Binocular vision naked eye 3D image processing method, device and system

Abstract

The application discloses a binocular vision naked-eye 3D image processing method, device and system, and particularly relates to the naked-eye 3D technical field; the application defines foreground and background areas, constructs a parameter extraction and optimal value quantization calculation system covering parallax, scene and splicing dimensions, and combines preset threshold values and weight factors in multiple scenes to realize comprehensive and accurate evaluation of naked-eye 3D image quality, solve the problem of one-sidedness of traditional single-index evaluation, and accurately locate the core crux affecting the stereoscopic effect.

Description

Technical Field

[0001] This invention relates to the field of naked-eye 3D technology, and more specifically, to a binocular vision naked-eye 3D image processing method, device, and system. Background Technology

[0002] With the iteration of display technology and the surge in demand for immersive experiences, glasses-free 3D technology has moved from the laboratory to large-scale commercial use, and is widely used in outdoor advertising screens, smart exhibitions, and other applications. The core advantage of glasses-free 3D technology lies in its ability to present stereoscopic visual effects without the need for auxiliary devices. Its key lies in reconstructing binocular parallax by simulating a real light field, allowing viewers to obtain a natural stereoscopic experience. However, in practical applications on large-screen displays, image quality directly determines the viewing experience, requiring a balance between parallax reasonableness, scene adaptability, and smooth splicing. This has become a core requirement for the technology's practical application.

[0003] However, the existing technology still has the following shortcomings:

[0004] First, traditional technologies lack a systematic, multi-dimensional quantitative evaluation system and rely heavily on a single indicator to judge image quality. They do not cover key dimensions such as parallax uniformity, parallax abrupt changes at stitching seams, and effective viewing angle, resulting in one-sided problem judgments and difficulty in accurately locating the core issues affecting stereoscopic effects.

[0005] Secondly, existing calibration methods rely heavily on manual trial and error, lacking a mechanism for reusing historical calibration experience. This is not only inefficient but also difficult to adapt to the dynamic image requirements of real-time projection on large screens. At the same time, the multi-unit collaborative optimization problem of splicing screens is prominent. Issues such as parallax consistency deviation and improper feathering width adaptation can easily lead to screen fragmentation, ghosting, or visual fatigue, seriously affecting the immersive experience.

[0006] To address this, a binocular vision naked-eye 3D image processing method, device, and system are proposed. Summary of the Invention

[0007] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a binocular vision naked-eye 3D image processing method, device and system.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A binocular vision naked-eye 3D image processing method includes the following steps:

[0010] Real-time image data analysis: Identify and preprocess the binocular parallax composite image projected on the large screen, extract parallax data, scene data and stitching data for comprehensive analysis to obtain image quality estimation, preset image quality estimation threshold, and mark images below the threshold as images to be processed;

[0011] Among them, the parallax data includes foreground parallax value, background parallax value and parallax uniformity; the scene data includes image brightness value, effective viewing angle range and image contrast value; and the stitching data includes the parallax abrupt change value at the edge of the stitching seam, the edge pixel feathering width and the parallax consistency deviation of the stitching unit.

[0012] Problem information integration: Integrate the core information of the images to be processed, generate standardized problem messages, and send them to the terminals of maintenance personnel in real time;

[0013] Matching and calibration strategy: After receiving the message, the operation and maintenance personnel match the image vector to be processed with the historical calibration strategy library and select the one with the highest similarity as the application strategy.

[0014] The specific process for determining the application strategy is as follows:

[0015] Based on the received standardized problem messages, the absolute differences between the disparity merit value, scene merit value, stitching merit value and their respective reference values ​​are calculated and used as the disparity deviation value, scene deviation value and stitching deviation value, respectively.

[0016] The parallax excellence value is obtained by comprehensively processing the foreground parallax deviation value, the background parallax deviation value, and the parallax uniformity.

[0017] The scene optimization value is obtained by comprehensively processing the standard deviation of image brightness, effective viewing angle range, and standard deviation of contrast.

[0018] The optimal stitching value is obtained by comprehensively processing the average disparity mutation value, feathering width value, and disparity consistency deviation value.

[0019] After multiplying the disparity deviation value, scene deviation value, and stitching deviation value by the corresponding preset weight factors, the best value type with the largest corresponding value is selected as the problem type. The deviation of each parameter and the problem type are extracted as core features and transformed into a structured feature vector according to preset rules.

[0020] Extract feature vectors from all cases in the historical tuning strategy library, and associate each case with corresponding tuning parameters and tuning effect labels;

[0021] The cosine similarity algorithm is used to calculate the similarity value between the feature vector of the image to be processed and the feature vector of each historical case. The adjustment strategy corresponding to the historical case with the highest similarity is selected as the application strategy for the image to be processed.

[0022] Image data re-inspection: The calibrated image is projected onto the large screen test area for re-inspection. If the image quality estimate is higher than the minimum allowable value, it is included in the playback queue; otherwise, it is sent to the management terminal.

