Event camera optical system parameter calibration method

Abstract

The invention relates to an event camera optical system parameter calibration method, and belongs to the technical field of optical system parameter calibration. The method comprises the five steps of data preprocessing, circle center initial positioning, data accurate processing, circle center secondary positioning and event camera internal parameter solving. And optical system parameter calibration of the event camera is realized by means of a traditional calibration board and a Zhang Zhengyou calibration algorithm. Meanwhile, weight information is introduced into the event camera for the first time, the inherent defect that the event camera lacks gray information is overcome, weighted application of the event camera is achieved, and the calibration precision is improved in the field of event camera optical system parameter calibration through a calibration plate. In addition, the method does not depend on other electronic equipment any more, and the application scene is more flexible and convenient.

Description

Technical Field

[0001] The present invention relates to the technical field of optical system parameter calibration, and in particular to a method for calibrating optical system parameters of an event camera, which is suitable for calibrating the internal parameters (such as focal length, principal point coordinates, distortion coefficient, etc.) of a dynamic vision system based on an event camera. Background Art

[0002] By emulating the human eye's imaging mode, event cameras can image only targets whose light intensity changes in the scene. Each pixel triggers events asynchronously based on changes in light intensity. At the same time, changes in light intensity, spatial position, and time are reflected through the Address-Event Representation (AER) method. Due to its unique imaging characteristics, event cameras have high dynamic range, high temporal resolution, low latency, and minimal data redundancy. Therefore, event cameras are widely used in scenarios such as high-speed motion, high-speed target detection and tracking, autonomous navigation systems, and indoor positioning systems. In these applications, since calibration accuracy directly affects the measurement results, high-precision calibration of event camera optical system parameters is an important prerequisite for these applications.

[0003] Camera calibration based on traditional frame cameras has been extensively researched for decades, resulting in the development of numerous algorithms and relatively mature solutions. Camera calibration is achieved by observing feature points on a calibration pattern, typically using a checkerboard, coplanar circles, and AprilTags grids. Furthermore, built-in libraries such as OpenCV and MATLAB provide open-source calibration toolboxes. Because event cameras are asynchronous sensors, the brightness changes of each pixel are asynchronously output as an event stream. Therefore, image-based feature point detection is not suitable for event stream data, making it impossible to calibrate the optical system parameters of event cameras using traditional calibration frameworks.

[0004] Initial methods for calibrating the optical system parameters of event cameras relied on stroboscopic calibration. This method exploits the sensitivity of event cameras to brightness changes. Using a high-frequency flashing light source (such as an LED) or an actively illuminated calibration pattern (a checkerboard pattern with LEDs), periodic brightness variations are generated in the temporal domain to trigger event generation. However, stroboscopic methods rely on a screen with a specific frequency to generate the calibration pattern, which presents a device dependency issue. To overcome this issue, researchers have used neural network image reconstruction methods for camera calibration. These methods reconstruct event stream data into grayscale images, extract feature points from the reconstructed images, and then use traditional camera calibration tools to calibrate the optical system parameters of event cameras. This method requires training for each set of optical system parameters, resulting in large data volumes and a lack of versatility. Furthermore, the reconstructed images may produce artifacts, which can affect calibration performance. Consequently, the relative motion-triggered event method has gained favor. This method reduces the complexity and data size of neural network methods while requiring no additional electronic equipment, increasing application flexibility. However, this method is susceptible to various forms of noise, which can affect the accuracy of event camera optical system parameter calibration. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for calibrating the optical system parameters of an event camera, which solves the problem that existing calibration technologies are not applicable to event stream data generated by event cameras and provides optical system parameters for event cameras in visual posture measurement. The present invention uses a traditional calibration plate and Zhang Zhengyou calibration algorithm to achieve optical system parameter calibration for event cameras. At the same time, the present invention introduces weight information into the event camera for the first time, which makes up for the inherent defect of the event camera's lack of grayscale information, realizes the weighted application of the event camera, and improves the calibration accuracy in the field of event camera optical system parameter calibration using a calibration plate. In addition, the present invention no longer relies on other electronic devices, and the application scenarios are more flexible and convenient.

