Star point detection algorithm based on event camera

Abstract

The invention relates to the technical field of image processing and sensors, in particular to a star point detection algorithm based on an event camera, and the algorithm comprises the following steps: S1, carrying out the preprocessing and noise filtering of an event stream; carrying out noise filtering on an original event stream of an event camera, and removing noise events which do not conform to space-time correlation through space-time density analysis and adaptive time window adjustment; s2, star point centroid extraction and trajectory analysis; and carrying out time slice interception and clustering processing on the event stream after noise reduction, calculating a weighted centroid of a star point event cluster, and fitting a motion trail of a star point. High precision of star point detection is ensured through calculation and dynamic adjustment, and noise is removed by setting a threshold according to camera parameters; the angular velocity is utilized to determine time slices, and the centroid is calculated and the trajectory is fitted through clustering and weighting, so that the detection precision is far higher than that of a traditional method, and star points can be accurately identified in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical fields of image processing and sensor technology, and particularly to a star point detection algorithm based on an event camera. Background Art

[0002] In fields such as astronomical observation and satellite navigation, star point detection is a key technology. Traditional star point detection methods mainly rely on optical cameras or CCD / CMOS cameras. These cameras work based on the principle of frame image acquisition and have many limitations. In environments with low light, such as at night or during deep space observation, their frame rate is low, making it difficult to capture fast-moving star points; at the same time, image noise will increase significantly, seriously affecting the recognition accuracy of star points. In high dynamic range scenarios, such as when the camera moves rapidly or the brightness of star points changes sharply, traditional cameras cannot work effectively either. This is because their image acquisition frequency is limited and they cannot respond to the rapid changes in the scene in a timely manner, resulting in blurred or lost star points, which greatly limits the development of related fields.

[0003] As a new type of sensor, an event camera adopts an event-driven working mode. When the brightness change in the scene exceeds a specific threshold, the event camera will record a tiny event, and its sampling frequency can be as high as millions, much higher than that of traditional image cameras. This makes the event camera have unique advantages in low light and high dynamic range scenarios and become an ideal tool for star point detection. However, the original data format of the event camera is very different from traditional image data. How to efficiently extract star point information from these data and achieve accurate detection and tracking has become an urgent problem to be solved. Therefore, it is of great practical significance to design an algorithm that can process event camera data and achieve accurate star point detection. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above problems and provide a star point detection algorithm based on an event camera. To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] A star point detection algorithm based on an event camera includes the following steps:

[0006] Step S1: Preprocessing of the event stream and noise filtering;

[0007] Perform noise filtering on the original event stream of the event camera, and remove noise events that do not conform to spatio-temporal correlation through spatio-temporal density analysis and adaptive time window adjustment;

[0008] The step of noise filtering for the original event stream of the event camera is crucial. In actual star point detection scenarios, the data collected by the event camera often contains a large amount of noise, such as false events generated by background interference, sensor errors, etc. By performing noise filtering through spatio-temporal density analysis and adaptive time window adjustment, these noise events can be effectively identified and removed. This makes the subsequent processed data cleaner, improving the quality and usability of the data. The event stream after removing noise provides a more reliable basis for subsequent star point detection, reducing the interference of noise on star point feature extraction and positioning, greatly enhancing the anti-interference ability of the algorithm in complex environments, and laying a solid foundation for accurate star point detection.

[0009] Step S2: Extraction and trajectory analysis of star point centroids;

[0010] Perform time slicing and clustering processing on the denoised event stream, calculate the weighted centroid of the star point event cluster, and fit the motion trajectory of the star point.

[0011] Processing the denoised event stream, including intercepting time slices, clustering, calculating the weighted centroid, and fitting the motion trajectory, is a key link in achieving accurate star point detection and tracking. Intercepting appropriate time slices can divide the continuous event stream into segments that are convenient for processing, making the star point features in each segment more obvious. The clustering operation can gather the events belonging to the same star point together, facilitating subsequent analysis. Calculating the weighted centroid takes into account the importance and distribution of events, and can more accurately determine the position of the star point. Fitting the motion trajectory can obtain the motion trend and law of the star point, realizing effective tracking of the star point. This series of operations enables the algorithm to accurately locate the position of the star point and master its motion information in a complex starry sky background, meeting the actual needs of high-precision star point detection and tracking, and providing strong support for applications such as starry sky monitoring and navigation.

[0012] Furthermore, in step S1, it includes the following steps:

[0013] Step S11: Spatio-temporal density calculation;

[0014] Calculate the spatio-temporal density D(e i ) of each event, and the formula is:

[0015]

[0016] If D(e i ) < T density , then eliminate this event.

