Adaptive periodic signal noise reduction method and system for event camera, and electronic equipment

Abstract

The invention discloses a self-adaptive periodic signal noise reduction method and system for an event camera and electronic equipment, and the method comprises the steps: S1, a spatial domain self-adaptive filtering step: carrying out spatial domain filtering on an original event stream output by the event camera, and generating a spatially filtered event stream; s2, a time domain density cutting step: performing event density analysis on the event stream after spatial filtering on a time axis, identifying and retaining a time interval higher than a set event density, and generating an event stream after time cutting; and S3, a pixel-level period phase filtering step: performing pixel-level period estimation and phase consistency filtering on the event stream after time cutting to generate a de-noised event stream. According to the method, the problems that the signal-to-noise ratio is too low due to high-density background noise interference when an event camera detects periodic flicker signals and an existing fixed parameter filtering method is poor in adaptability are solved, and the high-fidelity extraction capability of weak periodic signals in a complex noise environment is improved.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and sensor signal processing technology, specifically relating to a signal noise reduction method and system for an event camera in a periodic flickering scene, and an electronic device, particularly a high-fidelity noise reduction technology that integrates spatiotemporal characteristics and phase consistency and introduces a multi-layer adaptive mechanism. Background Technology

[0002] Event cameras are a novel type of visual sensor inspired by the biological retina, capable of asynchronously outputting pixel-level brightness change events with microsecond-level temporal resolution and a dynamic range exceeding 120 dB. This characteristic gives them unparalleled advantages over traditional cameras in scenarios such as high-speed motion capture, high dynamic range imaging, and periodic light signal detection. In recent years, event cameras have received widespread attention in applications requiring the detection of high-frequency periodic light signals, such as LED modulation communication, high-speed industrial stroboscopic measurement, neuroscience research, and AR / VR eye tracking.

[0003] However, event cameras have an inherent flaw in their operating principle: even in the absence of changes in lighting, the sensor spontaneously generates a large amount of background activity noise. When detecting weak, periodic flickering signals, the high-density background noise can easily overwhelm the real signal event, causing a sharp drop in the signal-to-noise ratio. This not only affects the accuracy of subsequent signal decoding but also incurs huge data storage and transmission overhead. This has become a core bottleneck restricting the practical deployment of event cameras in this type of application.

[0004] Existing technologies have proposed various event denoising methods, most of which are based on filtering the differences in spatiotemporal correlation between real events and background noise. Examples include event density filtering based on fixed thresholds, background activity filtering based on neighborhood statistics, and time-space-content correlation filtering incorporating polarity information. However, these methods face significant challenges when dealing with extreme scenarios where high-density noise and sparse periodic signals coexist: they generally do not explicitly utilize the structural characteristics of the periodic signal itself, and their filtering parameters are often fixed values ​​or require manual setting. When signal strength, noise levels, or flicker frequency change, filters with fixed parameters struggle to adapt adaptively, leading to a significant decrease in denoising performance—either leaving excessive noise or unnecessarily removing too much effective signal.

[0005] Therefore, there is an urgent need in this field for an event stream processing method that can adapt to environmental changes and still achieve high-fidelity noise reduction under extremely low signal-to-noise ratio conditions. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention proposes an adaptive periodic signal denoising method, system, and electronic device for event cameras. This addresses the issues of low signal-to-noise ratio due to high-density background noise interference and poor adaptability of existing fixed-parameter filtering methods when event cameras detect periodic flicker signals, thereby improving the high-fidelity extraction capability of weak periodic signals in complex noise environments.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An adaptive periodic signal noise reduction method for event cameras includes the following steps: S1, Spatial Domain Adaptive Filtering Step: Spatial domain filtering is performed on the original event stream output by the event camera to generate a spatially filtered event stream. The discrimination threshold of the spatial domain filtering is dynamically calculated based on the global statistical features of the filtering index values ​​of all pixels in the original event stream. S2, Temporal density segmentation step: Perform event density analysis on the time axis on the spatially filtered event stream, identify and retain time intervals with high event density, and generate a time-segmented event stream; S3, Pixel-level periodic phase filtering step: Perform pixel-level period estimation and phase consistency filtering on the time-divided event stream to generate a denoised event stream; wherein, the period estimation process includes a reliability-based decision mechanism, when the period estimation based on the first polarity event sequence is unreliable, the period estimation based on the second polarity event sequence or the period estimation of neighboring pixels is introduced to supplement or replace it; and, the discrimination threshold of the phase consistency filtering is dynamically adjusted based on the distribution characteristics of the phase histogram.

