An infrared spot centroid positioning optimization method based on adaptive multi-window tracking

Abstract

The application discloses an infrared light spot centroid positioning optimization method based on adaptive multi-window tracking and relates to the technical field of visual measurement. The light spot positioning precision and vibration measurement robustness are significantly improved by the adaptive multi-window centroid tracking optimization, mixed motion prediction, information entropy driven window adjustment, and morphological feature constraint occlusion detection, etc. The method uses an infrared LED lamp array as a target, combines an industrial camera and a data processing module, and realizes all-weather and real-time bridge structure monitoring. By high-precision sub-pixel positioning and multi-model weighted fusion prediction, the vibration trajectory of the bridge can be accurately tracked, and reliable attitude data is provided for structure health monitoring.

Description

Technical Field

[0001] This invention relates to the field of visual measurement technology, specifically to an optimized method for positioning the centroid of an infrared spot based on adaptive multi-window tracking. Background Technology

[0002] In the field of visual measurement, high-precision centroid positioning of infrared light spots is the core foundation for realizing three-dimensional coordinate calculation, attitude estimation and dynamic tracking.

[0003] However, real-world measurement environments are complex and variable, posing significant challenges to spot imaging: stray infrared sources or reflective interference exist in the background; the brightness, size, and shape of the spot itself dynamically change due to fluctuations in the light source, changes in target pose, atmospheric or optical factors; high-speed target movement can lead to blurring; and even partial occlusion can occur. Traditional positioning methods, such as fixed-threshold segmentation and single-window gray-scale centroid methods, perform poorly under these non-ideal conditions: global thresholds struggle to adapt to changes in spot brightness and complex background noise, easily leading to segmentation errors; tracking windows of fixed size and position cannot effectively respond to spot scaling and high-speed displacement—windows that are too small will lose the target, while windows that are too large will introduce excessive background interference; simple centroid calculations are extremely sensitive to uneven gray-scale distribution within the spot, blurred edges, or partial occlusion, causing a sharp decline in positioning accuracy and robustness. These shortcomings severely restrict the reliability and accuracy of visual measurement systems in dynamic and complex environments.

[0004] Therefore, there is an urgent need to develop a centroid positioning technology that can adaptively sense changes in light spot characteristics, intelligently suppress background noise, and maintain high accuracy and strong stability under various interferences. Summary of the Invention

[0005] The purpose of this invention is to provide an optimized method for infrared spot centroid localization based on adaptive multi-window tracking. This method aims to significantly improve the core performance of infrared target localization in visual measurement through innovative multi-window dynamic adjustment and localization strategy optimization, thereby solving the technical problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an infrared spot centroid localization optimization method based on adaptive multi-window tracking, comprising at least the following steps:

[0007] S1: Multi-model weighted fusion motion prediction. In response to the multi-modal nonlinear motion characteristics exhibited in bridge structure vibration, a hybrid motion predictor is proposed. The hybrid motion predictor achieves high-precision prediction of light spot trajectory by fusing the prediction results of linear motion model, quadratic motion model and circular motion model, combined with Bayesian dynamic weight adjustment mechanism.

[0008] S2: Occlusion detection with morphological feature constraints. In order to distinguish real target light spots from noise and identify occlusion artifacts, non-light spots are eliminated by constructing ellipse fitting degree, intensity distribution skewness and area attenuation rate through morphological feature constraints. Clustering regions of real light spots in feature space are determined by statistical learning, and judgment conditions are designed for anomaly detection.

[0009] S3: Information entropy-driven adaptive window adjustment. To overcome the influence of dynamic changes in target scale and background interference in bridge structure vibration measurement and to balance computational efficiency and tracking robustness, an information entropy-based window optimization mechanism is adopted. This mechanism quantifies the information richness of the target area and adjusts the tracking window size in real time to achieve the optimal balance between computational efficiency and tracking accuracy.

[0010] S4: A sub-pixel positioning technology based on a Gaussian spot model is proposed. To address the centroid positioning error caused by the dispersion of the spot in long-distance imaging during bridge monitoring, a two-dimensional asymmetric Gaussian light intensity distribution model is established, and sub-pixel-level positioning is achieved through maximum likelihood estimation, providing high-precision input for bridge micro-deformation analysis.