[0017] Calculating the spatio-temporal density of each event and removing noise events based on the density threshold is an effective noise identification method based on the spatio-temporal relationship of events. In the starry sky environment, events generated by star points have a certain spatio-temporal aggregation, while noise events are relatively dispersed. By calculating the spatio-temporal density, the aggregation degree of events can be quantified, so as to accurately identify noise events. After removing noise events, the data in the event stream is more concentrated on the real star point information, improving the purity of the data. This not only helps to more accurately extract star point features subsequently, but also reduces the computational complexity of the algorithm, improves the running efficiency of the algorithm, and thus optimizes the accuracy and speed of star point detection.

[0018] Step S12: Adaptive time window adjustment;

[0019] The time window Δt is adaptively adjusted according to the density of the event stream. The formula is:

[0020]

[0021] where α is set to 0.5 - 1.5 according to the environmental dynamics.

[0022] The size of the time window is adaptively adjusted according to the density of the event stream, enabling the algorithm to better adapt to event streams under different dynamic conditions. In starry sky observations, the movement speed of star points and the frequency of event generation may change due to various factors. When the event stream is denser, appropriately reducing the time window can more precisely capture the features of star points; while when the event stream is sparser, increasing the time window can ensure that sufficient event information is captured. This adaptive adjustment mechanism further optimizes the noise filtering effect, avoids problems such as incomplete noise filtering or loss of useful information caused by unreasonable time window settings, ensures the stable operation of the algorithm under various dynamic conditions, and improves the robustness and reliability of the algorithm.

[0023] Step S13: Secondary filtering of initial noise;

[0024] If the average density within the initial T init time of the event stream is lower than the threshold T init_density , then the events within this time period are discarded.

[0025] Checking the average density within the initial time of the event stream and discarding the period events below the threshold can specifically remove the noise that is difficult to filter by conventional methods in the initial part of the event stream. In the initial stage of the event camera startup, some unstable noise events may occur due to various reasons, and these noise events may affect the subsequent star point detection. By checking the average density within the initial time period, these abnormal noise events can be identified and discarded. This operation improves the noise filtering process, further enhances the data quality, reduces the impact of noise on star point detection, and enhances the reliability and accuracy of the algorithm.

[0026] Further, in step S2, the following steps are included:

[0027] Step S21: Time slice interception;

[0028] Determine the time slice size T according to the star point angular velocity ω slice , the formula is:

[0029]

[0030] where θ max is the preset maximum displacement angle;

[0031] Determining the time slice size according to the star point angular velocity provides appropriate data segments for subsequent clustering and centroid calculation. The angular velocity of the star point reflects its movement speed, and star points with different angular velocities have different displacements within the same time. If the time slice is set unreasonably, it may cause the movement of the star point within the time slice to be too large or too small, thus affecting the accuracy of clustering and centroid calculation. By determining the time slice size according to the star point angular velocity, it can ensure that the movement of the star point within each time slice is relatively stable, making the clustering and centroid calculation more accurate. This helps to more accurately determine the position of the star point, improve the accuracy of star point detection, and provide a more reliable basis for subsequent star point tracking and analysis.

[0032] Step S22: Mean shift clustering;

[0033] Perform mean shift clustering on the event stream within the time slice, with the kernel function being the Gaussian kernel and the kernel bandwidth h being dynamically adjusted:

[0034]

[0035] where, Var(C k ) is the coordinate variance of the cluster C k .

[0036] Performing mean shift clustering on the event stream within a time slice and dynamically adjusting the kernel bandwidth can effectively cluster events belonging to the same star point together. In a complex starry sky environment, events generated by star points may be interfered by various factors, resulting in a complex distribution of events. The mean shift clustering algorithm can automatically find the aggregation center of events according to the distribution characteristics of events, thus dividing events belonging to the same star point into the same cluster. Dynamically adjusting the kernel bandwidth can adaptively adjust the clustering range according to the distribution of events, making the clustering result more accurate. This operation can clearly distinguish different star points, providing a basis for accurately calculating the centroid of star points and enhancing the accuracy and reliability of star point detection.

[0037] Step S23: Weighted centroid calculation

[0038] Calculate the weighted centroid (x k , y k ) of each event cluster, with the weight ω i Based on the timestamp distribution:

[0039]

[0040] where μ t is the mean of the timestamps of the events within the cluster.

[0041] Calculating the weighted centroid of each event cluster based on the timestamp distribution takes into account the distribution characteristics of events in time. In actual star point detection, events generated by star points may not be evenly distributed in time, and events in certain time periods may better reflect the true position of the star point. By weighting the events, these important events can be highlighted, thus more accurately determining the position of the star point. This method is more accurate than simple average centroid calculation, improving the accuracy and reliability of star point positioning and providing stronger support for the precise tracking and analysis of star points.

