Dynamic self-adaptive sun glare interference elimination method based on event star sensor

Abstract

The application discloses a dynamic self-adaptive sun glare interference elimination method based on an event star sensor, and mainly solves the problem that a space detector cannot accurately predict a glare angle and effectively suppress noise interference generated thereby under a sun glare condition. First, a support vector regression (SVR) model is used to establish the relationship between a threshold value and an event rate under different glare angles. Second, a receiver operating characteristic (ROC) curve is used to measure the optimal threshold value under different glare angles. Finally, based on the analysis of the event stream characteristics in a unit time window, the SVR model is used to estimate the glare angle and adaptively adjust the response threshold value. The sun glare interference is effectively eliminated under a dynamic environment. Under a non-direct sun glare environment, the method designed in the application successfully realizes the guarantee of detection accuracy, the effective suppression of irrelevant light noise, the reduction of event processing cost and high practical value.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of computer vision, back light target detection, optimal control, etc., in particular to a space target detection method for dynamically and adaptively eliminating solar back light interference by using an event camera. BACKGROUND

[0002] Star sensor is a space attitude sensor that realizes space navigation based on star detection, and completes the attitude measurement and positioning of the carrier through star pattern matching. However, the navigation performance of the traditional star sensor is significantly limited when it is used in the environment of stray light, especially in the back light scene of the small angle incidence of the sun frequently faced by the space-based load in the on-orbit operation stage. The existing technical means are still insufficient for the problem of interference caused by the stray light of the sun in the back light scene to the star sensor.

[0003] The traditional solution focuses on adding a light shield in hardware and a software post-processing algorithm, but is generally ineffective in dealing with dynamic back light stray light environment due to the limitations of imaging effect, computational complexity or environmental adaptability. The event camera, as a new type of visual sensor, only records information when the brightness of a unit pixel changes by more than a threshold, providing a new idea for solving the above problems. The event camera is not limited by absolute light, and performs time domain brightness difference on background strong light and target signal, has advantages such as large dynamic range, high sensitivity, high sampling rate and low bandwidth, and can essentially improve the efficiency and data quality of signal acquisition in the back light environment.

[0004] In the space environment, due to the continuous change of the relative position of the detector, the space target and the sun, when the sun enters the field of view of the detector in the extreme environment, the focused sun photons will cause a high-frequency triggered leakage noise event dominated by junction leakage current. Even when the sun is not in the field of view, its light will enter the lens and produce irrelevant stray light noise event interference. Therefore, it is very difficult to detect the star target when the detector is running in the back light area disturbed by the strong light of the sun. Under this background, the event camera dynamic adaptive back light interference elimination method is expected to be an effective means to solve this problem. SUMMARY

[0005] The present application provides a dynamic adaptive solar back light interference elimination method based on an event star sensor, which mainly solves the problem that the space detector cannot accurately predict the back light angle and effectively suppress the noise interference generated thereby under the condition of solar back light. Through this method, the event trigger threshold can be adjusted according to real-time environmental data, and the performance of the detector in the face of complex light changes can be optimized.

[0006] The technical solution for achieving the purpose of the present application is as follows: a dynamic adaptive solar back light interference elimination method based on an event star sensor, which comprises the following steps:

[0007] Step (1): Constructing the inverse light angle prediction model based on SVR, defining the inverse light angle and the event trigger threshold, collecting the corresponding event data under different inverse light angles as the training set sample to train the model;

[0008] Step (2): Using the ROC curve to evaluate the detection performance of the system under different thresholds in the training set of step (1), and finding the optimal event trigger threshold corresponding to each inverse light angle;

[0009] Step (3): Defining a fixed time window, cyclically reading the event stream, calculating the event rate and reading the current hardware threshold parameter;

[0010] Step (4): Collecting the event rate data and hardware threshold parameter in step (3) as input, sending it into the pre-trained SVR model to predict the inverse light angle collected by the current device, and dynamically adjusting the appropriate threshold parameter according to the prediction result and the pre-evaluated ROC curve in step (2) and calling back.