[0006] The above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0007] The method for calibrating the optical system parameters of an event camera includes the following steps:

[0008] Step 1: Data preprocessing:

[0009] In a good experimental environment, ensure that the circular pattern calibration plate is fully imaged by the event camera and the circular pattern is completely within the camera's field of view. The event camera collects event data of the circular pattern calibration plate. The event generation model can be expressed by formula (1):

[0010] e k =(x k ,y k ,t k ,p k ) (1)

[0011] In formula (1), (x, y) is the pixel coordinate, t is the timestamp, and p k It's polarity, Indicates the increase or decrease of logarithmic intensity, ΔL(t k ) is the change in logarithmic intensity, ΔL(t k )=L(t k )-L(t k-1 ), c is the set threshold;

[0012] A single event point in the event stream can be expressed by formula (2):

[0013]

[0014] Where δ is the Dirichlet function. Considering that a single event carries very little information and is susceptible to noise interference, the event data stream is processed using a fixed time window slicing method. This method slices the spatiotemporal event stream data through a fixed time window, and then processes the event slices in various representation forms. Each pixel is represented by the number of events in the event slice, that is, the spatiotemporal data stream is mapped to the (x, y) plane with time characteristics, and converted into a traditional two-dimensional image representation method. This process can be expressed by formula (3)

[0015]

[0016] in is an image created within a fixed time window T.

[0017] Set a fixed time window T1, T1 satisfies formula (4),

[0018] T1=min T {T|Hough(I(x,y,T))≥N th} (4)

[0019] Where Hough(·) represents the gradient-based Hough circle detection algorithm, N th represents the minimum threshold for detecting circles, and I(x, y, T) represents the converted image. After the fixed time window T1 is set, the collected circular pattern calibration plate event stream data is sliced and segmented using the fixed time window method. The segmented data is then overlaid and transformed into a two-dimensional event frame image P1.

[0020] Step 2: Initial positioning of the circle center:

[0021] Perform pixel detection on the converted image P1. Obtain non-zero pixels in the image P1;

[0022] The gradient-based Hough circle detection method is used for all non-zero pixels. The theoretical description is as follows:

[0023] First, the Sobel operator is used to calculate the horizontal and vertical gradient components of all pixels (x, y) in the image I(x, y), which is expressed by formula (5):

[0024]

[0025] Among them S x and S y is the Sobel convolution kernel, which is expressed by formula (6):

[0026]

[0027] Then calculate the straight line where the gradient direction of each pixel point (x, y) lies, as shown in formula (7):

[0028] G y (x,y)·aG x (x,y)·b=G y (x,y)·xG x (x,y)·y (7)

[0029] Construct a two-dimensional parameter space (a, b) about the circle center and an accumulator H(a, b). All points (a, b) that satisfy equation (7) are accumulated in the parameter space to form a candidate circle center. Finally, the point corresponding to the maximum value of the accumulator H(a, b) is selected as the circle center (a0, b0):

[0030] (a0,b0)=max(H(a,b)) (8)

[0031] Calculate the distance r from all pixels to the center of the circle, and statistically calculate the histogram of the distance, and select the radius r0 with the highest frequency:

[0032]

[0033] This gives us the initial center (a0, b0) and radius r0.

[0034] Step 3: Accurate data processing:

[0035] With the detected initial circle center as the center point and the radius as the distance, a rectangular area is drawn on the image, and the data of the original event data stream corresponding to this rectangular area is propose, Expressed by formula (10)

[0036]

[0037] in Indicates A rectangular area with a center and a side length of 2r;

[0038] Event stream data According to its time information, intercept the event stream data with a time interval of T2 (T1>T2) The purpose of selecting a shorter time interval T2 when satisfying formula (11) is to reduce the impact of time uncertainty on subsequent circle center positioning;

[0039]

[0040] right The event stream data is accumulated in the time dimension and reconverted into the frame image P2.

[0041] Step 4: Secondary positioning of the circle center:

[0042] Perform gradient-based Hough circle detection on P2 to obtain non-zero pixels in image P2. Use the slobe operator to solve the gradient direction of all pixels, solve the straight line and normal based on the gradient direction, calculate the intersection coordinates of the normal and the value of the intersection accumulation in the parameter space, and extract the parameter space accumulator H(a,b).