[0042] Furthermore, in step S11, the parameter settings for the spatio-temporal density calculation are as follows:

[0043] σ s is the standard deviation of the spatial Gaussian kernel, and its value is determined according to the pixel resolution R pixel of the event camera: σ s = γ·R pixel , γ ∈ [0.5, 2.0];

[0044] σ t is the standard deviation of the temporal Gaussian kernel, and its value is set according to the dynamic range D event of the event camera: λ ∈ [0.1, 0.5];

[0045] The density threshold T densityDynamically adjusted by the following formula: η ∈ [0.2, 0.8].

[0046] Determine the relevant parameters in the spatio-temporal density calculation according to the pixel resolution and dynamic range of the event camera, and dynamically adjust the density threshold, so that the algorithm can adapt to event cameras with different performances. Different event cameras have different pixel resolutions and dynamic ranges, and these parameters will affect the spatio-temporal distribution and density of events. By determining the relevant parameters and density threshold according to the actual parameters of the camera, the algorithm can better adapt to the data collected by different cameras. This enhances the generality of the algorithm, enables the algorithm to be used on various types of event cameras, and expands the application range of the algorithm. At the same time, dynamically adjusting the density threshold can optimize the noise filtering effect according to the actual situation of the data, further improve the accuracy of star point detection, and ensure that the algorithm can obtain good detection results under different camera conditions.

[0047] Further, in step S12, the dynamic setting rule of the adjustment coefficient α in the adaptive time window adjustment is:

[0048] If the environmental dynamicity index V dynamic defined as the variance of the time intervals between adjacent events in the event stream) is greater than the threshold V th , then α = 0.5;

[0049] If V dynamic ≤ V th , then α = 1.5 - 0.5.

[0050] Dynamically set the adjustment coefficient α in the adaptive time window adjustment according to the environmental dynamicity index (variance of the time intervals between adjacent events), so that the algorithm can better adapt to different dynamic environments. In actual starry sky observations, the dynamicity of the environment may change. For example, the movement speed of star points, the frequency of event generation, etc. may be affected by factors such as weather and camera movement. The variance of the time intervals between adjacent events can reflect the dynamicity of the environment. When the variance is large, it indicates that the environmental dynamicity is strong. At this time, appropriately reducing the adjustment coefficient α can make the time window more flexible and better capture the rapid changes of star points; when the variance is small, it indicates that the environment is relatively stable, and increasing the adjustment coefficient α can make the time window more stable and reduce unnecessary adjustments. This dynamic setting mechanism improves the noise filtering efficiency, ensures that the algorithm can operate stably in complex and changeable environments, and improves the adaptability and reliability of the algorithm.

[0051] Further, in step S13, the threshold T init_density of the initial noise secondary filtering is calculated by the following formula:

[0052]

[0053] Among them, T is the length of the initial time period, and its value range is [10 ms, 50 ms].

[0054] The threshold for the secondary filtering of the initial noise is determined by calculating the average density of events within the initial time period of the event stream, providing a reasonable basis for the initial noise filtering. When the event camera starts up, some unstable noise events may be generated, and the density of these noise events may be different from that of normal star point events. By calculating the average density within the initial time period, a reasonable threshold can be obtained to determine whether the initial event is a noise event. Secondary filtering based on this threshold can effectively remove the initial noise and improve the data quality. This operation further enhances the anti-noise ability of the algorithm, reduces the impact of noise on subsequent star point detection and analysis, and ensures the accuracy and reliability of the algorithm throughout the entire operation process.

[0055] Furthermore, in step S21, the real-time estimation method for the angular velocity ω of the star point is as follows:

[0056] The historical angular velocity data is predicted and updated using a Kalman filter.

[0057] The state equation is: ω k+1 = ω k + Δt · a k + w k ,

[0058] The observation equation is: z k = w k + v k ,

[0059] where a k is the angular acceleration, ω k and v k are the process noise and the observation noise respectively, and Q and R are the covariance matrices.

[0060] The historical angular velocity data is predicted and updated using a Kalman filter to real-time estimate the angular velocity ω of the star point, which can accurately obtain the angular velocity information of the star point. During the star point detection and tracking process, the angular velocity of the star point is an important parameter, which will affect the interception of time slices and the fitting of the star point motion trajectory. The Kalman filter is a filter based on the optimal estimation theory, which can accurately predict and update the angular velocity of the star point using historical data and current observations. By real-time estimating the angular velocity of the star point, accurate data can be provided for the interception of time slices to ensure that the motion of the star point within each time slice can be accurately captured. At the same time, the accurate angular velocity information also helps to more accurately fit the motion trajectory of the star point, improve the accuracy of the star point position and motion trajectory calculation, and provide more reliable support for the tracking and analysis of the star point.

