Lens focusing method and device and zoom lens

Abstract

The invention relates to the technical field of lens focusing, and discloses a lens focusing method and device and a zoom lens. The method comprises the following steps: acquiring a focal length value and an aperture value of a zoom lens, and calculating a bending radius compensation amount of each radial region in an imaging surface; determining spatial position coordinates of each reflective interference area based on image data acquired by the zoom lens; generating a focusing state evaluation result based on the bending radius compensation amount and the spatial position coordinates; and generating a focusing control instruction according to the focusing state evaluation result and driving a focusing motor to move to a target position. According to the invention, the bending radius compensation amount of each radial area of the zoom lens under different focal lengths can be accurately calculated, the problem of out-of-focus of an edge area, which cannot be processed by a traditional focusing method, is effectively solved, and the full-view-field focusing consistency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lens focusing, and in particular to a lens focusing method, a lens focusing device and a zoom lens. Background Art

[0002] Traditional focusing technology primarily relies on single-point or multi-point distance measurement based on image clarity evaluation functions. This method performs well with fixed-focus lenses, but when used with zoom lenses, due to their complex optical structure and multiple lens groups, they produce varying degrees of field curvature at different focal lengths, resulting in the imaging surface exhibiting complex three-dimensional curved features. This makes it impossible for traditional focusing algorithms, which rely on flat surfaces, to simultaneously guarantee sharpness in both the center and edges of the image. Multiple reflections between lens groups create strong reflective interference on the imaging surface. These reflective areas typically have high contrast and sharpness values, making them easily misidentified by traditional focusing algorithms as true focus information. This can lead to misjudgments and oscillations in the focusing system, causing frequent out-of-focus or slow focusing issues in the autofocus system. Summary of the Invention

[0003] The main purpose of the present invention is to provide a lens focusing method, device and zoom lens. The present invention can accurately calculate the bending radius compensation amount of each radial area of ​​the zoom lens at different focal lengths, effectively solving the problem of defocusing in the edge area that traditional focusing methods cannot handle, and improving the focus consistency of the entire field of view.

[0004] To achieve the above object, the present invention provides a lens focusing method, comprising the following steps: Obtaining the focal length and aperture value of the zoom lens, and calculating the bending radius compensation amount of each radial area in the imaging plane; Determining the spatial position coordinates of each reflective interference area based on the image data collected by the zoom lens; generating a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates; A focus control instruction is generated according to the focus state evaluation result, and a focus motor is driven to move to a target position.

[0005] Optionally, in a first implementation of the first aspect of the present invention, obtaining the focal length and aperture value of the zoom lens and calculating the bending radius compensation amount of each radial area in the imaging plane includes: Collecting zoom ring angle data of a zoom lens and synchronously collecting aperture ring gear position data of the zoom lens; querying a preset focal length calibration mapping table according to the zoom ring angle data to obtain a focal length value, and querying a preset aperture calibration mapping table according to the aperture ring gear data to obtain an aperture value; The bending radius compensation amount of each radial area in the imaging plane is calculated according to the focal length value and the aperture value.

[0006] Optionally, in a second implementation of the first aspect of the present invention, calculating the bending radius compensation amount of each radial area in the imaging plane according to the focal length value and the aperture value includes: querying a preset field curvature model parameter table to obtain a current field curvature model parameter corresponding to the focal length value; Performing a field curvature correction calculation based on the focal length value and the current field curvature model parameters to obtain a curvature radius value of the imaging surface; Dividing the imaging surface into a plurality of radial regions according to a preset concentric ring segmentation rule, and determining a geometric position deviation amount of each radial region according to the bending radius value and the aperture value; A radial weight function is applied to the geometric position deviation to perform weighted calculation to obtain a bending radius compensation amount for each radial area in the imaging plane.

[0007] Optionally, in a third implementation of the first aspect of the present invention, determining the spatial position coordinates of each reflection interference area based on the image data collected by the zoom lens includes: Using a polarization analyzer to perform dual-channel separation processing on the image data collected by the zoom lens, respectively extracting a P polarization channel image and an S polarization channel image; Calculating a pixel-by-pixel intensity ratio of the P polarization channel image to the S polarization channel image to obtain a polarization intensity ratio image; Double threshold detection and local variance screening are performed on the polarization intensity ratio image to identify the spatial position coordinates of each reflection interference area.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, performing dual threshold detection and local variance screening on the polarization intensity ratio image to identify the spatial position coordinates of each reflective interference area includes: Comparing each pixel of the polarization intensity ratio image with a first threshold and a second threshold, respectively, marking a pixel as a candidate reflection pixel when the pixel value is greater than both the first threshold and the second threshold, and determining a first pixel distribution map based on the candidate reflection pixels; Based on the first pixel distribution map, a rectangular sliding window of a preset size is used to calculate the local mean and local variance pixel by pixel, and when the local variance exceeds a preset variance threshold, the pixel point is retained, otherwise the pixel point is eliminated to obtain a second pixel distribution map; First, performing an opening operation of a circular structuring element with a specified radius on the second pixel distribution map to remove isolated pixel points, and then performing a closing operation of the circular structuring element to fill the holes inside the pixel area, to obtain a reflective area distribution map; Performing a connected domain labeling analysis on the reflective area distribution map to obtain a plurality of reflective interference areas; The centroid coordinates of all pixel coordinates in each reflection interference area are calculated, and the centroid coordinates are used as the spatial position coordinates of each reflection interference area.

