imaging model. For image point p = [u, v], its corresponding 3D point is P = [X]. w ,Y w Z w P and p are represented by adding 1 as the last element to their augmented vectors. Therefore, in the context of the pinhole imaging model, the imaging process can be represented as:

[0041] sp=K[R t]P (12)

[0042] Where s is an arbitrary scaling factor, R is a rotation matrix, and t is a translation matrix;

[0043] K is the intrinsic matrix to be estimated, expressed as:

[0044]

[0045] Where p = [u,v] is the focal length along the two image axes, [u0,v0] is the principal point, and χ is the tilt factor;

[0046] To ensure the accuracy and reliability of the event camera calibration, radial distortion correction was performed on the pinhole imaging model; the normalized undistorted pixel [u,v] and the corresponding normalized distorted point [u d ,v d The relationship between them can be described as follows:

[0047]

[0048] Where, r 2 =u 2 +v 2 k1 and k2 are radial distortion coefficients. By establishing a one-to-one mapping between P and p through the detection of calibrated target feature points, the above equation can be solved to estimate the unknown camera parameters.

[0049] Then, a nonlinear optimization algorithm is used to optimize the event camera parameter values, as shown in the following formula:

[0050]

[0051] Where M represents the number of azimuths of the target; N represents the total number of feature points in a single target image; m ij This represents the j-th feature point in the i-th orientation. This represents the coordinates of the back-projected point obtained using the intrinsic parameters. The minimum value obtained by the above formula is the optimized event camera parameter.

[0052] The beneficial effects of adopting the above technical solutions are:

[0053] 1. The present invention provides an event camera calibration method based on Fourier transform. The pioneering application of Fourier transform in event camera calibration produces effective, reliable, and high-precision calibration results.

[0054] 2. The present invention provides an event camera calibration method based on Fourier transform, which uses phase distribution instead of intensity information to locate feature points. Compared with intensity information positioning, phase distribution positioning of feature points is more in line with the characteristics of event cameras, thereby improving the reliability and accuracy of event camera calibration.

[0055] 3. The present invention provides an event camera calibration method based on Fourier transform. Although the present invention focuses on circular arrays, the method can be easily extended to other periodic targets, such as grids, demonstrating its simplicity and flexibility. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the event camera calibration based on Fourier transform of this invention.

[0057] Figure 2 It is a feature detection framework based on Fourier transform;

[0058] Figure 3 This is a flowchart of extracting feature points using Fourier transform;

[0059] The diagram is labeled as follows: 1-Event camera, 2-LCD display screen, 3-Circular array target image. Detailed Implementation

[0060] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, in order to help those skilled in the art to have a more complete, accurate and in-depth understanding of the concept and technical solutions of the present invention, and to facilitate its implementation.

[0061] like Figures 1 to 3 As shown, this invention is an event camera calibration method based on Fourier transform, specifically including the following steps:

[0062] Step S1: Set up the event camera 1 calibration system; the system includes event camera 1, a vertically placed and movable and rotatable LCD display 2; the LCD display 2 flashes at a frequency to trigger events;

[0063] Step S10: In step S1, each pixel of event camera 1 works independently and asynchronously, and only outputs changes in brightness. The format of an event is a vector, represented as follows:

[0064] e = (x i ,y i ,t i ,p i (1)

[0065] Among them, (x i ,y i ) represents the pixel coordinates of the event, t i p represents the timestamp of the event. i Indicates the polarity of the event, i.e. whether the brightness has increased or decreased compared to the previous sample;

[0066] Event camera 1 detects the logarithmic change in light intensity illuminating each pixel individually. Let the brightness be I and the logarithmic change in light intensity be denoted as logI. Based on the change in the logarithmic change in light intensity, it asynchronously outputs a brightening or darkening event. The event output is represented as follows:

[0067] logI(x i ,ti )-logI(x i ,t i -Δt i )=±C (2)

[0068] Where, x i =(x i ,y i ) indicates that at t i The pixel position at time Δt represents a tiny time interval, and C represents a certain threshold. If the brightness information of the current pixel changes by more than the threshold C and increases, a positive event p is triggered. i = +1, if it is a decrease, then a negative event p is triggered. i =-1. Within the entire field of view of the camera, as long as there is a change in the value of a single pixel, an event will be output. All these events occur asynchronously. In the event camera (1) calibration system, the event is triggered by controlling the LCD screen to flash a circular array pattern at a certain frequency.