[0008] Preferably, in step S1, the process of performing spatial domain filtering on the raw event stream output by the event camera includes: Calculate the position of each pixel ( x , y ) filter index value S ( x , y The calculation formula is as follows: S ( x , y ) = N ( x , y ) / ( b + |Σ p i |) in, N ( x , y () represents the total number of events at that location. p i For the firsti The polarity of an event, b It is an adjustable positive parameter; Calculate all pixels S ( x , y The global statistical characteristics of the value; Based on the aforementioned global statistical features, the adaptive threshold for spatial domain filtering is dynamically calculated. i adaptive ; Each pixel S ( x , y ) value and the adaptive threshold i adaptive Compare and retain S ( x , y ) ≥ i adaptive All events at the specified pixel position are filtered out, and events at other positions are excluded.

[0009] Preferably, in step S1, the values ​​of all pixels are calculated. S ( x , y The global statistical features of the value are as follows: calculate the global statistical features of all pixels. S ( x , y mean of the values m s with standard deviation s s .

[0010] Preferably, in step S1, the adaptive threshold for spatial domain filtering is dynamically calculated based on the global statistical features. i adaptive ,include: When the global statistical feature is the mean m s with standard deviation s s At that time, according to the formula i adaptive = m s + k · s s Calculate the threshold, where, k These are predefined coefficients.

[0011] Preferably, in step S2, the process of performing event density analysis on the time axis of the spatially filtered event stream includes: The time axis of the spatially filtered event stream [ Tmin , T max Divide into K equally spaced sliding windows; Calculate the event density within each sliding window r ( t ); Calculate the median of the event density for all sliding windows. r median ; According to the formula t = β · r median Calculate the density cutting threshold t ,in, β This is the density scaling factor; Find events with a density continuously higher than a threshold t Maximum time interval [ T 0 , T 1 ]; The timestamp is stored in [ T 0 , T 1 Events within the specified interval are filtered out, while events outside the specified interval are excluded.

[0012] Preferably, in step S3, the pixel-level period estimation and phase consistency filtering process includes: S3.1, Time window polarity filtering: Analyze the single-pixel event sequence in the event stream after time segmentation, and suppress random polarity noise through polarity statistics within a short sliding window; S3.2, Period Estimation: For the event sequence after polarity filtering, detect the transition time intervals based on brightness enhancement events and brightness reduction events respectively, and execute the aforementioned confidence-based decision mechanism to obtain the final estimated period of the pixel. T est ; S3.3, Phase Calculation and Phase Density Filtering: Based on the final estimated period... T est The event is mapped to the phase space, the phase histogram is calculated, and the event is filtered based on a dynamically adjusted phase density threshold to retain events with consistent phase.

[0013] Preferably, in step S3.2, the process of executing the credibility-based decision-making mechanism includes: Calculate the estimated period based on the brightness enhancement event sequence and its coefficient of variation CV ON ; Calculate the estimated period based on the brightness attenuation event sequence and its coefficient of variation CV OFF ; Set the coefficient of variation threshold i CV ; like CV ON and CV OFF All less than i CV Then fusion and As the final estimated period for this pixel T est ; like CV ON and CV OFF Only one of them is smaller than i CV If so, then this reliable estimation period will be used as the final estimation period. T est ; like CV ON and CV OFF All are greater than or equal to i CV Then, from the neighboring pixels whose period estimates are reliable, take the mode or median of their estimated periods as the final estimated period for that pixel. T est .