[0009] Optionally, in a fifth implementation manner of the first aspect of the present invention, generating a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates includes: Determining the reflection influence range according to the spatial position coordinates of each reflection interference area, and setting a suppression weight coefficient for the pixel points within the reflection influence range according to the reflection intensity classification; Calculating the sharpness value of each pixel in the image data, and performing weighted suppression on the sharpness value in combination with the suppression weight coefficient to obtain sharpness evaluation data; According to the sharpness evaluation data, the sharpness value of each radial area is geometrically corrected using the corresponding bending radius compensation amount to obtain target sharpness data; The target sharpness data is weighted and summed to obtain a focus state evaluation result.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, calculating the sharpness value of each pixel in the image data, and performing weighted suppression on the sharpness value in combination with the suppression weight coefficient to obtain sharpness evaluation data includes: Performing noise preprocessing on the image data using a Gaussian filter to obtain a filtered image, and then performing edge detection on the filtered image using a Laplace second-order differential operator to obtain a gradient intensity value of each pixel point; Taking the absolute value of the gradient intensity value as the sharpness value of each pixel point, and constructing a first sharpness matrix according to the sharpness value of each pixel point; According to the spatial position coordinates and the reflection influence range of each reflection interference area, the first sharpness matrix is ​​multiplied by the corresponding suppression weight coefficient for numerical attenuation to obtain a second sharpness matrix; A weighted average is performed on all valid pixel points in the second sharpness matrix to obtain a weighted average value, and the weighted average value is used as sharpness evaluation data.

[0011] Optionally, in a seventh implementation of the first aspect of the present invention, generating a focus control instruction according to the focus state evaluation result and driving a focus motor to move to a target position includes: Calculating a proportional control amount and a differential control amount of the focus state evaluation result, and generating a first position adjustment amount according to the proportional control amount and the differential control amount; Calculating the sharpness gradient direction of the second sharpness matrix, determining the search step size and the moving direction according to the sharpness gradient direction, and dynamically adjusting the step speed according to the second sharpness matrix; adjusting the first position adjustment amount in real time according to the search step length, the moving direction, and the step speed to obtain a second position adjustment amount; Inputting the second position adjustment amount as an observation value into a Kalman filter, performing a recursive filtering operation through a state prediction equation and an observation update equation to obtain a focus position adjustment amount; The target moving distance and rotation direction of the focus motor are calculated according to the focus position adjustment amount, the target moving distance is converted into a driving pulse sequence and combined with the rotation direction to form a focus control instruction, and the focus motor is driven to move to the target position.

[0012] The present invention also provides a lens focusing device, comprising: An acquisition module is used to obtain the focal length and aperture value of the zoom lens and calculate the bending radius compensation amount of each radial area in the imaging surface; a reflection interference analysis module, configured to determine the spatial position coordinates of each reflection interference area based on the image data collected by the zoom lens; a focus state evaluation module, configured to generate a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates; A focus control module is used to generate a focus control instruction according to the focus state evaluation result and drive the focus motor to move to a target position.

[0013] The present invention also provides a zoom lens for implementing the steps of any of the above methods.

[0014] In summary, the technical solution provided by the present invention can accurately calculate the bending radius compensation amount of each radial area of ​​the zoom lens at different focal lengths through field curvature correction, effectively solves the problem of defocusing of edge areas that traditional focusing methods cannot handle, and significantly improves the focus consistency of the entire field of view. Polarized light intensity analysis technology is adopted, and the reflective interference area generated by multiple reflections inside the lens group is accurately identified through a dual-channel differential detection algorithm, and the negative impact of reflections on focus evaluation is effectively suppressed through an intelligent weight distribution mechanism, which greatly improves the focus stability in complex lighting environments. Combined with the joint processing mechanism of field curvature compensation and reflection suppression, the present invention realizes true adaptive focus control, can dynamically adjust the focus strategy according to actual optical conditions, can achieve fast response while ensuring focus accuracy, and at the same time introduces a Kalman filter to smooth the focus control instructions, effectively eliminating the influence of system noise and random fluctuations on focus accuracy, and improving the robustness and long-term stability of the entire lens focus. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 1 is a schematic diagram of the steps of a lens focusing method according to an embodiment of the present invention; Figure 2 It is a structural block diagram of a lens focusing device in one embodiment of the present invention.

[0016] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] Reference Figure 1 , this embodiment provides a lens focusing method, comprising the following steps: S1, obtaining the focal length and aperture value of the zoom lens, and calculating the bending radius compensation amount of each radial area in the imaging plane; In this embodiment, an angle-parameter mapping system for zoom lenses is established. This system maps the zoom ring angle to the focal length value, and the aperture ring position information to the actual aperture value. This creates queryable focal length and aperture calibration mapping tables. An angle sensor collects the zoom ring rotation angle of the zoom lens in real time, simultaneously capturing the aperture ring's current mechanical position data. The zoom ring angle is input into the focal length calibration mapping table, and the corresponding focal length value is calculated through interpolation. The aperture ring position data is input into the aperture calibration mapping table to obtain the current actual aperture value. Using an imaging curvature radius model based on the Petzval field curvature correction theory, combined with the focal length and aperture value, the field curvature compensation requirements for each radial position on the image sensor surface under current conditions are determined. The imaging plane is divided into multiple concentric ring regions. For each region, radial position parameters are calculated based on its distance from the optical axis center. The control parameters in the imaging curvature radius model are used to generate the ideal imaging plane offset at the current position, which is the curvature radius compensation for the current region.

[0019] S2, determining the spatial position coordinates of each reflection interference area based on the image data collected by the zoom lens; In this embodiment, a polarization analyzer with high extinction ratio characteristics is used to perform polarization separation processing on the raw image data captured by the zoom lens. The input image is split into two complementary channels based on the polarization direction of the light: a P-polarization channel image and an S-polarization channel image. The P-polarization channel primarily retains effective imaging information from the direction of direct light, while the S-polarization channel is more sensitive to interference components caused by specular reflections, thereby effectively distinguishing between reflected and direct light. A pixel-by-pixel intensity ratio calculation is performed on the corresponding pixels of the P-polarization channel image and the S-polarization channel image to produce a polarization intensity ratio image. The value of each pixel in the image represents the relative intensity index of its reflective characteristics, effectively revealing the boundaries and morphology of potential reflective interference areas in the image. A dual-threshold judgment mechanism is used to improve the accuracy of reflective detection. A main threshold is set to identify highly reflective areas where the polarization ratio exceeds the background level. Local variance is also introduced as an auxiliary judgment criterion. The variance level of the ratio map within the local area is calculated using a sliding window to eliminate false reflective points caused by noise or brightness fluctuations. When a pixel's polarization intensity ratio exceeds the main threshold and its neighborhood variance exceeds the set threshold, it is determined to be a valid pixel in the true reflection area. Through the above process, the spatial coordinates of all reflection interference areas in the image are located and output as a binary mask.