[0069] Step S2: Acquire event stream; Place the LCD display 2 within the imaging range of the event camera 1. The circular array pattern displayed on the LCD display 2 is regarded as a 2D calibration target. The event stream data output by the event camera 1 is converted and processed. The event stream data is accumulated into two-dimensional event frames within the time window and converted into image information to obtain event frame image data.

[0070] Step S20: In step S2, the event camera 1 outputs a series of events, including the timestamp of the event occurrence, the pixel coordinates of the event, and the polarity of the event occurrence. The event stream can be denoted as S, and is represented as follows:

[0071] S={e(x1,y1,t1,p1),e(x2,y2,t2,p2),...,e(x i ,y i ,t i ,p i )} (3)

[0072] A two-dimensional event frame can be denoted as F(x,y,T), which represents the accumulation of all events occurring at position (x,y) within a time window T;

[0073] Considering polarity information, positive and negative polarity events can be accumulated separately. The accumulation process is represented as follows:

[0074]

[0075] Where t0 is the start time of the event frame, and I(x,y) represents the generated image, which can be generated from the event frame in the following way:

[0076] I(x,y)=(f(F(x,y,T) + ,f(F(x,y,T) - ,f(F(x,y,T))) (5)

[0077]

[0078] Where, F(x,y,T) + and F(x,y,T) - Let F(x,y,T) represent the accumulation of positive and negative events, respectively, and let F(x,y,T) represent the accumulation of the total number of events. f is a transformation function. max It is the maximum event count in the event frame. This mapping normalizes the positive and negative polarity event counts to the range of 0 to 255 and puts them into the red and blue channels respectively, converting the event frame data into image data.

[0079] Step S3: Extract feature points; such as Figure 2 As shown, the wrap-around phase diagram in two directions is calculated using Fourier transform: φ x0 and φ y0 The feature points of the target image are located at φ x0 and φ y0 At the 2π phase value between; Figure 2 (d) shows the target image marked with feature points, where the red "+" indicates the location of the feature points;

[0080] Step S30: Step S3 uses a frequency domain analysis algorithm to extract the target feature points from the event frames accumulated over time in the event stream output by event camera 1, and uses Fourier transform on the periodic image to extract the frequency domain characteristics of the event data, thereby more accurately determining the parameters of the event camera.

[0081] The array of periodic targets can be viewed as a two-dimensional periodic signal, which can be obtained through Fourier transform as follows:

[0082]

[0083] Among them, H 0 It is a zero-order spectrum. and H represents the k-th order spectrum. * Indicates conjugate operation;

[0084] Step S2 uses a circular array pattern as an example as the 2D calibration target.

[0085] Figure 2 The framework for feature detection is shown, including Fourier Transform (FT), Inverse Fourier Transform (IFT), and Feature Detection (FD), with intensity calculation as follows:

[0086]

[0087] Where (x,y) represents an image point, (x c ,y c ) represents the center of the feature point, N is the distance between the centers of adjacent circles, which can be used as the period of the circular target, and R is the radius of each circle; Return the largest integer less than the input;

[0088] like Figure 2 As shown in (a), the circular array pattern is periodic in both the horizontal and vertical directions; when the camera's optical axis is perpendicular to the target plane, the circular target image can be regarded as a two-dimensional periodic signal, and its intensity function can be converted into a Fourier series:

[0089] I(x,y)=A0+A x0 cosφ x0 +A y0 cosφ y0 (10)

[0090] Where A0 represents the DC component, A x0 and A y0 Represents the alternating current component, φ x0 and φ y0 The two wrap-around phase maps represent the fundamental frequency along the x and y axes; using Fourier transform (FT), the Fourier spectrum of the circular target image can be obtained as follows:

[0091]

[0092] Where Q0 is the Fourier spectrum or zeroth-order spectrum of A0; Q x (f x -f x0 ,f y )express The Fourier spectrum or the fundamental spectrum along the x-axis. Q represents x The conjugate of; similarly, Q y (f x ,f y -f y0 )express The Fourier spectrum or the fundamental spectrum along the y-axis; Q represents y Conjugate;

[0093] Figure 2 (b) The Fourier spectrum of the target image is displayed, and the Fourier spectrum Q is separated by selecting an appropriate window. x and Q y And perform inverse Fourier transform (IFT) and Two wrapper phase maps φ from 0 to 2π are obtained. x0 and φ y0 ,like Figure 2 As shown in (c); Figure 2 (d) shows the two wrapped phases φ x0 and φ y0 The sum of , and the center of the circle at φ x0 and φ y0 Each circle contains a 2π phase value, and the center of the circle coincides with the 2π phase point. The position of the center is marked with a red "+". Therefore, the center of the circle can be located by extracting the 2π phase point.