[0014] Preferably, in step S3.3, the filtering process based on the dynamically adjusted phase density threshold includes: According to the final estimated period T est Using a phase reference point, the timestamp of each event is mapped to a phase space of [0, 360°) to obtain its phase value. f i ; Divide the phase space into K Each interval is counted, and the event phase histogram is plotted. H ( k ); Calculate the peak significance index of the phase histogram P The calculation formula is as follows: P = (max( H ) - mean( H )) / mean( H ); According to the peak significance indexP The phase density threshold coefficient is dynamically adjusted through a predefined mapping relationship. c Generate adaptive threshold coefficients c adaptive The mapping relationship is as follows: P The larger the value, the better. c adaptive The smaller the value; According to the formula e = c adaptive · max( H Calculate the phase density threshold e For each event, if its phase interval k Number of events H ( k )≥ ɛ If it is positive, it will be retained; otherwise, it will be discarded as noise.

[0015] The present invention also discloses an adaptive periodic signal noise reduction system for an event camera, used to perform the above method, comprising the following modules: Spatial Domain Adaptive Filtering Module: This module performs spatial domain filtering on the raw event stream output by the event camera to generate a spatially filtered event stream. The filtering threshold is dynamically calculated based on global statistical features. Temporal density segmentation module: used to perform temporal density analysis on the spatially filtered event stream, identify and retain time intervals with high event density, cut off low-density time intervals, and generate a time-segmented event stream; Pixel-level periodic phase filtering module: used to perform pixel-level period estimation and phase consistency filtering on the time-divided event stream to generate a noise-reduced event stream. It has a reliability decision mechanism for period estimation and a dynamic adjustment mechanism for phase threshold.

[0016] The present invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the processor and running on the processor, wherein the processor executes the program to implement the steps of the above-described method or the above-described system.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) In this invention, the spatial domain adaptive filtering step achieves automatic and accurate screening of signal pixels by dynamically calculating the threshold based on global statistical features, overcoming the limitation that a fixed threshold cannot adapt to different noise levels, and significantly improving the robustness of the initial filtering.

[0018] 2) In this invention, the pixel-level periodic phase filtering step introduces a reliability-based periodic decision mechanism and dynamic phase threshold adjustment, which can effectively deal with residual noise interference and still robustly estimate the signal period and achieve high-precision filtering in extreme scenarios, thus solving the problem that traditional methods are prone to failure when the signal is sparse.

[0019] 3) This invention forms a complete adaptive noise reduction solution through a three-level collaborative progressive filtering architecture of space, time and phase. It achieves nearly 100% noise removal and more than 99.8% signal retention under extremely low signal-to-noise ratio conditions, which greatly improves the output signal-to-noise ratio and provides reliable data protection for applications such as LED communication and high-speed industrial testing. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process of an adaptive periodic signal noise reduction method for an event camera, according to a preferred embodiment of the present invention.

[0021] Figure 2 This is a detailed flowchart illustrating the spatial domain adaptive filtering steps in a preferred embodiment of the present invention.

[0022] Figure 3 This is a flowchart illustrating the time-domain density cutting step in a preferred embodiment of the present invention.

[0023] Figure 4 This is a detailed flowchart of the pixel-level periodic phase filtering step in a preferred embodiment of the present invention.

[0024] Figure 5 The bar chart shows a quantitative performance comparison between the present invention and several existing mainstream comparison methods when the original event signal-to-noise ratio is -51.22 dB. Figure 5 Subgraphs (a), (b), and (c) respectively present a visual comparison of the noise reduction metrics (noise removal rate NRR, signal retention rate SRR, and output signal-to-noise ratio SNR) of the STPF of this invention with those of classic event filters (BAF, EDF, STCF, STCC, EBF, PFD-A).