[0011] Furthermore, the linear motion model is applicable to the uniform motion stage and is used to describe small deformations or steady-state vibrations of the structure. The linear motion model is shown in formula (1):

[0012] x k+1 =x k +v k Δt(1)

[0013] In the formula, x k Let v be the two-dimensional position vector of the light spot at time k, i.e., the pixel coordinates; k Let be the velocity vector at time k; Δt is the time step (s). The linear motion model has high computational efficiency, but low sensitivity to acceleration, thus leading to the introduction of the quadratic motion model.

[0014] The reference to the quadratic motion model is to characterize the vibration process with acceleration using the quadratic motion model, as shown in formula (2):

[0015]

[0016] In the formula, a k Let k be the acceleration vector (pixels / frame 2); through historical velocity difference estimation, the secondary motion model can capture the sudden change in vibration energy, but it is easily affected by high-frequency noise. Therefore, a circular motion model is introduced to reduce the influence of high-frequency noise. The circular motion model is shown in formula (3):

[0017]

[0018] In the formula, xc R represents the coordinates of the center of the circular motion; R represents the radius of the circular motion (in pixels); θ k ω is the angle at time k; ω is the angular velocity of the light spot (rad / frame);

[0019] The circular motion model has significant accuracy in predicting periodic motion. In addition, in order to adaptively match the time-varying characteristics of bridge structure vibration, a weight allocation strategy based on evidence reasoning is designed. The prediction results are fused through Bayesian weights, as shown in formula (4):

[0020]

[0021] In the formula, M represents the number of motion models, which is usually 3-5. Let i be the weights of model i at time k, satisfying

[0022] The weights are updated using real-time posterior probabilities, as shown in formula (5):

[0023]

[0024] In the formula, The value range is [0-1]; the likelihood function p(z) k |m i The observed values ​​and actual observed values ​​z of model i are represented by the model i. k The matching degree has a likelihood probability of [0-1];

[0025] Therefore, by using a multi-model weighted fusion prediction formula and known historical data, the coordinates of the light spot in the next frame image can be extracted. When the image is occluded, the predicted coordinates can be output until the image region is restored.

[0026] Furthermore, the ellipse fit (ε ∈ 0-1) measures the similarity between the light spot region and the ideal ellipse:

[0027]

[0028] In the formula, A is the area of ​​the light spot (pixel²); P is the perimeter of the light spot region (pixels);

[0029] Intensity skewness describes the symmetry of the light spot intensity distribution:

[0030]

[0031] In the formula, S is the intensity distribution skewness; μ is the mean intensity [0, 255]; σ is the standard deviation of intensity;

[0032] Area attenuation rate reflects the degree of attenuation of the current spot area relative to the reference area:

[0033]

[0034] Therefore, the occlusion condition can be converted into a morphological determination condition as follows:

[0035]

[0036] In the formula, ε0 is the expected value of the ellipse fit; S0 is the expected value of the skewness; δ ε δ S η is the feature change threshold; η is the spot area attenuation coefficient (0~1); and A0 is the spot reference area (pixel 2).

[0037] Furthermore, the information entropy-based window optimization mechanism includes at least the following steps:

[0038] First, calculate the information entropy of the tracking window W (with a size of w × w pixels):

[0039]

[0040] In the formula, w is the tracking window size (pixels); H(w) is the image information entropy within the window (bits); p k Let k be the probability density of gray level k.

[0041] With the target entropy value H target Based on this, establish a window size feedback control rule to dynamically adjust the tracking window size:

[0042]

[0043] In the formula, H target The target information entropy value (empirical parameter); w ＊ The optimal window size (in pixels).

[0044] Furthermore, S4 includes at least the following steps:

[0045] Considering background noise and optical system aberrations, a probability distribution model for the light spot intensity is established:

[0046]

[0047] In the formula, I(x, y) is the intensity value of the image at coordinates (x, y) [0-255]; S is the peak intensity of the light spot, generally greater than 0; (x0, y0) are the centroid coordinates of the light spot; σ x , σ y denoted as , where is the diffusion parameter of the light spot in the x and y directions; B is the background noise intensity.