[0043] Set the threshold Λ and extract the accumulator H0(a,b) that meets the threshold. H0(a,b) satisfies formula (12):

[0044] H0(a,b)=(H(a,b)≥Λ) (12)

[0045] Count all candidate points that meet the threshold condition under the current accumulator (a k ,b k ), and the corresponding accumulator value G(a k ,b k );

[0046] G(a k ,b k ) as a candidate point (a k ,b k ) and use formula (13) to find the coordinates of the circle center:

[0047]

[0048] In formula (13), H0 represents the accumulator that meets the threshold condition, G(a k ,b k ) represents the candidate circle (a k ,b k), (a0, b0) represents the coordinates of the circle center to be obtained, and this method is used to complete the secondary positioning of the circle center.

[0049] Step 5: Solve the internal parameters of the event camera:

[0050] Sort the extracted circle centers according to the order of feature points in Zhang Zhengyou's calibration algorithm;

[0051] Generate the three-dimensional coordinates of the center of the circle on the calibration plate plane in the world coordinate system;

[0052] According to the pinhole imaging principle, the homography matrix is constructed using Zhang Zhengyou calibration algorithm;

[0053] Adjust the orientation of the calibration plate to ensure that it presents different postures under the event camera, at least three postures, and repeat all the above steps to obtain multiple homography matrices. Solve them jointly to solve the intrinsic parameters of the event camera and realize the calibration of the event camera optical system parameters.

[0054] The beneficial effects of the present invention are:

[0055] First, it eliminates the reliance on dedicated electronic equipment for calibration of event camera optical system parameters, enabling flexible applications and a wider range of application scenarios.

[0056] Second, weight information is introduced into the event camera for the first time, which solves the inherent defect of the event camera without grayscale information. The weight is used to reduce the influence of noise, improve the accuracy of circle center extraction, and further improve the accuracy of event camera optical system parameter calibration.

[0057] Third, through precise data processing and a two-stage center positioning method, the impact of time inaccuracy on center positioning is reduced, the center positioning accuracy is improved, and the offset of the event camera principal point calibration result is further reduced;

[0058] Fourth, under non-ideal motion conditions (such as speed change and jitter), the present invention can still obtain stable event camera optical system parameters, and the reprojection error is stable at around 0.65 pixels, further meeting the needs of camera calibration in high-speed motion scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings described herein are used to provide further understanding of the present invention and constitute a part of this application. The illustrative examples of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0060] Figure 1 This is a flowchart of the overall process of calibrating the optical system parameters of the event camera of the present invention;

[0061] Figure 2The image is obtained by converting the event stream data of the circular pattern calibration plate of the present invention through a fixed time window;

[0062] Figure 3 Schematic diagram of the Hough circle detection method based on gradient of the present invention;

[0063] Figure 4 This is an image of the present invention converting event stream data after a fixed time into new event stream data after further shortening the time;

[0064] Figure 5 It is the circle center sorting diagram of the present invention;

[0065] Figure 6 A diagram showing the physical experiment setup of the present invention. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0067] See also Figures 1 to 6 As shown, the event camera optical system parameter calibration method of the present invention is based on the traditional camera calibration process. The traditional camera calibration process is mainly divided into six parts: grayscale image conversion, image detection, feature point extraction, generation of points in the world coordinate system, homography matrix solution, and camera parameter solution. Among them, feature point extraction occupies a significant position in the traditional camera calibration process. The accuracy of feature point extraction directly affects the accuracy of calibration. Therefore, the accuracy of feature point extraction is used as a reference for calibration. Feature point extraction also occupies an important position in the event camera optical system parameter calibration. The present invention focuses on the method of event camera feature point extraction, especially the method of circular feature point extraction, and proposes a circular feature point extraction algorithm suitable for event cameras, thereby completing the event camera optical system parameter calibration. The event camera optical system parameter calibration process includes five parts: data preprocessing, initial positioning of the circle center, precise data processing, secondary positioning of the circle center, and solution of event camera internal parameters.