[0023] Specifically, the analysis process for disparity data is as follows:

[0024] The image is used to identify targets, define the foreground and background regions, generate pixel coordinate masks for the two types of regions, and determine the selection range of feature points.

[0025] Randomly select several feature points in the foreground region, read the horizontal pixel coordinates of each point in the L and R images, calculate the pixel offset Δx of a single feature point, and then use the formula... ; This is the preset parallax conversion factor for outdoor large screens; The depth value corresponding to the feature point is converted into a physical disparity value. The arithmetic mean of the disparity values ​​of each feature point is taken as the foreground disparity value. A standard value for the foreground disparity value is preset. The absolute difference between the foreground disparity value and the standard value is calculated to obtain the foreground disparity deviation value.

[0026] A preset standard value for foreground disparity is used, and the absolute difference between the foreground disparity value and the standard value is calculated to obtain the foreground disparity deviation value.

[0027] Using the same logic as the foreground, select several feature points in the background area, calculate the pixel offset and convert it into disparity value, and take the average value as the background disparity value.

[0028] The background disparity deviation value is obtained by calculating the absolute difference between the background disparity value and the preset standard value.

[0029] Through formula The variance of the disparity values ​​of several feature points in the foreground region is obtained, where, The total number of feature points. For the disparity value of a single feature point, The mean foreground parallax is then used with the formula. The parallax uniformity is obtained.

[0030] Specifically, the process of analyzing scene data is as follows:

[0031] Calculate the grayscale values ​​of all pixels in the left eye view image and take the arithmetic mean. Extract the maximum brightness from the hardware parameters. Using the formula The image brightness value is obtained, and the absolute difference between the image brightness value and the pre-set image brightness standard value is calculated to obtain the image brightness standard deviation;

[0032] Input the raster period, screen width, and standard viewing distance of the corresponding scene, set the standard horizontal angle range, preset the interval angle, calculate the stereo separation degree of the split left and right eye images angle by angle, preset the separation degree standard, filter out the continuous angle interval that meets the standard, and calculate the symmetrical midpoint of the interval to obtain the effective viewing angle range.

[0033] Calculate the grayscale value pixel by pixel for the left-eye and right-eye view images and take the average value, then label them as follows: , Then, the average gray values ​​of the left and right eyes were calculated using the formula... Output A contrast ratio of 1.

[0034] The standard difference of contrast is obtained by calculating the absolute difference between the image contrast value and the preset standard contrast value.

[0035] Specifically, the process of analyzing the spliced ​​data is as follows:

[0036] By extracting local disparity sub-images with a width of several pixels on both sides of the stitching seam from the stitched binocular disparity composite image, pixel-level registration is performed on the two sub-images to ensure that the edge pixels of the two sub-images correspond one-to-one.

[0037] Traverse all corresponding pixels after registration, calculate the disparity value between the left and right disparity sub-images in spatial position based on the spatial position of the registered pixels, take the absolute value to obtain the disparity mutation value of each pixel, and take the average value to obtain the average disparity mutation value.

[0038] The formula for point-by-point calculation is: ,in, Let be the disparity jump value of the i-th pixel. Let be the disparity value of the i-th pixel in the left subimage. Let be the disparity value of the i-th pixel in the right sub-image;

[0039] Extract the average disparity mutation value, set the disparity mutation value intervals corresponding to the average disparity mutation value, and each disparity mutation value interval corresponds to a feathering width value. Match the average disparity mutation value with the corresponding interval to obtain the feathering width value.

[0040] For any two adjacent stitching units, identify the boundary region, uniformly select several groups of spatially corresponding pixels within the boundary region, count the disparity value of each group of corresponding pixels, calculate the absolute difference of the disparity values ​​of each group of corresponding pixels, and calculate the average value as the disparity consistency deviation value.

[0041] Specifically, the process of determining the image to be processed is as follows:

[0042] Set reference values ​​for parallax merit, scene merit, and stitching merit;

[0043] The image quality estimate is obtained by comprehensively processing the parallax merit, scene merit, and stitching merit.

[0044] The minimum allowable value corresponding to the preset image quality estimate is used to mark images whose image quality estimate is less than the minimum allowable value as images to be processed.

[0045] Specifically, the process of generating a standardized problem message is as follows:

[0046] The push message must include the image's unique identifier, parallax data, scene data, the original values ​​of the stitched data, the standard question type, and the specific data deviation value corresponding to the standard question.

[0047] H-1: Generate a unique ID for the image, extract disparity, scene and stitching measured parameters, and extract the reference threshold for the corresponding scene;

[0048] H-2: Compare the measured parameters with the reference thresholds and record the exceedance, deficiency, or compliance status in a unified format;

[0049] H-3: Based on the parameter deviation results, match the preset standard problem types and compile them into a problem list;

[0050] H-4: Integrates image ID, measured parameters, deviation description, and problem type to construct a standardized problem message, which is pushed to the maintenance personnel's terminal in real time.