[0048] Therefore, we define the likelihood function of the observed data and solve for the optimal solution of the spot coordinates by minimizing the negative logarithm:

[0049]

[0050] In the formula, is the optimal estimate of the spot centroid; N is the number of pixels involved in the calculation; I i I represents the actual observed grayscale value at pixel i; model Let be the gray value predicted by the model at pixel i.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] This invention significantly improves the accuracy of spot positioning and the robustness of vibration measurement through adaptive multi-window centroid tracking optimization, hybrid motion prediction, information entropy-driven window adjustment, and occlusion detection constrained by morphological features. The method utilizes an infrared LED array as a target, combined with an industrial camera and data processing module, to achieve all-weather, real-time bridge structure monitoring. Through high-precision sub-pixel positioning and multi-model weighted fusion prediction, it can accurately track the vibration trajectory of the bridge, providing reliable attitude data for structural health monitoring. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic diagram of the calculation process for an automatic measurement method of bridge structure attitude based on infrared visual tracking;

[0055] Figure 2 This is a schematic diagram of the LED array lamp imaging used in this invention;

[0056] Figure 3 This is a schematic diagram of the complete measurement process of the present invention. Detailed Implementation

[0057] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0058] The positioning error of the LED centroid spot was optimized based on probabilistic models, information entropy, hybrid predictors, and morphological determination. The adaptive multi-window centroid tracking optimization method consists of several parts: centroid localization based on a Gaussian spot model, window size decision based on information entropy optimization, a multi-model adaptive predictor, and occlusion detection constrained by morphological features. A schematic diagram of the specific calculation process is shown below. Figure 1 As shown.

[0059] See Figures 1-3 An optimization method for infrared spot centroid localization based on adaptive multi-window tracking includes at least the following steps:

[0060] S1: Multi-model weighted fusion motion prediction. In response to the multi-modal nonlinear motion characteristics exhibited in bridge structure vibration, a hybrid motion predictor is proposed. The hybrid motion predictor achieves high-precision prediction of light spot trajectory by fusing the prediction results of linear motion model, quadratic motion model and circular motion model, combined with Bayesian dynamic weight adjustment mechanism.

[0061] S2: Occlusion detection with morphological feature constraints. In order to distinguish real target light spots from noise and identify occlusion artifacts, non-light spots are eliminated by constructing ellipse fitting degree, intensity distribution skewness and area attenuation rate through morphological feature constraints. Clustering regions of real light spots in feature space are determined by statistical learning, and judgment conditions are designed for anomaly detection.

[0062] S3: Information entropy-driven adaptive window adjustment. To overcome the influence of dynamic changes in target scale and background interference in bridge structure vibration measurement and to balance computational efficiency and tracking robustness, an information entropy-based window optimization mechanism is adopted. This mechanism quantifies the information richness of the target area and adjusts the tracking window size in real time to achieve the optimal balance between computational efficiency and tracking accuracy.

[0063] S4: A sub-pixel positioning technology based on a Gaussian spot model is proposed. To address the centroid positioning error caused by the dispersion of the spot in long-distance imaging during bridge monitoring, a two-dimensional asymmetric Gaussian light intensity distribution model is established, and sub-pixel-level positioning is achieved through maximum likelihood estimation, providing high-precision input for bridge micro-deformation analysis.

[0064] The linear motion model is suitable for the uniform motion stage and is used to describe small deformations or steady-state vibrations of a structure. The linear motion model is shown in equation (1):

[0065] x k+1 =x k +v k Δt(1)

[0066] In the formula, x k Let v be the two-dimensional position vector of the light spot at time k, i.e., the pixel coordinates; k Let be the velocity vector at time k; Δt is the time step (s). The linear motion model has high computational efficiency, but low sensitivity to acceleration, thus leading to the introduction of the quadratic motion model.