[0068] Specific calibration process steps, such as Figure 1 As shown:

[0069] Step 1: Calibrate the optical system parameters of the event camera with the help of a symmetrical circular pattern calibration plate. Under a good experimental environment, ensure that the circular pattern calibration plate is fully imaged by the event camera and the circular pattern is completely within the camera's field of view. The event camera collects event data from the circular pattern calibration plate. The event generation model can be expressed by formula (1):

[0070] e k =(x k ,y k ,t k ,p k ) (1)

[0071] In formula (1), (x, y) is the pixel position, t is the timestamp information of the data generation, and p k It's polarity, Indicates the increase or decrease of logarithmic intensity, ΔL(t k ) is the change in logarithmic intensity, ΔL(t k )=L(t k )-L(t k-1 ), c is the set threshold;

[0072] A single event point in the event stream can be expressed by formula (2):

[0073]

[0074] where δ is the Dirichlet function.

[0075] Step 2: Considering that a single event carries very little information and is susceptible to noise interference, the event data stream is processed using a fixed time window slicing method. This method slices the spatiotemporal event stream through a fixed time window, and then processes the event slices in various representation forms. Each pixel is represented by the number of events in the event slice, that is, the spatiotemporal data stream is mapped to the (x, y) plane with time characteristics, and converted into a traditional two-dimensional image representation method. This process can be expressed by formula (3):

[0076]

[0077] in is the image created within a fixed time window T;

[0078] Set a fixed time window T1, T1 satisfies formula (4),

[0079] T1=min T {T|Hough(I(x,y,T))≥N th} (4)

[0080] Where Hough(·) represents the gradient-based Hough circle detection algorithm, N th Represents the minimum threshold for detecting the number of circles, and I(x, y, T) represents the converted image. After the fixed time window T1 is set, the collected circular pattern calibration plate event stream data is sliced and segmented using the fixed time window method, and the data within the segmentation is superimposed and converted into a two-dimensional event frame image P1. The two-dimensional event frame image is as follows: Figure 2 shown.

[0081] Step 3: Detect pixels of the converted image P1 to obtain non-zero pixels in the image P1.

[0082] Step 4: Use the gradient-based Hough circle detection method for all non-zero pixels. The theoretical description is as follows:

[0083] First, the Sobel operator is used to calculate the horizontal and vertical gradient components of all pixels (x, y) in the image I(x, y), which is expressed by formula (5):

[0084]

[0085] Among them S x and S y is the Sobel convolution kernel, which is expressed by formula (6):

[0086]

[0087] Then calculate the straight line where the gradient direction of each pixel point (x, y) lies, as shown in formula (7):

[0088] G y (x,y)·aG x (x,y)·b=G y (x,y)·xG x (x,y)·y (7)

[0089] Construct a two-dimensional parameter space (a, b) about the circle center and an accumulator H(a, b). All points (a, b) that satisfy equation (7) are accumulated in the parameter space to form a candidate circle center. Finally, the point corresponding to the maximum value of the accumulator H(a, b) is selected as the circle center (a0, b0):

[0090] (a0,b0)=max(H(a,b)) (8)

[0091] Calculate the distance r from all pixels to the center of the circle, and statistically calculate the histogram of the distance, and select the radius r0 with the highest frequency:

[0092]

[0093] Thus, the initial center (a0, b0) and radius r0 are obtained. The principle of the gradient-based Hough circle detection method is as follows: Figure 3 shown.

[0094] Step 5: Use the detected initial circle center as the center point and the radius as the distance to define a rectangular area on the image, and convert the data of the original event data stream corresponding to this rectangular area into propose, Expressed by formula (10)

[0095]

[0096] in Indicates A rectangular area with a center and a side length of 2r.

[0097] Step 6: Event Stream Data According to its time information, intercept the event stream data with a time interval of T2 (T1>T2) The purpose of selecting a shorter time interval T2 when satisfying formula (11) is to reduce the impact of time uncertainty on subsequent circle center positioning;

[0098]

[0099] right The event stream data is accumulated in the time dimension and converted back into the frame image P2. The frame image P2 is as follows: Figure 4 shown.