[0025] Figure 6 This is a block diagram of an adaptive periodic signal noise reduction system for an event camera, according to a preferred embodiment of the present invention. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0027] This embodiment provides an adaptive periodic signal noise reduction method for event cameras, referring to... Figure 1 As shown, the overall process includes the following three main steps: S1: Spatial Domain Adaptive Filtering: Spatial domain filtering is applied to the original event stream output by the event camera to generate a spatially filtered event stream. The adaptive threshold used for discrimination in spatial domain filtering is dynamically calculated based on the global statistical features of the filtering index values ​​of all pixels in the original event stream. S2: Temporal density segmentation: Perform event density analysis on the time axis on the spatially filtered event stream, identify and retain time intervals with higher than the set event density, and generate a time-segmented event stream; S3: Pixel-level periodic phase filtering: Pixel-level period estimation and phase consistency filtering are performed on the time-divided event stream to generate a denoised event stream; wherein, the period estimation process includes a reliability-based decision mechanism, when the period estimation based on the first polarity event sequence is unreliable, the period estimation based on the second polarity event sequence or the period estimation of neighboring pixels is introduced to supplement or replace it; the discrimination threshold of the phase consistency filtering is dynamically adjusted based on the distribution characteristics of the phase histogram.

[0028] The following provides a detailed explanation of each step.

[0029] Reference Figure 2 In step S1 of this embodiment, the spatial domain adaptive filtering process is as follows: S1.1, calculate the position of each pixel ( x , y ) filter index value S ( x , y The calculation formula is as follows: S ( x , y ) = N ( x , y ) / ( b + |Σ p i |), where, N ( x , y () represents the total number of events at that location. p i For the first i The polarity of an event, b It is an adjustable positive parameter.

[0030] S1.2, Calculate all pixels S ( x , y mean of the values m s and standard deviation s sThe calculation process is as follows: Suppose the image has a total of M valid pixels (i.e., N(x,y)>0 pixels), then the mean... m s The calculation formula is: Standard deviation s s The calculation formula is .

[0031] S1.3, according to formula i adaptive = m s + k · s s Dynamically calculate the adaptive threshold; in this embodiment, the coefficient... k Take 1.5.

[0032] S1.4, convert each pixel S ( x , y ) value and i adaptive Compare, retain S ( x , y ) ≥ i adaptive All events at the pixel position are filtered out, and the remaining events are output as a spatially filtered event stream.

[0033] Reference Figure 3 In step S2 of this embodiment, the process of time-domain density cutting is as follows: S2.1, the time axis of the spatially filtered event stream [ T min , T max Divided into multiple sliding windows, window length Δ T = α · ( T max - T min In this embodiment, the window scaling factor... α Take 0.1, and the sliding step size is Δ T / 5.

[0034] S2.2, Calculate the event density within each sliding window ∆T. r ( t The calculation process is as follows: For the j-th sliding window, its event density... r j The calculation formula is: r j = Nj / ∆ T , in, N j This represents the total number of events within the sliding window.

[0035] S2.3, Calculate the median of all sliding window event densities. r median The calculation process is as follows: First, the density values ​​of all sliding windows... r j ( j =1,2,..., K Arrange in ascending order to obtain the sequence. r 1≤ r 2≤ ... ≤ ρ K . K When it is an odd number, r median = r (K+1) / 2 When K is even, r median = ( r K / 2 + r K / 2 +1 ) / 2.

[0036] S2.4, Set the density cutting threshold t = β · r median In this embodiment, the density scaling factor β Take 0.5.

[0037] S2.5, find the maximum time interval where the event density is continuously higher than the threshold τ. T 0 , T 1 ].

[0038] S2.6, the timestamp is stored in [ T 0 , T 1 For events within a given interval, filter out events outside that interval and output the event stream after time segmentation.