[0020] S3, generating a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates; In this embodiment, the spatial coordinates of each reflective interference area identified by the polarization analysis module, combined with the reflective intensity level information output by the reflective detection network, are used to determine the actual impact range of each reflective area in the image. Pixels within the impact range are then divided into light, moderate, and heavy reflective areas based on a predefined grading strategy. Pixels corresponding to each level are then assigned different suppression weights, with light areas corresponding to higher suppression coefficients and heavy areas corresponding to lower suppression coefficients approaching zero. This effectively eliminates the interference of reflections on sharpness calculations. A sharpness calculation is performed on all pixels in the original image data, using high-frequency response methods such as the Laplace operator or Sobel gradient to extract image edge information as the basic pixel-level sharpness metric. The suppression weights are then applied to each pixel's sharpness value for weighted adjustment, selectively suppressing the sharpness contribution of reflective interference areas in the overall evaluation. Based on the curvature radius compensation corresponding to each radial region on the image sensor's imaging surface, the weighted sharpness evaluation data is geometrically corrected. The sharpness values ​​corresponding to pixels that deviate from the ideal imaging surface are mapped and adjusted according to the compensation model. This corrects the focus offset error caused by field curvature and restores the sharpness expression of each region on the theoretical focal plane. The position-corrected target sharpness data for each region is normalized and weighted summed based on its spatial position on the sensor and the radial weight coefficient, outputting the focus status evaluation result at a unified scale.

[0021] S4, generating a focus control instruction according to the focus state evaluation result and driving the focus motor to move to the target position.

[0022] In this embodiment, based on the focus state evaluation result generated for the current frame image, the change in the focus state evaluation result relative to the previous frame evaluation value is calculated. This is used to construct a proportional control variable and a differential control variable. The proportional control variable reflects the static deviation between the current focal plane and the optimal focus, while the differential control variable captures the dynamic response of the focus change trend. The two are linearly combined to form a first position adjustment variable. A second sharpness matrix of the current image is extracted and gradient analysis is performed on the second sharpness matrix to determine the sharpness gradient direction in the most sensitive areas of the image, indicating the focus path that will achieve the fastest improvement in image clarity. The search step size and movement direction are dynamically estimated based on the gradient direction, and the step speed is adaptively adjusted in real time based on the changing trend of the current sharpness matrix. This allows the first position adjustment variable to be automatically refined and adjusted in areas of the image with drastic sharpness changes, generating a second position adjustment variable with high spatial awareness. The second position adjustment variable is input as an observation into the Kalman filter state recursion model. A state estimation process, constructed through a combination of prediction and update steps, performs denoising and filtering on the second position adjustment variable, outputting a current, reliable focus position adjustment variable. The target movement distance and motor rotation direction are calculated based on the focus position adjustment amount, and the target movement distance is discretized into a drive pulse sequence. At the same time, the direction information is encoded into high and low level control signals, which together with the pulse sequence form a focus control instruction, which is transmitted to the focus motor drive unit to drive the focus motor to perform a rotation action, so that the lens moves along the optimal path to the target focus position.

[0023] In one example, obtaining a focal length and an aperture value of a zoom lens and calculating a bending radius compensation amount for each radial region in an imaging plane include: Collect the zoom ring angle data of the zoom lens and the aperture ring gear data of the zoom lens simultaneously; Querying a preset focal length calibration mapping table based on the zoom ring angle data to obtain a focal length value, and querying a preset aperture calibration mapping table based on the aperture ring gear data to obtain an aperture value; The bending radius compensation amount of each radial area in the imaging surface is calculated according to the focal length value and the aperture value.

[0024] In this example, an angle sensor and encoder system are integrated into the lens structure. The rotary encoder on the zoom ring outputs real-time angular change data during zoom operation, with a minimum resolution of 0.1 degrees. A mechanical gear position sensing unit is also located on the aperture ring, and combined with a Hall effect sensor or photoelectric switch, it enables instant identification of the aperture position. Each physical gear position change is accurately mapped to the corresponding aperture value. During the initialization phase, two calibration mapping tables are preset to describe the nonlinear relationship between the zoom ring angle and focal length value, and the mapping relationship between the aperture ring position and the actual incident light flux. The focal length calibration table is obtained by measuring the physical focal length values ​​at different angles and fitting them using a piecewise spline interpolation function. The aperture calibration table uses discrete notations based on the relationship between the opening and closing of the aperture blades and the corresponding aperture value. During actual operation, the current zoom angle value is read in real time from the zoom ring angle sensor. Using the zoom angle value as the independent variable, an interpolation search is performed in the focal length mapping table to obtain the physical focal length of the current lens. Simultaneously, the mechanical gear number of the current aperture ring is obtained from the aperture sensor module and the corresponding aperture value is matched in the aperture calibration table to complete the real-time calculation of the imaging system's imaging parameters. The imaging curvature radius model, constructed based on Petzval field curvature theory, is used. This model uses the current focal length and image height as input variables and lens characteristic parameters as control variables. Using the fitted model function, it calculates the theoretical imaging plane offset for each radial region under the current conditions. This model fully accounts for the nonlinear effects of focal length variation on field curvature and the weighted effect of image height on edge defocus, accurately describing the actual curvature of the imaging plane. The entire image sensor surface is divided into multiple concentric circular areas. The image height distance of the center point of each area is calculated respectively. The image height distance is substituted into the imaging bending radius model to obtain the imaging bending radius corresponding to each radial area. The focal plane offset is then calculated based on the geometric relationship between the theoretical focus of the imaging surface and the actual sensor plane. The focal plane offset is the compensation amount that needs to be applied to each radial area.