[0061] Further, in step S21, the least squares formula for the motion trajectory fitting is as follows:

[0062]

[0063] where θ k is the angular position of the star point at the k-th time slice, and the solution formula is:

[0064]

[0065] Using the least squares method to fit the star point motion trajectory and giving the corresponding solution formula can accurately describe the motion trend of the star point. In star point detection and tracking, understanding the motion trajectory of the star point is very important for predicting the future position of the star point and analyzing the motion law of the star point. The least squares method is a commonly used curve fitting method. It finds the fitting curve that best suits the data by minimizing the sum of the squares of the errors between the observed values and the fitting curve. By using the least squares method to fit the motion trajectory of the star point, a mathematical model of the star point motion can be obtained, thereby accurately describing the motion trend of the star point. The given solution formula provides a specific calculation method for practical applications, facilitating the implementation of the star point motion trajectory fitting in an actual system. This operation can provide strong support for the tracking and prediction of star points, meet the actual needs of high-precision star point detection and analysis, and has important application value in the fields of starry sky monitoring, navigation, etc.

[0066] Further, in step S22, the iteration termination condition of the mean shift clustering is: the displacement of the clustering center where ∈ = 0.01 pixel, or the maximum number of iterations K max = 20.

[0067] Setting the iteration termination conditions of the mean shift clustering, including the displacement of the clustering center and the maximum number of iterations, ensures the convergence and stability of the clustering process. During the mean shift clustering process, it is necessary to continuously iterate and update the clustering center until a certain condition is met. Without reasonable termination conditions, the clustering process may fall into an infinite loop or produce unstable results. By setting the displacement of the clustering center as the termination condition, it can be ensured that the clustering center gradually converges to a stable position during the iteration process, avoiding excessive fluctuations of the clustering center. At the same time, setting the maximum number of iterations can prevent the clustering process from proceeding infinitely due to certain special circumstances, ensuring the operation efficiency of the algorithm. This series of termination condition settings makes the clustering process more stable and reliable, improves the accuracy and reliability of the star point clustering results, and provides a more accurate basis for the subsequent calculation of the star point centroid and motion trajectory fitting.

[0068] The advantages of the present invention are:

[0069] 1. Through calculation and dynamic adjustment, the present invention ensures high-precision star point detection. In the preprocessing of the event stream, thresholds are set according to camera parameters to remove noise; in the centroid extraction stage, angular velocity is used to determine time slices, and the centroid is calculated through clustering and weighting, and finally the trajectory is fitted. Each link is precisely calculated, making the detection accuracy far exceed traditional methods and enabling accurate identification of star points in complex environments.

[0070] 2. In the present invention, the event camera collects data at high speed, and the algorithm closely matches its characteristics. The Kalman filter is used to estimate the angular velocity in real time, quickly determine time slices, and efficiently process data. Each step is closely connected, enabling timely acquisition of star point position information, realizing real-time detection and tracking, and meeting application scenarios with extremely high real-time requirements such as starry sky monitoring and navigation.

[0071] 3. The present invention designs a multi-step processing mechanism for noise. First, preliminary noise reduction is performed based on spatio-temporal correlation, then the time window is adaptively adjusted according to the density of the event stream for further filtering, and finally the initial part of the noise is filtered again. These steps work together to comprehensively eliminate background noise and interference, and can accurately detect star points even in complex environments, with performance superior to traditional detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The drawings constituting a part of this application are used to provide a further understanding of this application, making other features, objectives, and advantages of this application more obvious. The schematic embodiments and descriptions of the drawings of this application are used to explain this application and do not constitute an improper limitation of this application.

[0073] In the drawings:

[0074] Figure 1 is a flowchart of a star point detection algorithm based on an event camera in Embodiment 1.

[0075] Figure 2 is a flowchart of the preprocessing and noise filtering of the event stream in Embodiment 1.

[0076] Figure 3 is a flowchart of the extraction of the star point centroid and trajectory analysis in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0078] The present invention will be described in detail and specifically through specific embodiments to better understand the present invention. However, the following embodiments do not limit the protection scope of the present invention.

[0079] Embodiment 1

[0080] As Figures 1-3 shown, a star point detection algorithm based on an event camera includes the following steps:

[0081] Step S1: Preprocessing and noise filtering of the event stream;

[0082] Perform noise filtering on the original event stream of the event camera. By spatio-temporal density analysis and adaptive time window adjustment, remove noise events that do not conform to spatio-temporal correlation.