[0094] Step S4: Move the LCD display 2 back and forth or at different angles to obtain data of at least 3 poses. Repeat steps S1 to S3 to capture horizontal and vertical stripe images of different poses respectively.

[0095] Step S5: Based on the target image information, obtain the world coordinates corresponding to the feature points. Then, based on the correspondence between the feature points, use Zhang's calibration method to obtain the intrinsic and extrinsic parameters of the event camera. Use a nonlinear optimization algorithm to optimize the event camera parameter values.

[0096] Step S50: The calibration system for event camera 1 uses a pinhole imaging model. The imaging position of any point in space in the image can be represented by the pinhole imaging model. For image point p = [u, v], its corresponding 3D point is P = [X]. w ,Y w Z w P and p are represented by adding 1 as the last element to their augmented vectors. Therefore, in the context of the pinhole imaging model, the imaging process can be represented as:

[0097] sp=K[R t]P (12)

[0098] Where s is an arbitrary scaling factor, R is a rotation matrix, and t is a translation matrix;

[0099] K is the intrinsic matrix to be estimated, expressed as:

[0100]

[0101] Where p = [u,v] is the focal length along the two image axes, [u0,v0] is the principal point, and χ is the tilt factor;

[0102] To ensure the accuracy and reliability of the calibration of event camera 1, radial distortion correction was performed on the pinhole imaging model; the normalized undistorted pixel [u,v] and the corresponding normalized distorted point [u d ,v d The relationship between them can be described as follows:

[0103]

[0104] Where, r 2 =u 2 +v 2 k1 and k2 are radial distortion coefficients. By establishing a one-to-one mapping between P and p through the detection of calibrated target feature points, the above equation can be solved to estimate the unknown camera parameters.

[0105] Then, a nonlinear optimization algorithm is used to optimize the parameter values ​​of event camera 1, as shown in the following formula:

[0106]

[0107] Where M represents the number of azimuths of the target; N represents the total number of feature points in a single target image; m ij This represents the j-th feature point in the i-th orientation. This represents the coordinates of the back-projection point obtained using the intrinsic parameters. The minimum value obtained by the above formula is the optimized parameters of event camera 1.

[0108] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other situations without modification, are all within the protection scope of the present invention.

Claims

1. An event camera calibration method based on Fourier transform, characterized in that: Specifically, it includes the following steps: Step S1: Set up an event camera (1) calibration system; the system includes an event camera (1), a vertically placed and movable and rotatable LCD display (2); the LCD display (2) flashes at a frequency to trigger events; Step S2: Obtain the event stream; Place the LCD display (2) within the imaging range of the event camera (1), and regard the circular array pattern displayed on the LCD display (2) as a 2D calibration target. Convert the event stream data output by the event camera (1), accumulate the event stream data into two-dimensional event frames within the time window, and convert them into image information to obtain event frame image data. Step S3: Extract feature points; calculate the wrap-around phase map in two directions using Fourier transform: and The feature points of the target image are located at and Between At the phase value; Step S4: Move the LCD display (2) back and forth or at different angles to obtain data of at least 3 poses. Repeat steps S1 to S3 to capture horizontal and vertical stripe images of different poses respectively. Step S5: Based on the target image information, obtain the world coordinates corresponding to the feature points. Then, based on the correspondence between the feature points, use Zhang's calibration method to obtain the intrinsic and extrinsic parameters of the event camera. Use a nonlinear optimization algorithm to optimize the event camera parameter values.

2. The event camera calibration method based on Fourier transform according to claim 1, characterized in that: In step S1, each pixel of the event camera (1) works independently and asynchronously, and only outputs changes in brightness. The format of an event is a vector, represented as follows: (1) in, Represents the pixel coordinates of the event. Indicates the timestamp of the event. Indicates the polarity of the event, i.e. whether the brightness has increased or decreased compared to the previous sample; The event camera (1) individually detects the logarithmic change in the light intensity illuminating that pixel, assuming the brightness is... The logarithm of light intensity is denoted as It asynchronously outputs brightening or darkening events based on the logarithmic change in light intensity, with the event output represented as follows: (2) in, Indicates in Pixel position at time. Indicates a tiny time interval. This represents a certain threshold; if the change in the brightness information of the current pixel exceeds the threshold... And if it increases, then a positive event is triggered. If the value decreases, a negative event is triggered. Within the entire field of view of the camera, as long as there is a change in the value of a single pixel, an event will be output. All these events occur asynchronously. In the event camera (1) calibration system, the event is triggered by controlling the LCD screen to flash a circular array pattern at a certain frequency.