[0039] Reference Figure 4 In step S3 of this embodiment, the pixel-level periodic phase filtering process includes the following sub-steps: S3.1, Time Window Polarity Filtering: Perform short time window analysis on the single-pixel event stream. If a certain polarity event accounts for more than 80%, then retain the dominant polarity event; otherwise, treat all events in the window as noise and remove them.

[0040] S3.2, Period Estimation and Credibility Decision: S3.2.1, detect the time intervals of OFF→ON and ON→OFF transition events respectively. For the ON event period estimation, assume that the event is detected... L A series of consecutive ON events, with a timestamp sequence of: Then the periodic estimate The calculation formula is: Its coefficient of variation CV ON The calculation formula is: ,in and Let be the mean and standard deviation of the time interval for ON events, respectively. Similarly, for the estimation of the OFF event cycle, suppose M consecutive OFF events are detected, and their timestamp sequence is . ,but: Its coefficient of variation CV OFF The calculation formula is: ,in and These represent the mean and standard deviation of the time interval between ON events, respectively.

[0041] S3.2.2, Set the threshold for the coefficient of variation i CV =0.01.

[0042] S3.2.3, if CV ON and CV OFF All less than i CV Then the periods of both are combined as the final estimated period. T est .

[0043] S3.2.4, if CV ON and CV OFF Only one of them is smaller than i CV Then the reliable estimation period is used as T est .

[0044] S3.2.5, if CV ON and CV OFF All are greater than or equal to i CV Then, from the neighboring pixels whose period estimates are reliable, take the mode or median of their estimated periods as the final estimated period for that pixel. Test .

[0045] S3.3, Phase Calculation and Dynamic Phase Density Filtering: S3.3.1, using the most recent OFF→ON transition time as a reference point. t ref According to the formula f i = [( t i - t ref )mod T est / T est ] × 360° Maps events to phase space.

[0046] S3.3.2, divide the phase space into K Intervals (interval width ∆) f =0.1°), statistical phase histogram H ( k ).

[0047] S3.3.3, Calculate the peak significance index P = (max( H ) - mean( H )) / mean( H ).

[0048] S3.3.4, according to P Value dynamically adjusts the proportional coefficient c adaptive (For example, P When >1, c adaptive =0.5; P When ≤1, c adaptive =0.7).

[0049] S3.3.5, Calculate the phase density threshold e = c adaptive · max( H ).

[0050] S3.3.6, For each event, if its phase interval... k Number of events H ( k ) ≥ e If the noise is not present, it will be retained; otherwise, it will be removed as noise, and the final output will be a noise-reduced event stream.

[0051] The effectiveness and technical advantages of this invention will be verified using quantitative data below.

[0052] like Figure 5 As shown in the figure, a bar chart quantitatively compares the performance of the method of this invention with several mainstream comparative methods when the original event signal-to-noise ratio is -51.22 dB. This chart quantitatively demonstrates the superiority and robustness of the method of this invention in high-density noise scenarios through noise removal rate (NRR), signal retention rate (SRR), and output signal-to-noise ratio (SNR). Figure 5 Quantitative comparisons show that, at a signal-to-noise ratio of -51.22 dB, the present invention significantly outperforms traditional methods in terms of noise removal rate (NRR), signal retention rate (SRR), and output signal-to-noise ratio (SNR).

[0053] like Figure 6 As shown, this embodiment discloses an adaptive periodic signal noise reduction system for event cameras, used to perform the above method, including the following modules: Spatial Domain Adaptive Filtering Module: Used to perform spatial domain filtering on the raw event stream output by the event camera, generating a spatially filtered event stream; Temporal density segmentation module: used to perform temporal density analysis on the spatially filtered event stream, identify and retain time intervals with high event density, and generate a time-segmented event stream; Pixel-level periodic phase filtering module: used to perform pixel-level period estimation and phase consistency filtering on the time-divided event stream to generate a noise-reduced event stream.

[0054] Other aspects of this embodiment can be found in the above method embodiments.