[0083] The step of performing noise filtering on the original event stream of the event camera is crucial. In actual star point detection scenarios, the data collected by the event camera often contains a large amount of noise, such as false events generated by background interference, sensor errors, etc. By spatio-temporal density analysis and adaptive time window adjustment for noise filtering, these noise events can be effectively identified and removed. This makes the subsequent processed data cleaner, improving the quality and usability of the data. The event stream after removing noise provides a more reliable basis for subsequent star point detection, reducing the interference of noise on star point feature extraction and positioning, greatly enhancing the anti-interference ability of the algorithm in complex environments, and laying a solid foundation for accurately detecting star points.

[0084] Step S2: Extraction of star point centroids and trajectory analysis;

[0085] Perform time slicing and clustering processing on the denoised event stream, calculate the weighted centroid of the star point event clusters, and fit the motion trajectory of the star points.

[0086] Processing the denoised event stream, including intercepting time slices, clustering, calculating the weighted centroid, and fitting the motion trajectory, is a key link for realizing accurate star point detection and tracking. Intercepting appropriate time slices can divide the continuous event stream into segments that are convenient for processing, making the star point features in each segment more obvious. The clustering operation can gather events belonging to the same star point together for subsequent analysis. Calculating the weighted centroid takes into account the importance and distribution of events and can more accurately determine the position of the star point. And fitting the motion trajectory can obtain the motion trend and law of the star point, realizing effective tracking of the star point. This series of operations enables the algorithm to accurately locate the position of the star point and master its motion information in a complex starry sky background, meeting the actual needs of high-precision star point detection and tracking, and providing strong support for applications such as starry sky monitoring and navigation.

[0087] Further, in step S1, it includes the following steps:

[0088] Step S11: Spatiotemporal density calculation;

[0089] Calculate the spatiotemporal density D(e i ), and the formula is:

[0090]

[0091] If D(e i ) < T density , then eliminate this event.

[0092] Calculating the spatiotemporal density of each event and eliminating noise events based on the density threshold is an effective noise recognition method based on the spatiotemporal relationship of events. In the starry sky environment, the events generated by star points have a certain spatiotemporal aggregation, while noise events are relatively dispersed. By calculating the spatiotemporal density, the aggregation degree of events can be quantified, so as to accurately identify noise events. After eliminating noise events, the data in the event stream is more concentrated on the real star point information, improving the purity of the data. This not only helps to more accurately extract star point features subsequently, but also reduces the computational amount of the algorithm, improves the running efficiency of the algorithm, and thus optimizes the accuracy and speed of star point detection.

[0093] Step S12: Adaptive time window adjustment;

[0094] Adaptive adjust the time window Δt according to the density of the event stream, and the formula is:

[0095]

[0096] Among them, α is set to 0.5 - 1.5 according to the environmental dynamics.

[0097] Adaptive adjustment of the time window size according to the density of the event stream enables the algorithm to better adapt to the event stream under different dynamic conditions. In starry sky observation, the movement speed of star points and the frequency of event generation may change due to various factors. When the event stream is denser, appropriately reducing the time window can more finely capture the features of star points; while when the event stream is sparser, increasing the time window can ensure that sufficient event information is captured. This adaptive adjustment mechanism further optimizes the noise filtering effect, avoids the problem of incomplete noise filtering or loss of useful information caused by unreasonable time window setting, ensures the stable operation of the algorithm under various dynamic conditions, and improves the robustness and reliability of the algorithm.

[0098] Step S13: Secondary filtering of initial noise;

[0099] If the average density within the initial T init time of the event stream is lower than the threshold T init_density, then discard the events within this time period.

[0100] Check the average density within the initial time of the event stream and discard the period events below the threshold, which can specifically remove the noise that is difficult to filter by conventional methods in the initial part of the event stream. At the initial stage of the event camera startup, some unstable noise events may be generated due to various reasons, and these noise events may affect the subsequent star point detection. By checking the average density within the initial time period, these abnormal noise events can be identified and discarded. This operation improves the noise filtering process, further improves the data quality, reduces the impact of noise on star point detection, and enhances the reliability and accuracy of the algorithm.

[0101] Furthermore, in step S2, the following steps are included:

[0102] Step S21: Time slice interception;

[0103] Determine the time slice size T according to the star point angular velocity ω slice , the formula is:

[0104]

[0105] where θ max is the preset maximum displacement angle;

[0106] Determining the time slice size according to the star point angular velocity provides appropriate data segments for subsequent clustering and centroid calculation. The angular velocity of the star point reflects its movement speed, and star points with different angular velocities have different displacements in the same time. If the time slice is set unreasonably, it may cause the movement of the star point in the time slice to be too large or too small, thus affecting the accuracy of clustering and centroid calculation. By determining the time slice size according to the star point angular velocity, it can ensure that the movement of the star point within each time slice is relatively stable, making the clustering and centroid calculation more accurate. This helps to more accurately determine the position of the star point, improve the accuracy of star point detection, and provide a more reliable basis for subsequent star point tracking and analysis.