[0025] In one example, calculating the bending radius compensation amount of each radial area in the imaging plane according to the focal length value and the aperture value includes: Querying a preset field curvature model parameter table to obtain current field curvature model parameters corresponding to the focal length value; The field curvature correction calculation is performed by combining the focal length value and the current field curvature model parameters to obtain the curvature radius value of the imaging surface; The imaging surface is divided into multiple radial regions according to the preset concentric ring segmentation rule, and the geometric position deviation of each radial region is determined according to the bending radius value and the aperture value; The radial weight function is applied to the geometric position deviation to perform weighted calculation to obtain the bending radius compensation amount of each radial area in the imaging surface.

[0026] In this example, the algorithm module pre-sets a set of field curvature model parameter tables corresponding to different focal lengths. These tables are obtained during the system calibration phase by fitting measured data at a large number of focal lengths and image heights. Each parameter set covers the core optical characteristics that influence the curvature of the imaging surface, including constant terms, focal length-dependent terms, and image height adjustment terms. The table is indexed by focal length and contains parameter sets, supporting high-precision table lookup. After obtaining the current physical focal length, an interpolation query is performed on the field curvature model parameter table to extract the field curvature model parameter set at the corresponding focal length. This field curvature model parameter set is then input into the field curvature correction calculation module, where the imaging surface geometry is deduced. This yields the curvature radius of the imaging system at the corresponding focal length. This describes the spatial deviation of the ideal imaging surface from the planar sensor, and its variation reflects the lens' sensitivity to field curvature during zoom. The entire imaging surface is divided into several radial regions centered on the sensor optical axis according to a pre-set concentric ring partitioning rule. These regions are divided into equidistant or equal-area regions, each with a well-defined center coordinate and a fixed image height. After obtaining the bending radius, the geometric offset effect of the bending radius on different regions is derived in combination with the current aperture value. That is, the displacement of the imaging position of each radial region relative to the theoretical focal plane, namely the geometric position deviation, is calculated to represent the degree of defocus caused by field curvature in the current radial region. Taking into account the varying importance of different regions to the overall imaging quality, the geometric position deviation is weighted using a radial weight function. The weight function is an exponential attenuation function based on the inverse square of the image height, giving a higher weight to the center of the image and moderately reducing the contribution value of the edge regions. The bending radius compensation value for each radial region is output by multiplying the geometric position deviation by the weight coefficient region by region and normalizing the sum.

[0027] Among them, before combining the focal length value and the current field curvature model parameters to perform field curvature correction calculation, it also includes the step of performing environmental adaptability correction on the current field curvature model parameters, specifically including: collecting the temperature data, humidity data and air pressure data of the current shooting environment through the environmental sensor, and recording the working time and usage frequency of the zoom lens to obtain an environmental parameter data set; performing matrix operation on the environmental parameter data set and the preset environmental influence coefficient matrix to calculate the influence of temperature change on the refractive index of the lens, the influence of humidity change on the surface state of the lens and the influence of air pressure change on the distance between the lens groups, and obtaining the environmental correction factor; according to the environmental correction factor The sub-process numerically corrects the curvature coefficient, aberration coefficient and distortion coefficient in the current field curvature model parameters, and combines the lens aging compensation algorithm to compensate for the optical performance drift caused by long-term use, so as to obtain the field curvature model parameters after environmental correction; establishes a nonlinear mapping relationship between the environmental parameters and the field curvature parameter offset, and uses the cubic spline interpolation algorithm to predict and dynamically update the field curvature parameters under different environmental conditions in real time to obtain the adaptive field curvature model parameters; replaces the current field curvature model parameters with the adaptive field curvature model parameters, and updates the parameter records of the corresponding focal length values ​​in the preset field curvature model parameter table to obtain the field curvature calculation benchmark after dynamic correction.

[0028] In one example, determining the spatial coordinates of each reflection interference area based on image data collected by a zoom lens includes: A polarization analyzer is used to perform dual-channel separation processing on the image data collected by the zoom lens, and the P polarization channel image and the S polarization channel image are extracted respectively. Calculate the pixel-by-pixel intensity ratio of the P polarization channel image and the S polarization channel image to obtain a polarization intensity ratio image; Double threshold detection and local variance screening are performed on the polarization intensity ratio image to identify the spatial coordinates of each reflection interference area.

[0029] In this example, a polarization analyzer with motorized rotation capability is placed in the imaging optical path. The polarization analyzer integrates a P-polarization filter and an S-polarization filter, each used to extract component signals corresponding to different polarization directions of the incident light. During the actual acquisition process, the control module sequentially rotates the polarization filter to two complementary orientations, capturing an image frame in each orientation. The first frame is image data obtained through the P-polarization filter, and the second frame is image data obtained through the S-polarization filter. Because the P-polarization image primarily retains direct light and some diffuse reflections from the object surface, while the S-polarization image is more sensitive to specular reflections and multiple reflections, the difference between the two is used to distinguish between normal imaging areas and areas of potential reflective interference. The P- and S-polarization channel images are paired pixel by pixel and registered in image coordinate space. A pixel-level intensity ratio calculation is then performed. The P-polarization image pixel value at each coordinate position is divided by the S-polarization image pixel value at the corresponding position. This generates a two-dimensional grayscale image, the polarization intensity ratio image. Each pixel's grayscale value represents the difference in reflectance between the two polarization conditions at the corresponding spatial point. Stronger reflectance results in larger ratios, forming local highlights. A dual-threshold detection is performed on the polarization intensity ratio image. A primary threshold is set to identify pixels with polarization ratios above the background level, while an auxiliary variance threshold is set to exclude spurious high-ratio regions caused by uneven illumination or image noise. A sliding window method is then applied to the polarization intensity ratio image. The local variance is calculated within a fixed window around each pixel. If the polarization ratio of a pixel exceeds the primary threshold and the local variance of its neighborhood exceeds the set variance threshold, the pixel is considered a valid reflector. The coordinate information of all the reflected points is subjected to connected domain analysis to segment multiple reflective interference areas with actual physical significance. The spatial position center, boundary shape and coverage range of each area are calculated to form a reflective interference spatial distribution map.