[0067] The use of the quadratic motion model is to characterize the vibration process with acceleration. The quadratic motion model is shown in equation (2):

[0068]

[0069] In the formula, a k Let k be the acceleration vector (pixels / frame 2); through historical velocity difference estimation, the secondary motion model can capture the sudden change in vibration energy, but it is easily affected by high-frequency noise. Therefore, a circular motion model is introduced to reduce the influence of high-frequency noise. The circular motion model is shown in formula (3):

[0070]

[0071] In the formula, x c R represents the coordinates of the center of the circular motion; R represents the radius of the circular motion (in pixels); θ k ω is the angle at time k; ω is the angular velocity of the light spot (rad / frame);

[0072] The circular motion model has significant accuracy in predicting periodic motion. In addition, in order to adaptively match the time-varying characteristics of bridge structure vibration, a weight allocation strategy based on evidence reasoning is designed. The prediction results are fused through Bayesian weights, as shown in formula (4):

[0073]

[0074] In the formula, M represents the number of motion models, which is usually 3-5. Let i be the weights of model i at time k, satisfying

[0075] The weights are updated using real-time posterior probabilities, as shown in formula (5):

[0076]

[0077] In the formula, The value range is [0-1]; the likelihood function p(z) k |m i The observed values ​​and actual observed values ​​z of model i are represented by the model i. k The matching degree has a likelihood probability of [0-1];

[0078] Therefore, by using a multi-model weighted fusion prediction formula and known historical data, the coordinates of the light spot in the next frame image can be extracted. When the image is occluded, the predicted coordinates can be output until the image region is restored. This method significantly improves the robustness of trajectory prediction in bridge structure vibration measurement through multi-physics model collaboration and data-driven weight adjustment, providing a reliable data foundation for subsequent calculation of target attitude information.

[0079] Ellipse fit (ε is between 0 and 1), measures the similarity between the light spot region and the ideal ellipse:

[0080]

[0081] In the formula, A is the area of ​​the light spot (pixel²); P is the perimeter of the light spot region (pixels);

[0082] Intensity skewness describes the symmetry of the light spot intensity distribution:

[0083]

[0084] In the formula, S is the intensity distribution skewness; μ is the mean intensity [0, 255]; σ is the standard deviation of intensity;

[0085] Area attenuation rate reflects the degree of attenuation of the current spot area relative to the reference area:

[0086]

[0087] Therefore, the occlusion condition can be converted into a morphological determination condition as follows:

[0088]

[0089] In the formula, ε0 is the expected value of the ellipse fit, which is generally taken as 0.8 to 0.95 based on experience; S0 is the expected value of the skewness, which is generally taken as -0.5 to 0.5 based on experience; δ ε δ S η is the feature change threshold; η is the spot area attenuation coefficient (0~1); and A0 is the spot reference area (pixel 2).

[0090] The window optimization mechanism based on information entropy includes at least the following steps:

[0091] First, calculate the information entropy of the tracking window W (with a size of w × w pixels):

[0092]

[0093] In the formula, w is the tracking window size (pixels); H(w) is the image information entropy within the window (bits); p k Let k be the probability density of gray level k, with a value range of [0-1].

[0094] With the target entropy value H target Based on this, establish a window size feedback control rule to dynamically adjust the tracking window size:

[0095]

[0096] In the formula, H target The target information entropy value (empirical parameter), typically 4-6 bits; w ＊ The optimal window size (in pixels);

[0097] This method reduces average processing time by approximately 35% compared to the fixed-window algorithm. The mechanism achieves an adaptive balance between "focusing on target features and suppressing background noise" in complex engineering environments through information entropy-window size closed-loop control.

[0098] S4 includes at least the following steps:

[0099] Considering background noise and optical system aberrations, a probability distribution model for the light spot intensity is established:

[0100]

[0101] In the formula, I(x, y) is the intensity value of the image at coordinates (x, y) [0-255]; S is the peak intensity of the light spot, generally greater than 0; (x0, y0) are the centroid coordinates of the light spot; σ x , σ y denoted as , where is the diffusion parameter of the light spot in the x and y directions; B is the background noise intensity.

[0102] Therefore, we define the likelihood function of the observed data and solve for the optimal solution of the spot coordinates by minimizing the negative logarithm:

[0103]

[0104] In the formula, Here, N represents the optimal estimate of the spot centroid, and I represents the number of pixels involved in the calculation. i Let I be the actual observed gray value at pixel i. model Let be the grayscale value predicted by the model at pixel i. This method improves the positioning accuracy from ±0.5 pixels to ±0.1 pixels, greatly enhancing the positioning accuracy of the light spot center coordinates and providing accurate input conditions for solving the pose changes of the bridge structure.