[0055] The present invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the processor and running on the processor, wherein the processor executes the program to implement the steps of the above method or the system described above.

[0056] In summary, this invention discloses an adaptive periodic signal denoising method, system, and electronic device for event cameras. Addressing the technical problem of extremely low signal-to-noise ratio (SNR) in periodic flickering scenes, where event streams are susceptible to high-density background noise interference, this invention proposes a spatial-temporal-phase collaborative filtering framework. This invention requires no external triggering or prior frequency information, and achieves progressive noise removal through adaptive spatial filtering based on global statistical features, temporal axis density slicing, and pixel-level periodic phase filtering that integrates confidence decisions and dynamic threshold adjustments. This invention overcomes the performance degradation of traditional methods when signals are sparse and noise dynamically changes. Even under extremely low SNR conditions, it achieves nearly 100% noise removal and over 99.8% signal retention, significantly improving the output SNR and providing a high-fidelity data foundation for high-speed vision and precision optical signal detection applications.

[0057] It will be apparent to those skilled in the art that the above summary of the present invention does not imply that every exemplary embodiment or every implementation of the invention is described, and that various modifications and substitutions can be made to the invention without departing from its spirit and scope. Therefore, it is intended that the invention cover any modifications, substitutions, and equivalents of the invention that fall within the scope of the appended claims and their equivalents.

Claims

1. An adaptive periodic signal noise reduction method for event cameras, characterized in that, Includes the following steps: S1, Spatial Domain Adaptive Filtering Step: Spatial domain filtering is performed on the original event stream output by the event camera to generate a spatially filtered event stream; wherein, the adaptive threshold used for discrimination by spatial domain filtering is dynamically calculated based on the global statistical features of the filtering index values ​​of all pixels in the original event stream; S2, Time Domain Density Segmentation Step: Perform event density analysis on the time axis on the spatially filtered event stream, identify and retain time intervals with higher than the set event density, and generate a time-segmented event stream; S3, Pixel-level periodic phase filtering step: Perform pixel-level period estimation and phase consistency filtering on the time-divided event stream to generate a denoised event stream; wherein, the period estimation process includes a reliability-based decision mechanism, when the period estimation based on the first polarity event sequence is unreliable, the period estimation based on the second polarity event sequence or the period estimation of neighboring pixels is introduced to supplement or replace it; the discrimination threshold of the phase consistency filtering is dynamically adjusted based on the distribution characteristics of the phase histogram.

2. The adaptive periodic signal noise reduction method for event cameras according to claim 1, characterized in that, Step S1 is as follows: Calculate the position of each pixel ( x , y ) filter index value S ( x , y The calculation formula is: in, N ( x , y () represents the total number of events at that location. For the first i The polarity of an event, b It is an adjustable positive parameter; Calculate all pixels S ( x , y Global statistical characteristics of the value; Based on the aforementioned global statistical characteristics, the adaptive threshold for spatial domain filtering is dynamically calculated using linear or nonlinear functions. ; Each pixel S ( x,y ) value and the adaptive threshold θ adaptive Compare and retain S ( x,y ) ≥ θ adaptive All events at the specified pixel position are filtered out, and events at other positions are excluded.

3. The adaptive periodic signal noise reduction method for event cameras according to claim 2, characterized in that, In step S1, calculate the values ​​of all pixels. S ( x , y The global statistical features of the value are as follows: calculate the global statistical features of all pixels. S ( x , y mean of the values μ s with standard deviation σ s .

4. The adaptive periodic signal noise reduction method for event cameras according to claim 3, characterized in that, In step S1, the adaptive threshold for spatial domain filtering is dynamically calculated based on the global statistical features. θ adaptive ,include: When the global statistical feature is the mean μ s with standard deviation σ s At that time, according to the formula θ adaptive = μ s + k · σ s Calculate the threshold, where, k These are predefined coefficients.