[0030] Among them, before performing double threshold detection and local variance screening on the polarization intensity ratio image, it also includes the step of using a pre-trained reflection recognition neural network to perform intelligent pattern recognition processing on the polarization intensity ratio image, specifically including: performing sliding window segmentation on the polarization intensity ratio image according to a preset overlap rate, extracting multiple local image blocks of fixed size as neural network input samples, and obtaining a reflection candidate area sample set; inputting the reflection candidate area sample set into a pre-trained five-layer convolutional neural network for feature extraction, extracting local texture features, edge features and high-level semantic features in turn through the first convolutional layer, the second convolutional layer and the third convolutional layer to obtain multi-scale feature map data; inputting the multi-scale feature map data into the fully connected layer for feature fusion and dimensionality reduction processing, and The nonlinear transformation of the connection layer and the second fully connected layer calculates the reflection probability value, reflection center coordinates and reflection influence radius to obtain the reflection area feature vector; the reflection intensity level is classified based on the reflection area feature vector, and the detected reflection area is graded and marked as light reflection, moderate reflection and heavy reflection, and a corresponding suppression weight coefficient is assigned to each level to obtain graded reflection area data; a reflection confidence distribution map is generated according to the graded reflection area data, and the reflection confidence distribution map is weightedly fused with the polarization intensity ratio image to obtain an intelligently optimized polarization intensity ratio image; the intelligently optimized polarization intensity ratio image is used as input data for subsequent dual threshold detection and local variance screening processing, and the threshold parameters of the dual threshold detection are dynamically adjusted according to the graded reflection area data.

[0031] In one example, dual threshold detection and local variance screening are performed on the polarization intensity ratio image to identify the spatial coordinates of each reflective interference area, including: Comparing each pixel of the polarization intensity ratio image with a first threshold and a second threshold respectively, marking a pixel as a candidate reflection pixel when the pixel value of the pixel is greater than both the first threshold and the second threshold, and determining a first pixel distribution map based on the candidate reflection pixels; Based on the first pixel distribution map, a rectangular sliding window of a preset size is used to calculate the local mean and local variance pixel by pixel, and when the local variance exceeds a preset variance threshold, the pixel point is retained, otherwise the pixel point is eliminated to obtain a second pixel distribution map; First, an opening operation of a circular structuring element with a specified radius is performed on the second pixel distribution map to remove isolated pixels, and then a closing operation of the circular structuring element is performed to fill the holes inside the pixel area to obtain a reflective area distribution map; Conducting connected domain labeling analysis on the reflective area distribution map to obtain multiple reflective interference areas; The centroid coordinates of all pixel coordinates in each reflection interference area are calculated, and the centroid coordinates are used as the spatial position coordinates of each reflection interference area.

[0032] In this example, each pixel of the polarization intensity ratio image is compared with the first threshold and the second threshold respectively. When the pixel value of the pixel is greater than the first threshold and the second threshold at the same time, it is marked as a candidate reflective pixel. All pixels that meet the conditions are combined to form a binary mask image, that is, the first pixel distribution map. A rectangular sliding window of a fixed size is applied pixel by pixel on the first pixel distribution map, and the local mean and local variance of the included area are calculated in each window. The variance measures the degree of grayscale change in the area. When the local variance exceeds the preset variance threshold, it means that there is a polarization ratio difference in the area, and there is a possibility of real reflective interference. Therefore, the pixel is retained. Otherwise, it is regarded as a misjudgment caused by noise or image unevenness and is eliminated, and the second pixel distribution map after statistical screening is obtained. Based on the second pixel distribution map, two morphological processing operations are performed sequentially. An opening operation is performed on the image using a circular structuring element of a specified radius to remove isolated pixels and edge protrusions, making the region outlines clearer and more regular. The resulting image is then closed using the same structuring element to fill small holes within the region caused by missed low-value detections, ensuring a strong sense of enclosure and integrity in the reflective regions. A connected domain labeling analysis is then performed on the reflective region distribution map. Connected domain labeling is performed on the reflective region distribution map, dividing connected pixel sets into independent reflective interference regions using an eight-neighborhood or four-neighborhood approach. Each region is assigned a unique identifier, forming a multi-region structure list. Statistical processing is performed on the spatial coordinates of all pixels within each independent reflective interference region. The centroid coordinates of the region are calculated by taking the arithmetic mean of the horizontal and vertical coordinates of all pixels, describing the central location of the reflective interference. This provides strong spatial representation and visualization advantages, and the centroid coordinates are output as the spatial coordinates of each reflective interference region.

[0033] In one example, generating a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates includes: Determine the reflection influence range based on the spatial position coordinates of each reflection interference area, and set a suppression weight coefficient for the pixels within the reflection influence range according to the reflection intensity classification; Calculate the sharpness value of each pixel in the image data, and perform weighted suppression on the sharpness value in combination with the suppression weight coefficient to obtain the sharpness evaluation data; According to the sharpness evaluation data, the sharpness value of each radial area is geometrically corrected using the corresponding bending radius compensation amount to obtain the target sharpness data; The target sharpness data is weighted and summed to obtain the focus status evaluation result.