[0051] A binocular vision naked-eye 3D image processing device, comprising:

[0052] High-definition image acquisition card: Extracts binocular parallax composite images;

[0053] Computational processing equipment: for image preprocessing, parameter extraction, and matching application strategies;

[0054] Naked-eye 3D display screen: projects 3D images for post-calibration retesting;

[0055] Data storage device: Stores historical calibration strategy library and image processing records.

[0056] A binocular vision naked-eye 3D image processing system, comprising:

[0057] Image analysis module: The system recognizes and preprocesses binocular parallax composite images, extracts parallax data, scene data and stitching data for comprehensive analysis to obtain image quality estimation, presets image quality estimation threshold, and marks images below the threshold as images to be processed;

[0058] Information integration module: integrates the core information of the images to be processed, generates standardized problem push messages, and sends them to the terminals of operation and maintenance personnel in real time;

[0059] Strategy matching module: After receiving the message, the operation and maintenance personnel's terminal matches the image vector to be processed with the historical calibration strategy library and selects the one with the highest similarity as the application strategy;

[0060] Data re-inspection module: The adjusted image is projected onto the large screen test area for re-inspection. If the image quality estimate is higher than the minimum allowable value, it is included in the playback queue; if it is lower, it is sent to the management terminal.

[0061] The technical effects and advantages of this invention are as follows:

[0062] (1) By defining the foreground and background areas, this invention constructs a parameter extraction and quality quantification calculation system covering parallax, scene, and stitching dimensions. Combined with multi-scene preset thresholds and weight factors, it achieves a comprehensive and accurate evaluation of naked-eye 3D image quality, solves the problem of one-sided evaluation by traditional single indicators, and accurately locates the core issues affecting stereoscopic effect.

[0063] (2) This invention generates feature vectors by normalizing the problem type of the image to be processed, and combines the cosine similarity algorithm to match the historical calibration strategy library, and associates calibration parameters such as disparity conversion coefficient, raster period, feathering width and effect labels, thereby realizing the transformation from manual trial and error to intelligent selection, greatly shortening the calibration time of a single image, improving calibration accuracy, and adapting to the needs of real-time dynamic image processing.

[0064] (3) This invention achieves multi-unit collaborative optimization of splicing screen by pixel-level registration of splicing seams, adaptation of average parallax mutation value to feathering width, optimization of parallax consistency between adjacent units, and closed-loop mechanism of post-calibration re-inspection. This effectively reduces the risk of screen fragmentation, ghosting and stereo distortion, ensures the smoothness and naturalness of naked-eye 3D visual effect, and reduces visual fatigue of users. Attached Figure Description

[0065] Figure 1 This is a flowchart of a binocular vision naked-eye 3D image processing method according to the present invention.

[0066] Figure 2 This is a schematic diagram of a binocular vision naked-eye 3D image processing system according to the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example 1

[0069] like Figure 1 As shown, the steps of a binocular vision naked-eye 3D image processing method are as follows:

[0070] Real-time image data analysis: After preprocessing the images by projecting binocular parallax composite images onto a large screen, parallax data, scene data, and stitching data are extracted and comprehensively analyzed to obtain image quality estimates. A preset image quality estimate threshold is set, and images with image quality estimates lower than the corresponding threshold are taken as images to be processed.

[0071] Additional notes: Data preprocessing includes synchronously capturing the binocular parallax composite image currently projected onto the large screen using a high-definition image capture card (the acquisition frequency is consistent with the large screen refresh rate, 25-30 frames / second), calling the DepthNet model to separate the left eye view map (L map) and the right eye view map (R map), removing image noise and correcting pixel distortion, laying the foundation for subsequent data extraction;

[0072] Among them, the parallax data includes foreground parallax value, background parallax value and parallax uniformity; the scene data includes image brightness value, effective viewing angle range and image contrast value; and the stitching data includes the parallax abrupt change value at the edge of the stitching seam, the edge pixel feathering width and the parallax consistency deviation of the stitching unit.

[0073] Specifically:

[0074] By calling the YOLOv8 semantic segmentation algorithm to perform target recognition on the image, the foreground region (core visual elements such as advertising products and brand logos, accounting for ≥30%) and the background region (secondary elements such as sky and street) are defined, and pixel coordinate masks of the two types of regions are generated to determine the selection range of feature points.