[0011] Compared with the prior art, the present application has the following advantages: the present application proposes a dynamic adaptive sun inverse light interference elimination method based on an event star sensor, which has the following advantages: (1) using an event-based sensor, which utilizes the advantages of event cameras, such as high dynamic range and high time resolution, and is not limited by absolute illumination, thereby essentially improving the efficiency and quality of signal acquisition in inverse light environment; (2) using an inverse light angle prediction model based on SVR, which utilizes the excellent generalization ability of SVR to maintain high prediction accuracy even when there is a lot of noise in the data; (3) using a ROC curve to design a threshold optimization strategy, which determines the optimal threshold for each inverse light angle condition, and combines the SVR model to dynamically adaptively adjust the event trigger threshold according to real-time environmental data, thereby optimizing the performance of the detector in the face of complex light changes. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 The flowchart of the dynamic adaptive sun inverse light interference elimination method based on the event star sensor of the present application.

[0013] Figure 2 The threshold-event quantity relationship diagram for the event camera when the inverse light angle is 180°-90°.

[0014] Figure 3 The threshold-event quantity relationship diagram for the event camera when the inverse light angle is 160°-90°.

[0015] Figure 4 The ROC curve diagram for the event camera when the inverse light angle is 160°-90°. DETAILED DESCRIPTION

[0016] The application will be further described below with reference to the accompanying drawings.

[0017] The embodiment of the application provides a dynamic self-adaptive sun glare interference elimination method based on an event star sensor, and a method flow is as shown in Figure 1 The method comprises the following steps:

[0018] Step (1): constructing an SVR-based glare angle prediction model, defining a glare angle and an event trigger threshold, and collecting corresponding event quantity data under different glare angles as training set samples to train the model;

[0019] Step (2): using an ROC curve to evaluate the detection performance of the system under different thresholds in the training set in step (1), and finding out the optimal event trigger threshold corresponding to each glare angle;

[0020] Step (3): defining a fixed time window, cyclically reading in an event stream, calculating an event rate and reading a current hardware threshold parameter;

[0021] Step (4): taking the event rate data and the hardware threshold parameter collected in step (3) as input, inputting the pre-trained SVR model, predicting the glare angle collected by the current device, and dynamically adjusting a suitable threshold parameter according to the prediction result and the ROC curve pre-evaluated in step (2) and calling back.

[0022] Further, step (1) specifically comprises the following steps:

[0023] Step (11): setting up a glare experiment, defining a glare angle as an included angle between a light source and an imaging direction of an event camera, and setting the light source and a star point target position as fixed. Fixing the event camera on a high-precision three-axis turntable, the event camera collects all event data in a process of moving from a glare angle of 90° to a glare angle of 180°, repeatedly performing the experiment multiple times as shown in Figure 2 and Figure 3 The event data collected in the above experiment are taken as training set samples.

[0024] Step (12): in the training set samples, defining a glare angle and a threshold, and corresponding event quantity data under different glare angles. Constructing a feature set x and a target variable y; wherein x is a two-dimensional array, which comprises all combinations of the glare angle and the threshold; y is a one-dimensional array, which comprises event quantities corresponding to each combination in x;

[0025] Step (13): for the feature vector of the sample, wherein x i ={x i1 ,…x ik} is an influence factor of y i , and k is the number of influence factors. Then, the support vector regression function is:

[0026] f(x) = w T g(x) + bg(x)

[0027] where g(x) is a nonlinear function mapping the samples to a high dimensional space, w T is a coefficient of the independent variable function, and b is a bias;

[0028] Step (14): using the generalized optimal classification surface function considering the sub-sampling and the maximum classification interval, a dual optimization problem is constructed by using the Lagrange equation and the dual problem theory to obtain the values of w T and b in step (12), and the SVR function model is obtained as:

[0029]

[0030] where K(x i , x j ) is a kernel function, a i , a j and are Lagrange multipliers.

[0031] Further, step (2) specifically comprises the following steps:

[0032] Step (21): the ROC curve is represented in a coordinate graph, where the horizontal coordinate is the false positive rate (FPR) and the vertical coordinate is the true positive rate (TPR); the true positive rate is defined as the proportion of all actual positive class samples that are correctly identified as positive class, and the calculation formula is:

[0033]

[0034] where TN is the number of true negatives;

[0035] The false positive rate is defined as the proportion of all actual negative class samples that are incorrectly identified as positive class, and the calculation formula is:

[0036]

[0037] where FP is the number of false positives;

[0038] Step (22): for each inverse light angle, different TPR and FPR values are collected by changing the threshold value, and an ROC image is drawn as shown in Figure 4 ; each group (FPR, TPR) corresponds to a point in the ROC curve graph, and connecting these points forms the ROC curve for each inverse light angle; the optimal threshold value is the threshold value corresponding to the point closest to the upper left corner of the graph in the ROC curve, specifically, the threshold value that makes The smallest threshold value ensures that the true positive rate is maintained while the false positive rate is minimized.