[0107] Step S22: Mean shift clustering;

[0108] Perform mean shift clustering on the event stream within the time slice, with the kernel function being the Gaussian kernel and the kernel bandwidth h being dynamically adjusted:

[0109]

[0110] where, Var(C k ) is the coordinate variance of cluster C k .

[0111] Performing mean shift clustering on the event stream within a time slice and dynamically adjusting the kernel bandwidth can effectively cluster events belonging to the same star point together. In a complex starry sky environment, events generated by star points may be interfered by various factors, resulting in a complex distribution of events. The mean shift clustering algorithm can automatically find the aggregation center of events according to the distribution characteristics of events, thus dividing events belonging to the same star point into the same cluster. Dynamically adjusting the kernel bandwidth can adaptively adjust the clustering range according to the distribution of events, making the clustering result more accurate. This operation can clearly distinguish different star points, providing a basis for accurately calculating the centroid of star points and enhancing the accuracy and reliability of star point detection.

[0112] Step S23: Weighted centroid calculation

[0113] Calculate the weighted centroid (x k , y k ) of each event cluster, with the weight ω i Based on the timestamp distribution:

[0114]

[0115] where μ t is the mean of the timestamps of the events within the cluster.

[0116] Calculating the weighted centroid of each event cluster based on the timestamp distribution takes into account the distribution characteristics of events in time. In actual star point detection, events generated by star points may not be evenly distributed in time, and events in certain time periods may better reflect the true position of the star point. By weighting the events, these important events can be highlighted, thus more precisely determining the position of the star point. This method is more accurate than simple average centroid calculation, improving the accuracy and reliability of star point positioning and providing stronger support for the precise tracking and analysis of star points.

[0117] Furthermore, in step S11, the parameter settings for the spatio-temporal density calculation are as follows:

[0118] σ s is the standard deviation of the spatial Gaussian kernel, and its value is determined according to the pixel resolution R pixel of the event camera: σ s = γ·R pixel , γ ∈ [0.5, 2.0];

[0119] σ t is the standard deviation of the temporal Gaussian kernel, and its value is set according to the dynamic range D event of the event camera: λ ∈ [0.1, 0.5];

[0120] The density threshold T densityDynamically adjusted by the following formula: η ∈ [0.2, 0.8].

[0121] Determine the relevant parameters in the spatio-temporal density calculation according to the pixel resolution and dynamic range of the event camera, and dynamically adjust the density threshold, so that the algorithm can adapt to event cameras with different performances. Different event cameras have different pixel resolutions and dynamic ranges, and these parameters will affect the spatio-temporal distribution and density of events. By determining the relevant parameters and density threshold according to the actual parameters of the camera, the algorithm can better adapt to the data collected by different cameras. This enhances the generality of the algorithm, enables the algorithm to be used on various types of event cameras, and expands the application range of the algorithm. At the same time, dynamically adjusting the density threshold can optimize the noise filtering effect according to the actual situation of the data, further improve the accuracy of star point detection, and ensure that the algorithm can obtain good detection results under different camera conditions.

[0122] Further, in step S12, the dynamic setting rule of the adjustment coefficient α in the adaptive time window adjustment is:

[0123] If the environmental dynamic index V dynamic (defined as the variance of the time intervals between adjacent events in the event stream) is greater than the threshold V th , then α = 0.5;

[0124] If V dynamic ≤ V th , then α = 1.5 - 0.5.

[0125] Dynamically set the adjustment coefficient α in the adaptive time window adjustment according to the environmental dynamic index (variance of the time intervals between adjacent events), so that the algorithm can better adapt to different dynamic environments. In actual starry sky observations, the dynamic nature of the environment may change. For example, the movement speed of star points, the frequency of event generation, etc. may be affected by factors such as weather and camera movement. The variance of the time intervals between adjacent events can reflect the dynamic nature of the environment. When the variance is large, it indicates that the environment is more dynamic. At this time, appropriately reducing the adjustment coefficient α can make the time window more flexible and better capture the rapid changes of star points; when the variance is small, it indicates that the environment is relatively stable, and increasing the adjustment coefficient α can make the time window more stable and reduce unnecessary adjustments. This dynamic setting mechanism improves the noise filtering efficiency, ensures that the algorithm can operate stably in complex and changing environments, and improves the adaptability and reliability of the algorithm.

[0126] Further, in step S13, the threshold T init_density of the initial noise secondary filtering is calculated by the following formula:

[0127]

[0128] Among them, T is the length of the initial time period, and its value range is [10 ms, 50 ms].