[0075] Several feature points are randomly selected in the foreground region. The number of selections is preset by technicians and must be greater than 3 and a positive integer. The horizontal pixel coordinates of each point in the L and R images are read, and the pixel offset Δx of a single feature point is calculated using the formula. ( This is the preset parallax conversion factor for outdoor large screens, usually set to 0.02; The depth value corresponding to the feature point (depth value range 0-255) is converted to physical disparity value. Then, the arithmetic mean of the disparity values ​​of each feature point is taken as the foreground disparity value. ;

[0076] A preset standard value for foreground disparity is used to calculate the foreground disparity deviation value by taking the absolute difference between the foreground disparity value and the standard value. ;

[0077] Several feature points are selected in the background area. The number of points is preset by technicians and must be greater than 3 and be a positive integer. The pixel offset is calculated and converted into a disparity value. The average value is taken as the background disparity value. The background disparity deviation value is obtained by calculating the absolute difference between the background disparity value and the standard value. ;

[0078] Through formula The variance of the disparity values ​​of several feature points in the foreground region is obtained. ,in, The total number of feature points. For the disparity value of a single feature point, The mean foreground parallax is then used with the formula. Obtain parallax uniformity ;

[0079] In addition, parallax uniformity can reflect the smoothness of the stereoscopic effect in the core area (foreground) of naked-eye 3D. High uniformity can prevent core elements (such as advertising products) from ghosting or stereoscopic distortion due to abrupt changes in parallax, making the viewing experience natural and comfortable.

[0080] After normalizing the foreground disparity deviation value, background disparity deviation value, and disparity uniformity, the formula is used. The disparity figure of merit is obtained by weighted calculation. ;in , , These are the corresponding preset weighting factors;

[0081] For example, taking outdoor large screen scenarios:

[0082] Preset standard values: foreground parallax 10 arcminutes, background parallax 4 arcminutes, parallax uniformity 90%;

[0083] Actual collected values: foreground parallax 12 arcminutes, background parallax 5 arcminutes, parallax uniformity 85%;

[0084] Deviations: Foreground deviation 2 arcminutes, background deviation 1 arcminute, parallax uniformity deviation 5%;

[0085] Normalized results: Foreground 0.6, Background 0.8, Parallax uniformity 0.94;

[0086] Final parallax estimate: 78 points, which can guarantee a smooth 3D effect for the foreground advertisement with no obvious ghosting, meeting the requirements for outdoor naked-eye 3D display;

[0087] Calculate the grayscale values ​​of all pixels in the left-eye view image and take the arithmetic mean. Extract the maximum brightness from the hardware parameters. (Outdoor large screens are typically rated at 1500 cd / ㎡, while indoor screens are rated at 800 cd / ㎡), using the formula The image brightness value is obtained, and preset image brightness standard values ​​are used for different scenes. The standard deviation of image brightness is calculated by the absolute difference between the image brightness value and the standard value. ;

[0088] Using optical simulation tools, the screen's grating period, screen width, and standard viewing distance for the corresponding scene are input. A standard horizontal angle range is set, and a preset interval angle is used to calculate the stereo separation degree of the split left and right eye images angle by angle. A preset separation degree standard is set, and continuous angle intervals that meet the standard are selected. The effective viewing angle range is obtained by calculating the symmetrical midpoint of the interval. ;

[0089] For example, taking an outdoor naked-eye 3D screen as an example, input its grating period of 0.1mm, screen width of 8m and standard viewing distance of 5-50m into the optical simulation tool, set the horizontal angle range of -45° to +45° with 1° intervals, calculate the stereo separation degree of the split left and right eye images angle by angle, preset the separation degree ≥80% as the effective standard, filter out the continuous angle range of -16° to 16° that meets the standard, calculate the symmetrical midpoint of this range, and finally obtain the effective viewing angle range of the screen as ±16°;

[0090] According to the disparity shift rule, the binocular disparity composite image is divided into independent left-eye and right-eye view images. The grayscale value of each pixel in the two images is calculated and the average value is taken, and they are labeled as follows: , Then, the average gray values ​​of the left and right eyes were calculated using the formula... Output A contrast ratio of 1.

[0091] Preset standard contrast values ​​for different scenes, and calculate the standard difference of contrast by taking the absolute difference between the image contrast value and the standard contrast value. ;

[0092] For example, assuming the left-eye and right-eye view maps are separated from the binocular parallax composite image, the following calculations are made:

[0093] Average gray value of left eye =140;

[0094] Average gray value of the right eye =100;

[0095] After substituting into the formula, we get =0.33;

[0096] The output image contrast ratio is 0.33:1, indicating that there is a certain difference in brightness between the left and right eye view images;

[0097] The default standard contrast ratio for outdoor naked-eye 3D large screen scenes is 0.25:1 (to meet the visual comfort requirements of outdoor strong light environments); the absolute difference between the actual screen contrast ratio of 0.33:1 and this standard value is calculated to obtain the standard difference of contrast ratio.

[0098] After normalizing the standard deviation of image brightness, effective viewing angle, and standard deviation of contrast, the formula was used. Obtain scene excellence value ,in , , These are the corresponding preset weighting factors;

[0099] From the stitched binocular parallax composite image, a local parallax sub-image with a width of several pixels on both sides of the stitching seam is extracted. The number of sub-images is preset by the technicians and must be greater than 3 and be a positive integer. Pixel-level registration is performed on the two sub-images to ensure that the edge pixels of the two sub-images correspond one-to-one.