[0039] Further, the step (3) comprises the following specific steps:

[0040] Step (31): using the event camera to take pictures for target detection under the condition of sun backlight;

[0041] Step (32): setting an initial threshold value and defining a fixed time window, cyclically reading in the event stream, decoding the address event stream and storing in the quadruple e i =(x i ,y i ,t i ,p i ) where (x i ,y i ) is the event pixel coordinate, t i is the event timestamp, and p i is the event polarity;

[0042] Step (33): for the event stream of the fixed time window, calculating the event amount in unit time, i.e. the event rate, and reading the current hardware threshold parameter.

[0043] Further, the step (4) takes the event rate data collected in the step (3) and the hardware threshold parameter as input, sends them into the pre-trained SVR model, predicts the backlight angle collected by the current device, and according to the prediction result, dynamically adjusts the appropriate threshold parameter and calls back according to the ROC curve pre-evaluated in the step (22).

[0044] The above is the preferred embodiment of the present application, and the above examples are only used to help understand the method of the present application and its core idea, but the embodiment of the present application is not limited by the above content. For those skilled in the art, according to the idea of the present application, the specific embodiment and application range will be changed, any change, modification, substitution, combination, simplification made without departing from the spirit and principle of the present application should be equivalent replacement method, and all are included in the protection scope of the present application.

Claims

1. A dynamic adaptive solar backlight interference cancellation method based on an event star sensor, characterized in that, Includes the following steps: Step (1): Construct a backlight angle prediction model based on SVR, define the backlight angle and event trigger threshold, and collect the event quantity data corresponding to different backlight angles as training set samples to train the model; Step (2): Use ROC curves to evaluate the target detection performance of the training set in step (1) under different thresholds at multiple angles, and find the corresponding optimal event trigger threshold for each backlight angle; Step (3): Define a fixed time window, read the event stream in a loop, calculate the event rate, and read the current hardware threshold parameters; Step (4): The event rate data and hardware threshold parameters collected in step (3) are used as inputs and fed into the pre-trained SVR model to predict the backlight angle collected by the current device. According to the prediction result, the appropriate threshold parameters are dynamically adjusted and the callback is performed based on the ROC curve pre-evaluated in step (2). Step (1) specifically includes the following steps: Step (11): Set up a backlight experiment, define the backlight angle as the angle between the light source and the imaging direction of the event camera, and set the positions of the light source and the star point target to be fixed; fix the event camera on a high-precision three-axis turntable, and collect all event data during the process of the event camera moving from a backlight angle of 90° to a backlight angle of 180°, and repeat the experiment multiple times; use the event data collected in the above experiment as training set samples; Step (12): Construct an SVR backlight angle prediction model and train the model based on the training set samples; For the training set data at different thresholds under each backlight angle in step (2), under a unique backlight angle, the true positive rate (TPR) and false positive rate (FPR) are calculated for the data at each threshold. The TPR and FPR are calculated for all backlight angles at different thresholds, and ROC curves are plotted for different angles. For each backlight angle, the optimal threshold corresponds to the point closest to the upper left corner of the ROC curve, i.e., minimizing the threshold. The point at which the light is emitted, and the threshold corresponding to that point is the optimal threshold for that backlight angle. Step (3) specifically includes the following steps: Step (31): Use an event camera to perform target detection and shooting under backlight conditions; Step (32): Set an initial threshold and define a fixed time window, read the event stream in a loop, decode the address event stream, and store it in the quadruple e. i =(x i ,y i ,t i ,p i In ), where (x i ,y i ) represents the pixel coordinates of the event, t i p is the event timestamp. i The polarity of the event; Step (33): For an event stream with a fixed time window, calculate the number of events per unit time, i.e. the event rate, and read the current hardware threshold parameters.

Images