3. The event camera calibration method based on Fourier transform according to claim 2, characterized in that: In step S2, the event camera (1) outputs a series of events, including the timestamp of the event, the pixel coordinates of the event, and the polarity of the event. The event stream can be denoted as... , means as follows: (3) A two-dimensional event frame can be denoted as , indicating within the time window Inside, location The accumulation of all events that have occurred; Considering polarity information, positive and negative polarity events can be accumulated separately. The accumulation process is represented as follows: (4) in, It is the start time of the event frame, let's assume... This represents the generated image, which can be generated from the event frame in the following ways: (5) (6) (7) in, and These represent the accumulation of positive and negative polarity events, respectively. This represents the cumulative total number of events. It is a transformation function. It is the maximum event count in the event frame. This mapping normalizes the positive and negative polarity event counts to the range of 0 to 255 and puts them into the red and blue channels respectively, converting the event frame data into image data.

4. The event camera calibration method based on Fourier transform according to claim 1, characterized in that: Step S3 employs a frequency domain analysis algorithm to extract the target's feature points from the event frames accumulated over time in the event stream output by the event camera (1). Fourier transform is used on the periodic image to extract the frequency domain characteristics of the event data, thereby more accurately determining the parameters of the event camera. The array of periodic targets can be viewed as a two-dimensional periodic signal, which can be obtained through Fourier transform as follows: (8) in, It is a zero-order spectrum. and Represents the k-th order spectrum. Indicates conjugate operation; Step S2 uses a circular array pattern as an example as the 2D calibration target. The intensity calculations for Fourier Transform (FT), Inverse Fourier Transform (IFT), and Feature Detection (FD) are as follows: (9) in, Represents an image point, The center of the circle is represented by the feature point. The distance between the centers of adjacent circles can be used as the period of a circular target. The radius of each circle; Return the largest integer less than the input; The circular array pattern is periodic in both the horizontal and vertical directions; when the camera's optical axis is perpendicular to the target plane, the circular target image can be considered as a two-dimensional periodic signal, and its intensity function can be converted into a Fourier series: (10) in, Represents the DC component. and Indicates the communication component. and The two wrap-around phase maps represent the fundamental frequency along the x and y axes; using Fourier transform (FT), the Fourier spectrum of the circular target image can be obtained as follows: (11) in, for Fourier spectrum or zeroth order spectrum; express Fourier spectrum or along The fundamental spectrum of the axis express conjugate; similarly, express Fourier spectrum or along The fundamental spectrum of the axis; express Conjugate; Separate the Fourier spectrum by selecting an appropriate window. and And perform the inverse Fourier transform (IFT). and , get from 0 to Two package phase diagrams and And the center of the circle can be obtained. and All of them have Phase value, and the center of the circle is Phase points are consistent, thus allowing for extraction The phase point is used to locate the center of the circle.

5. The event camera calibration method based on Fourier transform according to claim 2, characterized in that: The event camera (1) calibration system adopts a pinhole imaging model. The imaging position of any point in space in the image can be represented by the pinhole imaging model. For image points The corresponding 3D point is ; and By adding 1 as the last element to represent their augmented vectors, the imaging process can therefore be represented in the context of the pinhole imaging model as follows: (12) in, It is any scaling factor. It is a rotation matrix. It is a translation matrix; The intrinsic matrix to be estimated is expressed as: (13) in, It is the focal length along the two image axes. It is the main point. It is the tilt factor; To ensure the accuracy and reliability of the event camera (1) calibration, radial distortion correction was performed on the pinhole imaging model; normalized undistorted pixels were then used. With the corresponding normalized distortion point The relationship between them can be described as follows: (14) in, , and The radial distortion coefficient is established by calibrating the detection of target feature points. and By establishing a one-to-one mapping between them, the above equations can be solved to estimate the unknown camera parameters; Then, a nonlinear optimization algorithm is used to optimize the parameter values ​​of the event camera (1), as shown in the following formula: (15) in, Indicates the azimuth of the target being collected; This represents the total number of feature points in a target image. Indicates the first The first position in the direction One feature point, This represents the coordinates of the back-projected point obtained using intrinsic parameters. This represents the rotation parameter of the i-th orientation in the world coordinate system; Let represent the world coordinates corresponding to the j-th projection point in the i-th orientation. The minimum value obtained by the above formula is the optimized event camera (1) parameter.

Images