[0100] Iterate through all corresponding pixels after registration. Based on the spatial position of the registered pixels, calculate the disparity value between the left and right disparity sub-images in spatial position point by point, take the absolute value to obtain the disparity abrupt change value for each pixel, and take the average value to obtain the average disparity abrupt change value. ;

[0101] The formula for point-by-point calculation is: ,in, Let be the disparity jump value of the i-th pixel. Let be the disparity value of the i-th pixel in the left subimage. Let be the disparity value of the i-th pixel in the right sub-image;

[0102] For example, in the stitched disparity map, three pixels on each side of the stitching seam are cropped, and after registration, the disparity values ​​of the corresponding pixels are as follows:

[0103] Left side: 12, 15, 9

[0104] Right side: 10, 14, 11

[0105] The calculated average parallax jump value is approximately 1.67, indicating that the parallax jump of this seam is very slight, and the seam is almost invisible to the human eye.

[0106] The average disparity abrupt change value is extracted. Based on the characteristics of large-screen scenes, disparity abrupt change value intervals are defined for each set of values. Each set of disparity abrupt change value intervals corresponds to a set of feathering width values. The feathering width value is obtained by matching the average disparity abrupt change value with the corresponding interval. ;

[0107] For example,

[0108] The average parallax abrupt change value of a certain seam is 1.8 parallax units → the corresponding feathering width value is 2.5;

[0109] The average parallax abrupt change value of a certain seam is 3.5 parallax units → the corresponding feathering width value is 4;

[0110] The average parallax abrupt change value of a certain seam is 5.2 parallax units → the corresponding feathering width value is 6.5;

[0111] For any two adjacent stitching units, identify the boundary region. Within the boundary region, uniformly select several groups of spatially corresponding pixels. Calculate the disparity value of each group of corresponding pixels, and then calculate the absolute difference between the disparity values ​​of each group and use the average value as the disparity consistency deviation value. ;

[0112] Taking two adjacent splicing units A and B on the left and right sides of the large screen as an example:

[0113] Identify the vertical boundary region between the two, and uniformly select 5 groups of pixels whose spatial positions are completely corresponding within the region;

[0114] Read the disparity value of each group of points. For example, the disparity values ​​of cell A are 12, 15, 11, 14, and 13, and the corresponding values ​​of cell B are 13, 14, 12, 14, and 12.

[0115] The absolute difference of disparity was calculated for each group, resulting in 1, 1, 1, 0, 1;

[0116] Calculate the average absolute difference: (1+1+1+0+1)÷5=0.8, to obtain the disparity consistency deviation value between units A and B;

[0117] After normalizing the mean disparity abrupt change value, feathering width value, and disparity consistency deviation value, the formula was used. Obtain the splicing excellence value ,in , , These are the corresponding preset weighting factors;

[0118] Based on differences in scene characteristics, reference values ​​are set for parallax merit, scene merit, and stitching merit, and are respectively labeled as follows: , , ;

[0119] After normalizing the parallax, scene, and stitching merit values, the formula is used. Obtain image quality valuation ,in , , These are the corresponding preset weighting factors;

[0120] The minimum allowable value corresponding to the preset image quality estimation is used to mark images whose image quality estimation is less than the minimum allowable value as images to be processed.

[0121] Problem information integration: The core information of the images to be processed is integrated to form a standardized problem push message, which is then sent to the terminal of the operation and maintenance personnel;

[0122] The push message must include the image's unique identifier (image ID), parallax data, scene data, the original values ​​of the stitched data, the standard question type, and the specific data deviation value corresponding to the standard question (e.g., "Foreground parallax value 18 arcminutes, exceeding the threshold limit by 3 arcminutes").

[0123] Specifically:

[0124] H-1: Generate a unique ID for the image, extract disparity, scene and stitching measured parameters, and extract the reference threshold for the corresponding scene;

[0125] H-2: Compare the measured parameters with the reference thresholds and record the exceedance, deficiency, or compliance status in a unified format;

[0126] H-3: Based on the parameter deviation results, match the preset standard problem types and compile them into a problem list;

[0127] H-4: Integrates image ID, measured parameters, deviation description, and problem type to construct image problem information, which is then pushed to the maintenance personnel's terminal in real time.

[0128] For example, generate image ID20260123-LED-006-00101 and extract the measured parameters: foreground disparity value 18 arcminutes, effective viewing angle ±10°, and average disparity abrupt change value 5.2; retrieve outdoor scene reference thresholds: foreground disparity ≤15 arcminutes, effective viewing angle ≥±16°, and average disparity abrupt change value ≤3.0.

[0129] After comparison, the following records were obtained: the foreground disparity exceeded the upper limit of the threshold by 3 arc minutes, the effective viewing angle was lower than the lower limit of the threshold by ±6°, and the average disparity mutation value exceeded the upper limit of the threshold by 2.2.