[0105] Based on the above embodiments, the following specific applications are proposed:

[0106] See Figure 2 Hardware deployment: Target array LED infrared lights, the spacing between each array LED bead is known by measurement, wavelength 850nm±20nm, array number can be N×M, emission angle is generally 15°, 30°, 45°, 60°, etc., can achieve all-weather 24-hour uninterrupted emission, imaging quality such as... Figure 2 As shown. The camera is an industrial digital camera with a resolution of 7000(H)×7000(V), featuring a built-in infrared light channel and a pixel size of 3.2 micrometers. The computing unit is a portable 8-core industrial control board. The industrial camera and computing unit are integrated into a self-made camera measurement device. The target is fixed on the surface of the bridge structure being measured, with the light emission direction perpendicular to the self-made measurement device. The self-made measurement device is fixed to a rigid body on the ground, and its six degrees of freedom are constrained.

[0107] Software processing flow: The industrial camera is calibrated to remove the influence of distortion on the measured structure. By setting parameters in the industrial camera control software, such as the acquisition window size, acquisition frame rate, and file storage location, the subsequent acquisition process is ensured to be fully automated.

[0108] The software calculation program is activated to preprocess the acquired images. A weighted fusion prediction step obtains the predicted coordinates of the current frame. Morphological feature constraints, including "ellipse fit," "intensity distribution skewness," and "area attenuation rate," are extracted to determine whether the target in the current frame is occluded. If occluded, the predicted coordinates are directly output; otherwise, the optimal calculation window size for the current state is further determined. After determining the optimal window, the sub-pixel-level centroid coordinates of the spot are solved using an asymmetric Gaussian model. The displacement (Δx, Δy, Δz) and rotation angle (α, β, γ) are output using the RANSAC-PNP method and virtual vision servoing.

[0109] In summary:

[0110] The present invention has the following significant improvement effects:

[0111] By employing an asymmetric Gaussian spot model and maximum likelihood estimation, the positioning error is reduced from ±0.5 pixels in the traditional centroid method to ±0.1 pixels, achieving a displacement resolution of 31.25 micrometers, which can detect millimeter-level micro-vibrations and rotations.

[0112] This invention is based on occlusion detection using three-dimensional morphological features (ellipse fitting degree, intensity skewness, and area attenuation rate), with a false judgment rate of less than 3.5%. It is also adaptable to all weather conditions, resistant to sunlight interference, and maintains good coordinate positioning accuracy under rain, fog, and sudden changes in strong light, enabling 24-hour monitoring.

[0113] By employing a window dynamic adjustment method driven by information entropy, the window processing speed is significantly improved, greatly saving computational time costs. Simultaneously, relying on a multi-model weighted fusion motion prediction method, sufficient measurement accuracy can be maintained even when the target is occluded.

[0114] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An optimization method for infrared spot centroid localization based on adaptive multi-window tracking, characterized in that: At least the following steps are included: S1: Multi-model weighted fusion motion prediction. In response to the multi-modal nonlinear motion characteristics exhibited in bridge structure vibration, a hybrid motion predictor is proposed. The hybrid motion predictor achieves high-precision prediction of light spot trajectory by fusing the prediction results of linear motion model, quadratic motion model and circular motion model, combined with Bayesian dynamic weight adjustment mechanism. S2: Occlusion detection with morphological feature constraints. In order to distinguish real target light spots from noise and identify occlusion artifacts, non-light spots are eliminated by constructing ellipse fitting degree, intensity distribution skewness and area attenuation rate through morphological feature constraints. Clustering regions of real light spots in feature space are determined by statistical learning, and judgment conditions are designed for anomaly detection. S3: Information entropy-driven adaptive window adjustment. To overcome the influence of dynamic changes in target scale and background interference in bridge structure vibration measurement and to balance computational efficiency and tracking robustness, an information entropy-based window optimization mechanism is adopted. This mechanism quantifies the information richness of the target area and adjusts the tracking window size in real time to achieve the optimal balance between computational efficiency and tracking accuracy. S4: A sub-pixel positioning technology based on a Gaussian spot model is proposed. To address the centroid positioning error caused by the dispersion of the spot in long-distance imaging during bridge monitoring, a two-dimensional asymmetric Gaussian light intensity distribution model is established, and sub-pixel-level positioning is achieved through maximum likelihood estimation, providing high-precision input for bridge micro-deformation analysis.