[0129] The threshold for the second filtering of the initial noise is determined by calculating the average density of events within the initial time period of the event stream, providing a reasonable basis for the initial noise filtering. When the event camera starts up, some unstable noise events may be generated, and the density of these noise events may be different from that of normal star point events. By calculating the average density within the initial time period, a reasonable threshold can be obtained to determine whether the initial event is a noise event. Secondary filtering based on this threshold can effectively remove the initial noise and improve the data quality. This operation further enhances the anti-noise ability of the algorithm, reduces the impact of noise on subsequent star point detection and analysis, and ensures the accuracy and reliability of the algorithm throughout the operation process.

[0130] Furthermore, in step S21, the real-time estimation method for the angular velocity ω of the star point is as follows:

[0131] A Kalman filter is used to predict and update the historical angular velocity data.

[0132] The state equation is: ω k+1 = ω k + Δt·a k + w k ,

[0133] The observation equation is: z k = w k + v k ,

[0134] where a k is the angular acceleration, ω k and v k are the process noise and the observation noise respectively, and Q and R are covariance matrices.

[0135] Using a Kalman filter to predict and update the historical angular velocity data and real-time estimate the angular velocity ω of the star point can accurately obtain the angular velocity information of the star point. During the star point detection and tracking process, the angular velocity of the star point is an important parameter, which will affect the interception of time slices and the fitting of the star point motion trajectory. The Kalman filter is a filter based on the optimal estimation theory, which can accurately predict and update the angular velocity of the star point by using historical data and current observations. By real-time estimating the angular velocity of the star point, accurate data can be provided for the interception of time slices to ensure that the star point motion within each time slice can be accurately captured. At the same time, the accurate angular velocity information also helps to more accurately fit the motion trajectory of the star point, improve the accuracy of the star point position and motion trajectory calculation, and provide more reliable support for the tracking and analysis of the star point.

[0136] Further, in step S21, the least squares formula for the motion trajectory fitting is as follows:

[0137]

[0138] where θ k is the angular position of the star point at the k-th time slice, and the solution formula is:

[0139]

[0140] Using the least squares method to fit the star point motion trajectory and giving the corresponding solution formula can accurately describe the motion trend of the star point. In star point detection and tracking, understanding the motion trajectory of the star point is very important for predicting the future position of the star point and analyzing the motion law of the star point. The least squares method is a commonly used curve fitting method. It finds the fitting curve that best suits the data by minimizing the sum of the squares of the errors between the observed values and the fitting curve. By using the least squares method to fit the motion trajectory of the star point, a mathematical model of the star point motion can be obtained, thereby accurately describing the motion trend of the star point. The given solution formula provides a specific calculation method for practical applications, facilitating the implementation of the star point motion trajectory fitting in the actual system. This operation can provide strong support for the tracking and prediction of the star point, meet the actual needs of high-precision star point detection and analysis, and has important application value in the fields of starry sky monitoring, navigation, etc.

[0141] Further, in step S22, the iteration termination condition of the mean shift clustering is: the displacement of the clustering center where ∈ = 0.01 pixel, or the maximum number of iterations K max = 20.

[0142] Setting the iteration termination conditions of the mean shift clustering, including the displacement of the clustering center and the maximum number of iterations, ensures the convergence and stability of the clustering process. In the mean shift clustering process, it is necessary to continuously iterate and update the clustering center until a certain condition is met. Without reasonable termination conditions, the clustering process may fall into an infinite loop or produce unstable results. By setting the displacement of the clustering center as the termination condition, it can be ensured that the clustering center gradually converges to a stable position during the iteration process, avoiding excessive fluctuations of the clustering center. At the same time, setting the maximum number of iterations can prevent the clustering process from proceeding infinitely due to certain special circumstances, ensuring the operation efficiency of the algorithm. This series of termination condition settings makes the clustering process more stable and reliable, improves the accuracy and reliability of the star point clustering results, and provides a more accurate basis for the subsequent star point centroid calculation and motion trajectory fitting.

[0143] Embodiment 2

[0144] Example 1: Star Point Detection in Low Light Environment

[0145] Parameter Settings:

[0146] σ s = 1.5 × R pixel (R pixel = 0.1mm), σ t = 0.3ms.

[0147] α = 1.0, T init = 20ms, T init_density = 5events / ms.

[0148] Implementation Steps:

[0149] The event camera (model: Prophesee Gen4) collects the original event stream and transmits it to the embedded platform (NVIDIA Jetson AGX Xavier) through the USB 3.0 interface.

[0150] Perform spatio-temporal density filtering to eliminate events with a density lower than T density = 10.

[0151] Intercept the time slice T slice = 50m and perform mean shift clustering (β = 0.8).

[0152] Calculate the weighted centroid and fit the trajectory, and output the star point coordinates and angular velocity.