[0034] In this example, the spatial centroid coordinates of each reflective interference region are used as a criterion. In combination with parameters such as the pixel distribution density, the mean and maximum local polarization intensity ratios, and other parameters within each region, a reflection intensity grading system is constructed. Reflective regions are classified into three levels: mild, moderate, and severe. Based on the corresponding influence radius model, a radius-adjustable influence region boundary is constructed for each reflective region, centered at the centroid. Each pixel in the image space is traversed, and its reflection level is determined based on whether the pixel falls within the influence range of a reflective region and its distance from the reflection center. A corresponding suppression weight coefficient is assigned. Mildly reflective regions are given a higher weight to retain some sharpness signal, moderately reflective regions are given a lower weight to reduce the interference, and severe regions are set to approximately zero to completely eliminate their sharpness contribution. Sharpness values ​​are calculated for all pixels in the entire image, and the sharpness characteristics of each pixel region are extracted using methods such as the Laplace operator response, local gradient amplitude, or frequency domain energy distribution. Each pixel's sharpness value is weighted point-by-point using a suppression weighting factor. Each pixel's sharpness value is multiplied by the corresponding suppression factor to generate sharpness data that accounts for reflections. The image is divided into several concentric radial regions. Based on the lens's current focal length and aperture, the corresponding bending radius compensation value for each radial region is extracted from a field curvature model lookup table, describing the geometric offset between the theoretical focal plane and the sensor imaging plane. This bending radius compensation value is used to spatially adjust the sharpness evaluation data. The sharpness values ​​of pixels in the offset region are geometrically corrected or interpolated according to the compensation distance, simulating the theoretical focus position. All corrected target sharpness data are then uniformly weighted and summed. Weights are set based on the region's image height distance, image center importance, or user-defined region priority. This summation process aggregates the sharpness status of the entire image, generating a comprehensive numerical result that serves as the focus status assessment for the current frame.

[0035] In one example, the sharpness value of each pixel in the image data is calculated, and the sharpness value is weightedly suppressed in combination with the suppression weight coefficient to obtain sharpness evaluation data, including: Use Gaussian filter to preprocess the image data for noise to obtain the filtered image, and then apply Laplace second-order differential operator to perform edge detection on the filtered image to obtain the gradient intensity value of each pixel; The absolute value of the gradient intensity value is used as the sharpness value of each pixel point, and a first sharpness matrix is ​​constructed according to the sharpness value of each pixel point; According to the spatial position coordinates and the reflection influence range of each reflection interference area, the first sharpness matrix is ​​multiplied by the corresponding suppression weight coefficient for numerical attenuation to obtain a second sharpness matrix; A weighted average is performed on all valid pixel points in the second sharpness matrix to obtain a weighted average value, and the weighted average value is used as sharpness evaluation data.

[0036] In this example, Gaussian filtering is performed on the input image data to achieve noise suppression. By constructing a two-dimensional Gaussian kernel with a standard deviation parameter, a convolution operation is performed on the original image, effectively smoothing high-frequency noise while retaining the main structural features. This gives the image greater stability and processing consistency without destroying edge information. The Laplace second-order differential operator is applied to the image preprocessed by Gaussian filtering. The Laplace second-order differential operator enhances the edge response of areas with significant grayscale changes in the image by performing second-order derivative calculations on the image. It can capture sharp edges, texture boundaries, and tiny details in the image with high sensitivity, and outputs a gradient image containing positive and negative values, whose values ​​reflect the severity of the grayscale curvature of the pixel in the image. The absolute value of each pixel value in the gradient image is taken to obtain a non-negative sharpness amplitude map, and the first sharpness matrix is ​​constructed. Each pixel value in the matrix represents its local sharpness intensity in the image space. The higher the value, the greater the contribution of the pixel to the edge structure. Referring to the spatial position coordinates and influence radius information of the reflective interference area identified by polarization analysis, a corresponding suppression weight coefficient is assigned to each pixel within the reflection influence range according to the corresponding level of reflection suppression strategy. For example, the heavy area corresponds to a weight close to zero, the moderate area uses an intermediate weight, and the non-reflective area weight is set to one. The weight coefficient is multiplied pixel by pixel with the sharpness value of the corresponding position in the first sharpness matrix to achieve effective attenuation of the local sharpness expression caused by the reflective interference, forming a second sharpness matrix with a suppression effect. All valid pixels are screened out in the second sharpness matrix, and a weighted average calculation is performed. The sharpness values ​​of all pixels are numerically weighted and accumulated according to their spatial position or preset area weight, and normalized by the number of valid pixels to obtain a scalar value representing the clarity of the entire image as sharpness evaluation data.

[0037] In one example, generating a focus control instruction according to a focus state evaluation result and driving a focus motor to move to a target position includes: Calculating a proportional control amount and a differential control amount of a focus state evaluation result, and generating a first position adjustment amount according to the proportional control amount and the differential control amount; Calculating the sharpness gradient direction of the second sharpness matrix, determining the search step size and the moving direction according to the sharpness gradient direction, and dynamically adjusting the step speed according to the second sharpness matrix; The first position adjustment amount is adjusted in real time according to the search step length, the moving direction and the step speed to obtain a second position adjustment amount; The second position adjustment amount is input into the Kalman filter as an observation value, and a recursive filtering operation is performed through the state prediction equation and the observation update equation to obtain the focus position adjustment amount; The target moving distance and rotation direction of the focus motor are calculated based on the focus position adjustment amount, the target moving distance is converted into a driving pulse sequence and combined with the rotation direction to form a focus control instruction, which drives the focus motor to move to the target position.