[0130] Matching standard problem types: excessive foreground parallax, too narrow effective viewing angle, severe parallax abrupt change at splicing seams, summarized into a problem list.

[0131] Integrate all image problem information and push it to the maintenance personnel's terminal in real time;

[0132] Matching application strategy: After receiving the image problem information push on the terminal, the operation and maintenance personnel will match the image vector to be processed with the historical calibration strategy library and select the one with the highest similarity as the application strategy for the image to be processed.

[0133] Specifically:

[0134] Based on the received image problem information, the absolute difference between the disparity merit value, scene merit value, and stitching merit value and the corresponding reference value is calculated as the disparity deviation value, scene deviation value, and stitching deviation value, respectively.

[0135] After multiplying the disparity deviation value, scene deviation value, and stitching deviation value by the corresponding preset weight factors, the type with the largest corresponding good value is selected as the problem type. The deviation of each parameter and the problem type are extracted as core features and transformed into a structured feature vector according to preset rules.

[0136] For example, one-hot encoding can be used to classify problem types, and parameter deviation values ​​can be normalized.

[0137] Extract feature vectors from all historical cases in the strategy library, and associate each case with corresponding calibration parameters (such as disparity conversion coefficient adjustment value, raster period optimization value, feathering width determination value) and calibration effect label (meeting standards / not meeting standards).

[0138] The cosine similarity algorithm is used to calculate the similarity value between the feature vector of the image to be processed and the feature vector of each historical case. The adjustment strategy corresponding to the historical case with the highest similarity is selected as the application strategy for the image to be processed.

[0139] For example, consider the problem of excessive foreground parallax and an excessively narrow effective viewing angle on outdoor naked-eye 3D large screens:

[0140] Extract the core features of the image to be processed: the parameter deviation is "foreground disparity value 18 arcminutes, exceeding the threshold by 3 arcminutes; effective viewing angle ±10°, narrower than the threshold by 6°", and the problem type is "foreground disparity exceeds the standard, effective viewing angle is too narrow"; perform One-Hot encoding on the problem type (foreground disparity exceeds the standard = 1, effective viewing angle is too narrow = 1, splicing seam abrupt change = 0), normalize the deviation value (disparity deviation 3 / 5 = 0.6, viewing angle deviation 6 / 10 = 0.6), and generate the feature vector [1,1,0,0.6,0.6].

[0141] Retrieve historical strategy library examples: Example 1 has a feature vector of [1,1,0,0.5,0.5], associated with the adjustment parameters "parallax conversion coefficient k adjusted to 0.017, raster period adjusted to 0.09mm", and the effect label is "meets the standard"; Example 2 has a feature vector of [1,0,0,0.6,0], associated with the adjustment parameter "k adjusted to 0.018", and the effect label is "meets the standard".

[0142] The cosine similarity algorithm was used to calculate that the similarity between the vector to be processed and Case 1 was 0.98, and the similarity between Case 2 and Case 2 was 0.85. The adjustment parameters of Case 1, which had the highest similarity, were selected as the application strategy for this application.

[0143] Image data re-inspection: The calibrated image is re-projected onto the large screen test area for re-inspection; if the image quality estimate is greater than the minimum allowable value, the corresponding image is included in the large screen playback queue; if the image quality estimate is less than the minimum allowable value, the image is sent to the corresponding management personnel terminal.

[0144] Devices for binocular vision naked-eye 3D image processing include:

[0145] High-definition image acquisition card: Captures binocular parallax composite images in real time at 25-30 frames per second (consistent with the frame rate of large screens), providing synchronous, high-quality raw data for subsequent processing;

[0146] Computing and processing equipment: Equipped with core algorithms such as DepthNet and YOLOv8, it completes image preprocessing, multi-dimensional parameter extraction, excellence value calculation and tuning strategy matching, and is the core decision-making unit of the entire process;

[0147] Maintenance personnel terminal: Receives standardized messages containing image ID, deviation details, and problem type in real time, intuitively displaying image problems and assisting staff in quick management and decision-making;

[0148] Naked-eye 3D display screen: The main screen is used for commercial scenarios (outdoor advertising, smart exhibitions) to project 3D images, while the test area screen is used for re-testing after calibration to verify whether the effect meets the standards;

[0149] Data storage devices: store historical calibration strategy libraries, multi-scenario preset parameters, and image processing records, providing data support for intelligent matching, process traceability, and optimization;

[0150] Color analyzer: Extracts RGB color values ​​from image sampling areas and converts them into CIE1976Lab coordinates, calculates color deviation values, and provides data support for scene quality assessment;

[0151] Optical simulation equipment: Input the screen grating period, width and standard viewing distance, calculate the stereo separation from angle to angle, filter the effective angle range and accurately determine the effective viewing angle range;

[0152] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0153] Example 2

[0154] Please see Figure 2As shown, based on the binocular vision naked-eye 3D image processing method provided in Embodiment 1 of this application, Embodiment 2 of this application proposes a binocular vision naked-eye 3D image processing system. Embodiment 2 is merely a preferred embodiment of Embodiment 1, and the implementation of Embodiment 2 will not affect the individual implementation of Embodiment 1.