[0100] Step 7: Perform gradient-based Hough circle detection on the image again. Obtain non-zero pixels in image P2, use the slobe operator to solve the gradient direction of all pixels, solve the straight line and normal based on the gradient direction, calculate the intersection coordinates of the normal and the intersection cumulative value in the parameter space, and extract the parameter space accumulator H(a,b);

[0101] Set the threshold Λ and extract the accumulator H0(a,b) that meets the threshold. H0(a,b) satisfies formula (12):

[0102] H0(a,b)=(H(a,b)≥Λ) (12)

[0103] Count all candidate points that meet the threshold condition under the current accumulator (a k ,b k ), and the corresponding accumulator value G(a k ,b k ).

[0104] Step 8: G(a k ,bk ) as a candidate point (a k ,b k ) and use formula (13) to find the coordinates of the circle center:

[0105]

[0106] In formula (13), H0 represents the accumulator that meets the threshold condition, G(a k ,b k ) represents the candidate circle (a k ,b k ), (a0, b0) represents the coordinates of the circle center to be obtained, and this method is used to complete the secondary positioning of the circle center.

[0107] Step 9: Sort the extracted circle centers according to the order of feature points in Zhang Zhengyou's calibration algorithm. The sorting diagram is as follows: Figure 5 shown.

[0108] Step 10: Generate the three-dimensional coordinates of the center of the circle on the calibration plate plane in the world coordinate system.

[0109] Step 11: Based on the pinhole imaging principle, use Zhang Zhengyou calibration algorithm to construct the homography matrix.

[0110] Step 12: Adjust the orientation of the calibration plate to ensure that it presents different postures under the event camera, at least three postures, and repeat all the above steps to obtain multiple homography matrices, solve them jointly, and then solve the intrinsic parameters of the event camera to achieve calibration of the event camera optical system parameters.

[0111] The above method calibrates the optical system parameters of an event camera. It maintains a certain level of calibration accuracy even under complex conditions such as noise interference and non-ideal motion. Notably, this method does not rely on a specific calibration plate or complex motion control equipment; it can be directly calibrated using a circular pattern calibration plate, greatly enhancing the flexibility of its application scenarios.

[0112] like Figure 6 As shown in the figure, the present invention uses a traditional circular pattern calibration plate and utilizes a slide rail to arbitrarily control the relative movement speed between the calibration plate and the event camera. The event camera collects event stream data, which is intercepted and superimposed into an image through fixed time window slicing. The center of the image is extracted to solve the optical system parameters of the event camera. The optical system parameters of the event camera are shown in Table 1. The numerical values of the parameters show that this method is feasible in practical application scenarios.

[0113] Table 1 Event camera optical system parameters

[0114] <![CDATA[F u (pixel)]]> <![CDATA[F v (pixel)]]> <![CDATA[U0(pixel)]]> <![CDATA[V0(pixel)]]> RMSE(pixel) 1711.386154 1712.05952 641.695026 358.6797251 0.657694732

[0115] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements to the present invention are intended to fall within the scope of protection of the present invention.

Claims

1. A method for calibrating optical system parameters of an event camera, characterized by: The following steps are involved: Step 1: Data preprocessing; Step 2: Initial positioning of the circle center; Step 3: Accurate data processing: With the detected initial circle center as the center point and the radius as the distance, a rectangular area is drawn on the image, and the data of the original event data stream corresponding to this rectangular area is propose, It is expressed by the following formula in Indicates A rectangular area with a center and a side length of 2r; Event stream data According to its time information, intercept the event stream data with a time interval of T2 (T1>T2) The purpose of selecting a shorter time interval T2 is to reduce the impact of time uncertainty on subsequent circle center positioning when the following formula is satisfied: right The event stream data is accumulated in the time dimension and converted back into the frame image P2; Step 4: Secondary positioning of the circle center; Step 5: Solve the internal parameters of the event camera.