5. The adaptive periodic signal noise reduction method for event cameras according to claim 1, characterized in that, Step S2 includes: The time axis of the spatially filtered event stream [ T min , T max Divided into K An equidistant sliding window; Calculate the event density within each sliding window ρ ( t ); Calculate the median of the event density for all sliding windows. ρ median ; According to the formula τ = β · ρ median Calculate the density cutting threshold τ ,in, β This is the density scaling factor; Find events with a density continuously higher than a threshold τ Maximum time interval [ T 0 , T 1 ]; The timestamp is stored in [ T 0 , T 1 Events within the specified interval are filtered out, while events outside the specified interval are filtered out.

6. The adaptive periodic signal noise reduction method for event cameras according to claim 1, characterized in that, Step S3 includes: S3.1, Time window polarity filtering: Analyze the single-pixel event sequence in the event stream after time segmentation, and suppress random polarity noise through polarity statistics within a short sliding window; S3.2, Period Estimation: For the event sequence after polarity filtering, detect the transition time intervals based on brightness enhancement events and brightness reduction events respectively, and execute the aforementioned confidence-based decision mechanism to obtain the final estimated period of the pixel. T est ; S3.3, Phase Calculation and Phase Density Filtering: Based on the final estimated period... T est The event is mapped to the phase space, the phase histogram is calculated, and the event is filtered based on a dynamically adjusted phase density threshold to retain events with consistent phase.

7. The adaptive periodic signal noise reduction method for an event camera according to claim 6, characterized in that, In step S3.2, the process of executing the credibility-based decision-making mechanism includes: Calculate the estimated period based on the brightness enhancement event sequence and its coefficient of variation CV ON ; Calculate the estimated period based on the brightness attenuation event sequence and its coefficient of variation CV OFF ; Set the coefficient of variation threshold θ CV ; like CV ON and CV OFF All less than θ CV Then fusion and As the final estimated period for this pixel T est ; like CV ON and CV OFF Only one of them is smaller than θ CV If so, then this reliable estimation period will be used as the final estimation period. T est ; like CV ON and CV OFF All are greater than or equal to θ CV Then, from the neighboring pixels whose period estimates are reliable, take the mode or median of their estimated periods as the final estimated period for that pixel. T est .

8. The adaptive periodic signal noise reduction method for an event camera according to claim 6, characterized in that, In step S3.3, the filtering process based on the dynamically adjusted phase density threshold includes: According to the final estimated period T est Using a phase reference point, the timestamp of each event is mapped to a phase space of [0, 360°) to obtain its phase value. φ i ; Divide the phase space into K Each interval is counted, and the event phase histogram is plotted. H ( k ); Calculate the peak significance index of the phase histogram P The calculation formula is as follows: P = (max( H ) - mean( H )) / mean( H ); According to the peak significance index P The phase density threshold coefficient is dynamically adjusted through a predefined mapping relationship. γ Generate adaptive threshold coefficients γ adaptive The mapping relationship is as follows: P The larger the value, the better. γ adaptive The smaller the value; According to the formula ε = γ adaptive · max( H Calculate the phase density threshold ε For each event, if its phase interval k Number of events H ( k )≥ ɛ If it is positive, it will be retained; otherwise, it will be discarded as noise.

9. An adaptive periodic signal noise reduction system for an event camera, for performing the method as described in any one of claims 1-8, characterized in that, Includes the following modules: Spatial Domain Adaptive Filtering Module: Used to perform spatial domain filtering on the raw event stream output by the event camera, generating a spatially filtered event stream; Temporal density segmentation module: used to perform temporal density analysis on the spatially filtered event stream, identify and retain time intervals with high event density, and generate a time-segmented event stream; Pixel-level periodic phase filtering module: used to perform pixel-level period estimation and phase consistency filtering on the time-divided event stream to generate a noise-reduced event stream.

10. An electronic device, comprising a memory, a processor, and a computer program stored in the processor and running thereon, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8 or the system as described in claim 9.

Images