[0155] Specifically, Embodiment 2 of this application provides a binocular vision naked-eye 3D image processing system, comprising:

[0156] Image analysis module: The system recognizes and preprocesses binocular parallax composite images, extracts parallax data, scene data and stitching data for comprehensive analysis to obtain image quality estimation, presets image quality estimation threshold, and marks images below the threshold as images to be processed;

[0157] Information integration module: integrates the core information of the images to be processed, generates standardized problem push messages, and sends them to the terminals of operation and maintenance personnel in real time;

[0158] Strategy matching module: After receiving the message, the operation and maintenance personnel's terminal matches the image vector to be processed with the historical calibration strategy library and selects the one with the highest similarity as the application strategy;

[0159] Data re-inspection module: The adjusted image is projected onto the large screen test area for re-inspection. If the image quality estimate is higher than the minimum allowable value, it is included in the playback queue; if it is lower, it is sent to the management terminal.

[0160] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0161] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0162] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0163] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0164] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0165] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0166] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable ATA hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0167] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A binocular vision naked-eye 3D image processing method, characterized in that, Includes the following steps: Real-time image data analysis: Identify and preprocess the binocular parallax composite image projected on the large screen, extract parallax data, scene data and stitching data for comprehensive analysis to obtain image quality estimation, preset image quality estimation threshold, and mark images below the threshold as images to be processed; Among them, the parallax data includes foreground parallax value, background parallax value and parallax uniformity; the scene data includes image brightness value, effective viewing angle range and image contrast value; and the stitching data includes the parallax abrupt change value at the edge of the stitching seam, the edge pixel feathering width and the parallax consistency deviation of the stitching unit. Problem information integration: Integrate the core information of the images to be processed, generate standardized problem messages, and send them to the terminals of maintenance personnel in real time; Matching and calibration strategy: After receiving the message, the operation and maintenance personnel match the image vector to be processed with the historical calibration strategy library and select the one with the highest similarity as the application strategy. The specific process for determining the application strategy is as follows: Based on the received standardized problem messages, the absolute differences between the disparity merit value, scene merit value, stitching merit value and their respective reference values ​​are calculated and used as the disparity deviation value, scene deviation value and stitching deviation value, respectively. The parallax excellence value is obtained by comprehensively processing the foreground parallax deviation value, background parallax deviation value, and parallax uniformity. The scene optimization value is obtained by comprehensively processing the standard deviation of image brightness, effective viewing angle range, and standard deviation of contrast. The optimal stitching value is obtained by comprehensively processing the average disparity mutation value, feathering width value, and disparity consistency deviation value. After multiplying the disparity deviation value, scene deviation value, and stitching deviation value by the corresponding preset weight factors, the best value type with the largest corresponding value is selected as the problem type. The deviation of each parameter and the problem type are extracted as core features and transformed into a structured feature vector according to preset rules. Extract the feature vectors of all cases in the historical tuning strategy library, and associate each case with the corresponding tuning parameters and tuning effect labels; The cosine similarity algorithm is used to calculate the similarity value between the feature vector of the image to be processed and the feature vector of each historical case. The adjustment strategy corresponding to the historical case with the highest similarity is selected as the application strategy for the image to be processed. Image data re-inspection: The calibrated image is projected onto the large screen test area for re-inspection. If the image quality estimate is higher than the minimum allowable value, it is included in the playback queue; otherwise, it is sent to the management terminal.

2. The binocular vision naked-eye 3D image processing method according to claim 1, characterized in that, The process of analyzing disparity data is as follows: The image is used to identify targets, define the foreground and background regions, generate pixel coordinate masks for the two types of regions, and determine the selection range of feature points. Randomly select several feature points in the foreground region, read the horizontal pixel coordinates of each point in the L and R images, calculate the pixel offset Δx of a single feature point, and then use the formula... ; This is the preset parallax conversion factor for outdoor large screens; The depth value corresponding to the feature point is converted into a physical disparity value, and the arithmetic mean of the disparity values ​​of each feature point is taken as the foreground disparity value. A preset standard value for foreground disparity is used, and the absolute difference between the foreground disparity value and the standard value is calculated to obtain the foreground disparity deviation value. A preset standard value for foreground disparity is used, and the absolute difference between the foreground disparity value and the standard value is calculated to obtain the foreground disparity deviation value. Using the same logic as the foreground, select several feature points in the background area, calculate the pixel offset and convert it into disparity value, and take the average value as the background disparity value. The background disparity deviation value is obtained by calculating the absolute difference between the background disparity value and the preset standard value. Through formula The variance of the disparity values ​​of several feature points in the foreground region is obtained, where, The total number of feature points. For the disparity value of a single feature point, The mean foreground parallax is then used with the formula. The parallax uniformity is obtained.