[0038] In this example, a proportional control variable is constructed based on the difference between the focus state evaluation result and the previous frame evaluation result, reflecting the static deviation between the current focus and the ideal imaging state. Simultaneously, the first-order time derivative is calculated based on the rate of change of the evaluation values ​​of two adjacent frames on the time axis to obtain a differential control variable, quantifying the direction and speed of dynamic changes in the current focus trend. The proportional and differential terms are weightedly superimposed to form a first position adjustment variable. Image gradient direction analysis is performed in the second sharpness matrix to identify the local area with the most sensitive image sharpness response. By vector summing and normalizing the pixel gradient directions within the local area, the main direction of the sharpness gradient of the entire image is determined, and this direction serves as the guiding direction for adjustment of the motor actuator. At the same time, based on the sharpness change intensity and gradient amplitude in the area where the main direction is located, combined with the sharpness increase and decrease trend from the center to the edge of the image, the search step size and movement direction are dynamically estimated, and an adaptive adjustment strategy is introduced to control the step speed in real time. When the image sharpness improves rapidly, the step speed is accelerated to improve the adjustment efficiency. When the sharpness change slows down or approaches saturation, the step speed is actively reduced to improve the focus accuracy, thereby avoiding overshoot or oscillation problems. The estimated search step size, movement direction and the current first position adjustment are combined and calculated to superimpose a real-time compensation value based on the image content response to form a second position adjustment value. The second position adjustment value is input into the Kalman filter as the observation value. Through the joint recursive calculation of the state prediction equation and the observation update equation, the historical state, estimation error covariance, process noise and observation noise are jointly applied to the current state prediction process to output a focus position adjustment value after filtering and correction, which effectively suppresses the adverse effects of occasional image disturbances or single-frame outliers on focus positioning. The target movement distance and rotation direction required by the actual motor are calculated based on the focus position adjustment amount, and the target movement distance is converted into the number of step pulses to form a motor drive pulse sequence; at the same time, the direction information is encoded in the form of high and low levels, combined with the pulse data to form a motor control instruction, and sent to the motor controller through the drive interface to drive the motor to perform directional precise movement, so that the lens mechanism can complete the position closed-loop adjustment along the target focus direction.

[0039] Among them, before calculating the target moving distance and rotation direction of the focus motor according to the focus position adjustment amount, it also includes the steps of performing time series analysis and fusion optimization on the focus status data of multiple frames in a row, specifically including: establishing a time series data cache queue to store the focus status evaluation results, reflective interference area distribution data and bending radius compensation of N consecutive frames, and assigning a weight coefficient based on time attenuation to each frame of data to obtain a multi-frame time series data set; performing trend analysis on the focus status evaluation results in the multi-frame time series data set, calculating the focus trend slope through the first-order difference, calculating the focus acceleration through the second-order difference, identifying the change pattern and motion trajectory of the focus state, and obtaining the focus trend feature vector; performing a focus trend analysis based on the focus trend feature vector Focus behavior prediction: The focus state development trend of several frames in the future is predicted through a linear regression algorithm, and the prediction error is corrected in combination with the actual measurement value of the current frame to obtain the focus state prediction data; the multi-frame time series data set is time-series fused through a weighted average algorithm, where the recent frame data has a higher weight and the long-term frame data has a lower weight. At the same time, the fusion weight distribution is dynamically adjusted according to the focus stability index to obtain the focus evaluation result after time series fusion; a comprehensive judgment is made based on the focus evaluation result after time series fusion and the focus state prediction data. When the deviation between the prediction result and the fusion result exceeds the preset threshold, the focus error correction mechanism is activated. Otherwise, the fusion result is used as the final focus control basis to obtain the optimized focus position adjustment amount.

[0040] The reflective influence range is determined according to the spatial position coordinates of each reflective interference area, and before setting the suppression weight coefficient for the pixel points within the reflective influence range according to the reflection intensity classification, it also includes an adaptive weight dynamic allocation step based on scene recognition and lighting analysis, specifically including: scene type recognition of the current image data, calculating the brightness distribution characteristics through image histogram analysis, calculating the texture complexity characteristics through gradient statistical analysis, and calculating the color saturation characteristics through color space conversion analysis to obtain a scene feature descriptor; scene classification is performed based on the scene feature descriptor, and the shooting scene is divided into indoor low-light scenes, outdoor strong-light scenes, backlit scenes, uniformly illuminated scenes and complex lighting scenes, and a corresponding basic weight allocation strategy is preset for each scene type to obtain a scene-related weight distribution strategy. template; analyze the global illumination intensity distribution of the P-polarization channel image and the S-polarization channel image, calculate the illumination uniformity index, contrast index and dynamic range index, and evaluate the complexity of the current illumination conditions based on these indices to obtain the illumination complexity evaluation index; adaptively adjust the parameters in the weight distribution template according to the illumination complexity evaluation index, increase the reflection suppression weight coefficient when the illumination complexity is high, and reduce the reflection suppression weight coefficient when the illumination complexity is low, and adjust the attenuation parameter of the radial weight function at the same time to obtain the adaptively optimized weight distribution strategy; apply the adaptively optimized weight distribution strategy to the pixel weight setting within the reflection influence range, and establish a weight distribution history record for subsequent weight strategy learning and optimization to obtain the scene-adaptive suppression weight coefficient distribution.

[0041] Reference Figure 2 , this embodiment provides a lens focusing device, comprising: Acquisition module 1, used to obtain the focal length and aperture value of the zoom lens, and calculate the bending radius compensation amount of each radial area in the imaging surface; A reflection interference analysis module 2 is used to determine the spatial position coordinates of each reflection interference area based on the image data collected by the zoom lens; A focus state evaluation module 3 is configured to generate a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates; The focus control module 4 is configured to generate a focus control instruction according to the focus state evaluation result and drive the focus motor to move to a target position.

[0042] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0043] An embodiment of the present invention further provides a zoom lens, which is used to implement the above method.

[0044] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0045] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, is also included in the patent protection scope of the present invention.

Claims

1. A lens focusing method, characterized in that: include: Obtaining the focal length and aperture value of the zoom lens, and calculating the bending radius compensation amount of each radial area in the imaging plane; Determining the spatial position coordinates of each reflective interference area based on the image data collected by the zoom lens; generating a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates; A focus control instruction is generated according to the focus state evaluation result, and a focus motor is driven to move to a target position.

2. The lens focusing method according to claim 1, wherein: The obtaining of the focal length and aperture value of the zoom lens and the calculation of the bending radius compensation amount of each radial area in the imaging plane includes: Collecting zoom ring angle data of a zoom lens and synchronously collecting aperture ring gear position data of the zoom lens; querying a preset focal length calibration mapping table according to the zoom ring angle data to obtain a focal length value, and querying a preset aperture calibration mapping table according to the aperture ring gear data to obtain an aperture value; The bending radius compensation amount of each radial area in the imaging plane is calculated according to the focal length value and the aperture value.