[0153] Test Results: Noise filtering efficiency: 92.3%, centroid positioning error: 0.08 pixels, processing delay: 8.5 milliseconds.

[0154] Example 3: Real-time Tracking in High Dynamic Scenarios

[0155] Parameter Adjustment:

[0156] α = 0.5 (high dynamicity), θ max = 2°, K max = 15.

[0157] Hardware Optimization: Use CUDA to accelerate mean shift clustering, allocate 256 threads per cluster for thread blocks, enable SIMD instructions to optimize spatio-temporal density calculation, and increase the throughput by 30%.

[0158] Application Scenario: In a satellite tracking system, the star point trajectory is output in real time, and the angular velocity estimation error < 0.5%.

[0159] The specific embodiments of the present invention have been described in detail above, but they are only examples, and the present invention is not equivalent to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, equivalent transformations and modifications made without departing from the spirit and scope of the present invention should all be covered within the scope of the present invention.

Claims

1. A star point detection algorithm based on event camera, characterized by: The following steps are involved: Step S1: event stream preprocessing and noise filtering; Noise filtering is performed on the original event stream of the event camera, and noise events that do not conform to the spatiotemporal correlation are removed through spatiotemporal density analysis and adaptive time window adjustment; Step S2: Extraction of star point centroid and trajectory analysis; The denoised event stream is subjected to time slicing and clustering, the weighted centroid of the star point event cluster is calculated, and the motion trajectory of the star point is fitted.

2. The star point detection algorithm based on event camera according to claim 1, characterized in that: In step S1, the following steps are included: Step S11: space-time density calculation; Calculate the spatiotemporal density D(e) of each event i ), the formula is: If D(e i ) <T density , then remove the event. Step S12: Adaptive time window adjustment; The time window Δt is adaptively adjusted according to the density of the event stream. The formula is: Among them, α is set to 0.5~1.5 according to the dynamics of the environment. Step S13: secondary filtering of initial noise; If the event stream starts with T init The average density during the time is lower than the threshold T init_density , the events in this time period are discarded.

3. The star point detection algorithm based on event camera according to claim 1, characterized in that: In step S2, the following steps are included: Step S21: time slice interception; Determine the time slice size T according to the star point angular velocity ω slice , the formula is: Among them, θ max is the preset maximum displacement angle; Step S22: mean shift clustering; The event stream within the time slice is clustered using mean shift. The kernel function is a Gaussian kernel, and the kernel bandwidth h is dynamically adjusted: Among them, Var(C k ) is cluster C k The coordinate variance of . Step S23: Weighted centroid calculation Calculate the weighted centroid of each event cluster (x k ,y k ), weight ω i Distribution based on timestamp: Among them, μ t is the mean of the event timestamps within the cluster.

4. The event camera-based star point detection algorithm according to claim 2, characterized in that: In step S11, the parameters for calculating the spatiotemporal density are set as follows: σ s is the standard deviation of the spatial Gaussian kernel, whose value depends on the pixel resolution R of the event camera. pixel Determine: s =γ·R pixel ,γ∈[0.5,2.0];σ t is the temporal Gaussian kernel standard deviation, whose value is based on the dynamic range D of the event camera. event set up: The density threshold T density Dynamically adjusted by the following formula:

5. The star point detection algorithm based on event camera according to claim 2, characterized in that: In step S12, the dynamic setting rule of the adjustment coefficient α in the adaptive time window adjustment is: If the environmental dynamic index V dynamic is defined as the variance of the time intervals between adjacent events in the event stream) is greater than the threshold V th , then α=0.5; If V dynamic ≤V th , then α=1.5-0.

5.

6. The event camera-based star point detection algorithm according to claim 2, characterized in that: In step S13, the threshold T of the secondary filtering of the initial noise is init_density Calculated by the following formula: Where T is the length of the initial time period, and its value range is [10ms, 50ms].

7. The star point detection algorithm based on event camera according to claim 2, characterized in that: In step S21, the real-time estimation method of the star point angular velocity ω is: The Kalman filter is used to predict and update the historical angular velocity data. The state equation is: The observation equation is: Among them, a k is the angular acceleration, ω k and v k are process noise and observation noise respectively, Q and R are covariance matrices.

8. The event camera-based star point detection algorithm according to claim 3, characterized in that: In step S21, the least squares formula for the motion trajectory fitting is: Among them, θ k is the star point angular position of the kth time slice, and the solution formula is:

9. The event camera-based star point detection algorithm according to claim 3, characterized in that: In step S22, the iteration termination condition of the mean shift clustering is: the cluster center displacement Where ∈ = 0.01 pixel, or the maximum number of iterations K max =20.

Images