3. The binocular vision naked-eye 3D image processing method according to claim 1, characterized in that, The specific process of scene data analysis is as follows: Calculate the grayscale values ​​of all pixels in the left eye view image and take the arithmetic mean. Extract the maximum brightness from the hardware parameters. Using the formula The image brightness value is obtained, and the absolute difference between the image brightness value and the pre-set image brightness standard value is calculated to obtain the image brightness standard deviation; Input the raster period, screen width, and standard viewing distance of the corresponding scene, set the standard horizontal angle range, preset the interval angle, calculate the stereo separation degree of the split left and right eye images angle by angle, preset the separation degree standard, filter out the continuous angle interval that meets the standard, and calculate the symmetrical midpoint of the interval to obtain the effective viewing angle range. Calculate the grayscale value pixel by pixel for the left-eye and right-eye view images and take the average value, then label them as follows: , Then, the average gray values ​​of the left and right eyes were calculated using the formula... ; Output A contrast ratio of 1. The standard difference of contrast is obtained by calculating the absolute difference between the image contrast value and the preset standard contrast value.

4. The binocular vision naked-eye 3D image processing method according to claim 1, characterized in that, The specific process of analyzing the spliced ​​data is as follows: By extracting local disparity sub-images with a width of several pixels on both sides of the stitching seam from the stitched binocular disparity composite image, pixel-level registration is performed on the two sub-images to ensure that the edge pixels of the two sub-images correspond one-to-one. Traverse all corresponding pixels after registration, calculate the disparity value between the left and right disparity sub-images in spatial position based on the spatial position of the registered pixels, take the absolute value to obtain the disparity mutation value of each pixel, and take the average value to obtain the average disparity mutation value. The formula for point-by-point calculation is: ,in, Let be the disparity jump value of the i-th pixel. Let be the disparity value of the i-th pixel in the left subimage. Let be the disparity value of the i-th pixel in the right sub-image; Extract the average disparity mutation value, set the disparity mutation value intervals corresponding to the average disparity mutation value, and each disparity mutation value interval corresponds to a feathering width value. Match the average disparity mutation value with the corresponding interval to obtain the feathering width value. For any two adjacent stitching units, identify the boundary region, uniformly select several groups of spatially corresponding pixels within the boundary region, count the disparity value of each group of corresponding pixels, calculate the absolute difference of the disparity values ​​of each group of corresponding pixels, and calculate the average value as the disparity consistency deviation value.

5. The binocular vision naked-eye 3D image processing method according to claim 1, characterized in that, The specific process for determining the image to be processed is as follows: Set reference values ​​for parallax merit, scene merit, and stitching merit; The image quality estimate is obtained by comprehensively processing the parallax merit, scene merit, and stitching merit. The minimum allowable value corresponding to the preset image quality estimate is used to mark images whose image quality estimate is less than the minimum allowable value as images to be processed.

6. The binocular vision naked-eye 3D image processing method according to claim 1, characterized in that, The specific process for generating a standardized problem message is as follows: The push message must include the image's unique identifier, parallax data, scene data, the original values ​​of the stitched data, the standard question type, and the specific data deviation value corresponding to the standard question. H-1: Generate a unique ID for the image, extract disparity, scene and stitching measured parameters, and extract the reference threshold for the corresponding scene; H-2: Compare the measured parameters with the reference thresholds and record the exceedance, deficiency, or compliance status in a unified format; H-3: Based on the parameter deviation results, match the preset standard problem types and compile them into a problem list; H-4: Integrates image ID, measured parameters, deviation description, and problem type to construct a standardized problem message, which is pushed to the maintenance personnel's terminal in real time.

7. A binocular vision naked-eye 3D image processing device, employing the binocular vision naked-eye 3D image processing method described in claims 1-6, comprising: High-definition image acquisition card: Extracts binocular parallax composite images; Computational processing equipment: for image preprocessing, parameter extraction, and matching application strategies; Naked-eye 3D display screen: projects 3D images for post-calibration retesting; Data storage device: Stores historical calibration strategy library and image processing records.

8. A binocular vision naked-eye 3D image processing system, applying the binocular vision naked-eye 3D image processing method described in claims 1-6, comprising: Image analysis module: The system recognizes and preprocesses binocular parallax composite images, extracts parallax data, scene data and stitching data for comprehensive analysis to obtain image quality estimation, presets image quality estimation threshold, and marks images below the threshold as images to be processed; Information integration module: integrates the core information of the images to be processed, generates standardized problem push messages, and sends them to the terminals of operation and maintenance personnel in real time; Strategy matching module: After receiving the message, the operation and maintenance personnel's terminal matches the image vector to be processed with the historical calibration strategy library and selects the one with the highest similarity as the application strategy; Data re-inspection module: The adjusted image is projected onto the large screen test area for re-inspection. If the image quality estimate is higher than the minimum allowable value, it is included in the playback queue; if it is lower, it is sent to the management terminal.

Images