2. The infrared spot centroid localization optimization method based on adaptive multi-window tracking according to claim 1, characterized in that: The linear motion model is applicable to the uniform motion stage and is used to describe small deformations or steady-state vibrations of the structure. The linear motion model is shown in formula (1): x k+1 =x k +v k Δt(1) In the formula, x k Let v be the two-dimensional position vector of the light spot at time k, i.e., the pixel coordinates; k Let be the velocity vector at time k; Δt is the time step (s). The linear motion model has high computational efficiency, but low sensitivity to acceleration, thus leading to the introduction of the quadratic motion model. The reference to the quadratic motion model is to characterize the vibration process with acceleration using the quadratic motion model, as shown in formula (2): In the formula, a k Let k be the acceleration vector at time k. Through historical velocity difference estimation, the secondary motion model can capture the sudden change in vibration energy. However, it is easily affected by high-frequency noise. Therefore, a circular motion model is introduced to reduce the influence of high-frequency noise. The circular motion model is shown in formula (3): In the formula, x c R is the coordinate of the center of the circular motion; R is the radius of the circular motion; θ k Let ω be the angle at time k; ω be the angular velocity of the light spot. The circular motion model has significant accuracy in predicting periodic motion. In addition, in order to adaptively match the time-varying characteristics of bridge structure vibration, a weight allocation strategy based on evidence reasoning is designed. The prediction results are fused through Bayesian weights, as shown in formula (4): In the formula, M represents the number of motion models; Let i be the weights of model i at time k, satisfying The weights are updated using real-time posterior probabilities, as shown in formula (5): In the formula, The value range is [0-1]; the likelihood function p(z) k |m i The observed values ​​and actual observed values ​​z of model i are represented by the model i. k The matching degree has a likelihood probability of [0-1]; Therefore, by using a multi-model weighted fusion prediction formula and known historical data, the coordinates of the light spot in the next frame image can be extracted. When the image is occluded, the predicted coordinates can be output until the image region is restored.

3. The infrared spot centroid localization optimization method based on adaptive multi-window tracking according to claim 2, characterized in that: Ellipse fit, which measures the similarity between the light spot area and the ideal ellipse: In the formula, A is the area of ​​the light spot; P is the perimeter of the light spot region; Intensity skewness describes the symmetry of the light spot intensity distribution: In the formula, S is the intensity distribution skewness; μ is the mean intensity [0, 255]; σ is the standard deviation of intensity; Area attenuation rate reflects the degree of attenuation of the current spot area relative to the reference area: Therefore, the occlusion condition can be converted into a morphological determination condition as follows: In the formula, ε0 is the expected value of the ellipse fit; S0 is the expected value of the skewness; δ ε δ S The threshold for feature changes; η is the light spot area attenuation coefficient, and A0 is the light spot reference area.

4. The infrared spot centroid localization optimization method based on adaptive multi-window tracking according to claim 3, characterized in that: The information entropy-based window optimization mechanism includes at least the following steps: First, calculate the information entropy of the tracking window W: In the formula, w is the tracking window size; H(w) is the image information entropy within the window; p k Let k be the probability density of gray level k. With the target entropy value H target Based on this, establish a window size feedback control rule to dynamically adjust the tracking window size: In the formula, H target The target information entropy value; w * This is the optimal window size.

5. The infrared spot centroid localization optimization method based on adaptive multi-window tracking according to claim 4, characterized in that: The S4 includes at least the following steps: Considering background noise and optical system aberrations, a probability distribution model for the light spot intensity is established: In the formula, I(x, y) is the intensity value of the image at coordinates (x, y) [0-255]; S is the peak intensity of the light spot; (x0, y0) is the centroid coordinate of the light spot; σ x , σ y denoted as , where is the diffusion parameter of the light spot in the x and y directions; B is the background noise intensity. Therefore, we define the likelihood function of the observed data and solve for the optimal solution of the spot coordinates by minimizing the negative logarithm: In the formula, is the optimal estimate of the spot centroid; N is the number of pixels involved in the calculation; I i I represents the actual observed grayscale value at pixel i; model Let be the gray value predicted by the model at pixel i.

Images