2. The method for calibrating optical system parameters of an event camera according to claim 1, wherein: The data preprocessing described in step 1 is: Ensure that the circular pattern calibration plate is fully imaged by the event camera and the circular pattern is entirely within the camera's field of view. The event camera collects event data of the circular pattern calibration plate. The event generation model is expressed as follows: e k =(x k ,y k ,t k ,p k ) Where (x, y) is the pixel coordinate, t is the timestamp, and p k It's polarity, Indicates the increase or decrease of logarithmic intensity, ΔL(t k ) is the change in logarithmic intensity, ΔL(t k )=L(t k )-L(t k-1 ), c is the set threshold; A single event point in the event stream is represented by the following formula: Where: δ is the Dirichlet function; the event data stream is processed using a fixed time window slicing method, slicing the spatiotemporal event stream data through a fixed time window, and then image processing is performed on the event slices in various representation forms. Each pixel is represented by the number of events in the event slice, that is, the spatiotemporal data stream is mapped to the (x, y) plane based on its time characteristics and converted into a two-dimensional image representation method; this process is represented by the following formula: in is the image created within a fixed time window T; Set a fixed time window T1, T1 satisfies the following formula, T1=min T {T|Hough(I(x,y,T))≥N th } Where Hough(·) represents the gradient-based Hough circle detection algorithm, N th Represents the minimum threshold for detecting the number of circles, and I(x, y, T) represents the image after conversion. After the fixed time window T1 is set, the collected circular pattern calibration plate event stream data is sliced and segmented using the fixed time window method, and the data within the segmentation is superimposed and converted into a two-dimensional event frame image P1.

3. The method for calibrating optical system parameters of an event camera according to claim 1, wherein: The initial positioning of the circle center described in step 2 is specifically: Perform pixel detection on the converted image P1; obtain non-zero pixels in the image P1; The gradient-based Hough circle detection method is used for all non-zero pixels, as described below: First, the Sobel operator is used to calculate the horizontal and vertical gradient components of all pixels (x, y) in the image I(x, y), which is expressed as follows: G x (x,y)=I*S x G y (x,y)=I*S y Among them S x and S y is the Sobel convolution kernel, which is expressed as follows: Then calculate the straight line where the gradient direction of each pixel point (x, y) lies, as follows: G y (x,y)·a-G x (x,y)·b=G y (x,y)·x-G x (x,y)·y Construct a two-dimensional parameter space (a, b) about the circle center and an accumulator H(a, b). All points (a, b) that satisfy the above equation are accumulated in the parameter space to form a candidate circle center. Finally, the point corresponding to the maximum value of the accumulator H(a, b) is selected as the circle center (a0, b0): (a0,b0)=max(H(a,b)) Calculate the distance r from all pixels to the center of the circle, and statistically calculate the histogram of the distance, and select the radius r0 with the highest frequency: This gives us the initial center (a0, b0) and radius r0.

4. The method for calibrating optical system parameters of an event camera according to claim 1, wherein: The secondary positioning of the circle center described in step 4 is specifically as follows: Perform gradient-based Hough circle detection on P2 to obtain non-zero pixels in image P2. Use the slobe operator to solve the gradient direction of all pixels, solve the straight line and normal based on the gradient direction, calculate the intersection coordinates of the normal and the value of the intersection accumulation in the parameter space, and extract the parameter space accumulator H(a,b); Set the threshold Λ and extract the accumulator H0(a,b) that meets the threshold. H0(a,b) satisfies the following formula: H0(a,b)=(H(a,b)≥Λ) Count all candidate points that meet the threshold condition under the current accumulator (a k ,b k ), and the corresponding accumulator value G(a k ,b k ); G(a k ,b k ) as a candidate point (a k ,b k ) weight, use the following formula to find the coordinates of the circle center Where H0 represents the accumulator that meets the threshold condition, G(a k ,b k ) represents the candidate circle (a k ,b k ) corresponds to the accumulator value on the circle, (a0, b0) represents the coordinates of the circle center to be obtained, and the secondary positioning of the circle center is completed.

5. The method for calibrating optical system parameters of an event camera according to claim 1, wherein: The solution to the intrinsic parameters of the event camera described in step 5 is as follows: the extracted circle centers are sorted according to the order of the feature points in the Zhang Zhengyou calibration algorithm; the three-dimensional coordinates of the circle center on the calibration plate plane in the world coordinate system are generated; based on the pinhole imaging principle, the Zhang Zhengyou calibration algorithm is used to construct a homography matrix; the orientation of the calibration plate is adjusted to ensure that the calibration plate presents different postures under the event camera, at least three postures, and all the above steps are repeated to obtain multiple homography matrices, which are solved jointly to solve the intrinsic parameters of the event camera and achieve calibration of the event camera optical system parameters.

Images