3. The lens focusing method according to claim 2, wherein: Calculating the bending radius compensation amount of each radial area in the imaging plane according to the focal length value and the aperture value includes: querying a preset field curvature model parameter table to obtain a current field curvature model parameter corresponding to the focal length value; Performing a field curvature correction calculation based on the focal length value and the current field curvature model parameters to obtain a curvature radius value of the imaging surface; Dividing the imaging surface into a plurality of radial regions according to a preset concentric ring segmentation rule, and determining a geometric position deviation amount of each radial region according to the bending radius value and the aperture value; A radial weight function is applied to the geometric position deviation to perform weighted calculation to obtain a bending radius compensation amount for each radial area in the imaging plane.

4. The lens focusing method according to claim 1, wherein: The determining of the spatial position coordinates of each reflection interference area based on the image data collected by the zoom lens includes: Using a polarization analyzer to perform dual-channel separation processing on the image data collected by the zoom lens, respectively extracting a P polarization channel image and an S polarization channel image; Calculating a pixel-by-pixel intensity ratio of the P polarization channel image to the S polarization channel image to obtain a polarization intensity ratio image; Double threshold detection and local variance screening are performed on the polarization intensity ratio image to identify the spatial position coordinates of each reflection interference area.

5. The lens focusing method according to claim 4, wherein: The performing of dual threshold detection and local variance screening on the polarization intensity ratio image to identify the spatial position coordinates of each reflection interference area includes: Comparing each pixel of the polarization intensity ratio image with a first threshold and a second threshold, respectively, marking a pixel as a candidate reflection pixel when the pixel value is greater than both the first threshold and the second threshold, and determining a first pixel distribution map based on the candidate reflection pixels; Based on the first pixel distribution map, a rectangular sliding window of a preset size is used to calculate the local mean and local variance pixel by pixel, and when the local variance exceeds a preset variance threshold, the pixel point is retained, otherwise the pixel point is eliminated to obtain a second pixel distribution map; First, performing an opening operation of a circular structuring element with a specified radius on the second pixel distribution map to remove isolated pixel points, and then performing a closing operation of the circular structuring element to fill the holes inside the pixel area, to obtain a reflective area distribution map; Performing a connected domain labeling analysis on the reflective area distribution map to obtain a plurality of reflective interference areas; The centroid coordinates of all pixel coordinates in each reflection interference area are calculated, and the centroid coordinates are used as the spatial position coordinates of each reflection interference area.

6. The lens focusing method according to claim 1, wherein: The generating a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates includes: Determining the reflection influence range according to the spatial position coordinates of each reflection interference area, and setting a suppression weight coefficient for the pixel points within the reflection influence range according to the reflection intensity classification; Calculating the sharpness value of each pixel in the image data, and performing weighted suppression on the sharpness value in combination with the suppression weight coefficient to obtain sharpness evaluation data; According to the sharpness evaluation data, the sharpness value of each radial area is geometrically corrected using the corresponding bending radius compensation amount to obtain target sharpness data; The target sharpness data is weighted and summed to obtain a focus state evaluation result.

7. The lens focusing method according to claim 6, wherein: The calculating the sharpness value of each pixel in the image data, and performing weighted suppression on the sharpness value in combination with the suppression weight coefficient to obtain sharpness evaluation data, includes: Performing noise preprocessing on the image data using a Gaussian filter to obtain a filtered image, and then performing edge detection on the filtered image using a Laplace second-order differential operator to obtain a gradient intensity value of each pixel point; Taking the absolute value of the gradient intensity value as the sharpness value of each pixel point, and constructing a first sharpness matrix according to the sharpness value of each pixel point; According to the spatial position coordinates and the reflection influence range of each reflection interference area, the first sharpness matrix is ​​multiplied by the corresponding suppression weight coefficient for numerical attenuation to obtain a second sharpness matrix; A weighted average is performed on all valid pixel points in the second sharpness matrix to obtain a weighted average value, and the weighted average value is used as sharpness evaluation data.

8. The lens focusing method according to claim 7, wherein: The generating of a focus control instruction according to the focus state evaluation result and driving the focus motor to move to a target position includes: Calculating a proportional control amount and a differential control amount of the focus state evaluation result, and generating a first position adjustment amount according to the proportional control amount and the differential control amount; Calculating the sharpness gradient direction of the second sharpness matrix, determining the search step size and the moving direction according to the sharpness gradient direction, and dynamically adjusting the step speed according to the second sharpness matrix; adjusting the first position adjustment amount in real time according to the search step length, the moving direction, and the step speed to obtain a second position adjustment amount; Inputting the second position adjustment amount as an observation value into a Kalman filter, performing a recursive filtering operation through a state prediction equation and an observation update equation to obtain a focus position adjustment amount; The target moving distance and rotation direction of the focus motor are calculated according to the focus position adjustment amount, the target moving distance is converted into a driving pulse sequence and combined with the rotation direction to form a focus control instruction, and the focus motor is driven to move to the target position.

9. A lens focusing device, characterized in that: The lens focusing device is configured to implement the steps of the lens focusing method according to any one of claims 1 to 8, comprising: An acquisition module is used to obtain the focal length and aperture value of the zoom lens and calculate the bending radius compensation amount of each radial area in the imaging surface; a reflection interference analysis module, configured to determine the spatial position coordinates of each reflection interference area based on the image data collected by the zoom lens; a focus state evaluation module, configured to generate a focus state evaluation result based on the bending radius compensation amount and the spatial position coordinates; A focus control module is used to generate a focus control instruction according to the focus state evaluation result and drive the focus motor to move to a target position.

10. A zoom lens, characterized in that: The zoom lens is used to implement the steps of the lens focusing method according to any one of claims 1 to 8.

Images