Multi-view panoramic image fusion display method and system based on DA intelligent interconnected screen

Abstract

This invention relates to the field of image sensing technology and discloses a method and system for multi-view panoramic image fusion display based on a DA smart interconnected screen. The method includes: acquiring images of the vehicle's surrounding environment; correcting pixels of synchronous time frames to obtain multi-view images; converting to grayscale images and gradient aggregation to generate boundary feature maps; extracting curvature difference features according to resolution; calculating correction values ​​based on curvature differences to construct a mapping table; resampling the images to obtain curvature correction maps; determining overlapping view areas and equalizing brightness to obtain boundary-optimized images; extracting color features to form cross-view texture restoration to obtain fusion preparation images. This method can solve the problem of boundary mismatch in multi-view images.

Description

Technical Field

[0001] This invention relates to the field of image sensing technology, and in particular to a multi-view panoramic image fusion display method and system based on a DA smart interconnected screen. Background Technology

[0002] Image sensing technology utilizes image acquisition devices to receive, convert, transmit, and represent optical information in a target scene. It is widely used in intelligent transportation, driver assistance systems, industrial inspection, security monitoring, and human-machine interaction. In vehicle intelligence applications, image fusion display methods are widely used in DA intelligent interconnected screens. These screens process environmental images captured by cameras at the front, rear, left, and right sides of the vehicle and output them to the driver's display interface, providing the driver with visualized information about the vehicle's surroundings.

[0003] Existing multi-view image fusion display methods applied to DA smart interconnected screens, such as CN117237237A, involve capturing environmental images from the front, rear, left, and right cameras of the vehicle, performing time synchronization, distortion correction, and registration processing on each image, mapping the processed multi-view images onto the same display plane, and fusing them according to preset stitching rules within the overlapping areas of adjacent views. Finally, the fused image is output to the DA smart interconnected screen. However, due to differences in the installation position, pitch angle, and lens characteristics of each camera, even after basic correction, inconsistent boundary curvature still exists in the overlapping areas of images from different perspectives.

[0004] In summary, existing technologies suffer from mismatches in the boundary stitching of multi-view images. Summary of the Invention

[0005] This invention provides a method and system for multi-view panoramic image fusion display based on DA smart interconnected screen to solve the problem of mismatch in the boundary connection of multi-view images.

[0006] Firstly, to address the aforementioned technical problems, this invention provides a multi-view panoramic image fusion display method based on a DA smart interconnected screen, comprising: acquiring images of the vehicle's surrounding environment; performing pixel correction on the images of the vehicle's surrounding environment to obtain preliminarily corrected multi-view image data; extracting boundary feature maps from the multi-view image data and performing hierarchical curvature analysis on the boundary feature maps to obtain curvature difference features; performing pixel resampling on the multi-view image data based on the curvature difference features to obtain a curvature-corrected image; determining overlapping regions of viewpoints based on the curvature-corrected image, extracting local exposure features of the overlapping regions of viewpoints, and performing brightness equalization on the overlapping regions of viewpoints based on the local exposure features to obtain a boundary-optimized image with consistent brightness; extracting color distribution feature maps of common areas from the boundary-optimized image, forming cross-view texture data based on the color distribution feature maps, and repairing the boundary-optimized image based on the cross-view texture data to obtain a fusion preparation image.

[0007] Secondly, the present invention provides a multi-view panoramic image fusion display system based on a DA smart interconnected screen, comprising: a data acquisition module for acquiring images of the vehicle's surrounding environment; a preliminary correction module for performing pixel correction on the images of the vehicle's surrounding environment to obtain preliminarily corrected multi-view image data; a curvature analysis module for extracting boundary feature maps of the multi-view image data and performing hierarchical curvature analysis on the boundary feature maps to obtain curvature difference features; a curvature correction module for performing pixel resampling on the multi-view image data based on the curvature difference features to obtain a curvature-corrected image; a boundary optimization module for determining the view overlap region based on the curvature-corrected image, extracting the local exposure features of the view overlap region, and performing brightness equalization on the view overlap region based on the local exposure features to obtain a boundary-optimized image with consistent brightness; and an image fusion module for extracting color distribution feature maps of common areas from the boundary-optimized image, forming cross-view texture data based on the color distribution feature maps, and repairing the boundary-optimized image based on the cross-view texture data to obtain a fusion preparation image.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention first performs time synchronization and geometric distortion correction on the images of the vehicle's surrounding environment, then extracts the boundary feature map and performs hierarchical curvature analysis to obtain curvature difference features, and constructs a curvature mapping table to resample the multi-view images. Since the differences in the installation position and lens characteristics of different cameras will be quantified as curvature differences and directly corrected, the misalignment and contour stretching of the splicing boundary can be reduced from the geometric shape, and finally the smooth alignment of multi-view images in the overlapping area can be achieved. (2) Based on curvature correction, this invention further extracts the local exposure features of the overlapping area of ​​the viewpoint, including the mean brightness, standard deviation and the position of the main peak of the histogram, and generates the brightness adjustment value of each window accordingly, and performs distance-weighted gradual correction along the splicing boundary; by compressing the brightness difference between adjacent viewpoints, the brightness jump at the splicing seam is avoided, and the brightness transition of the overlapping area is made more natural and continuous. (3) The present invention extracts the common area from the boundary optimization image, statistically analyzes the RGB distribution of each window to obtain the color distribution feature map, determines the window with small color deviation as the color corresponding window and extracts cross-view texture data, and then repairs the broken area with texture matching degree below the threshold by using neighboring pixel distance weighting; by eliminating color difference and texture breakage, the color and texture of different viewpoints at the splicing boundary are kept consistent, and finally a visually seamless fusion preparation image is obtained. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the multi-view panoramic image fusion display method based on DA smart interconnected screen provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a multi-view panoramic image fusion display system based on a DA smart interconnected screen provided in the second embodiment of the present invention. Detailed Implementation

[0010] 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.

[0011] Reference Figure 1 The first embodiment of the present invention provides a multi-view panoramic image fusion display method based on a DA smart interconnected screen, including the following steps: S11, acquire images of the vehicle's surrounding environment; S12, perform pixel correction on the image of the environment surrounding the vehicle to obtain pre-corrected multi-view image data; S13, extract the boundary feature map of the multi-view image data, and perform hierarchical curvature analysis on the boundary feature map to obtain curvature difference features; S14, perform pixel resampling on the multi-view image data based on the curvature difference characteristics to obtain a curvature-corrected image; S15, determine the overlapping area of ​​the viewpoint based on the curvature-corrected image, extract the local exposure features of the overlapping area of ​​the viewpoint, and perform brightness equalization on the overlapping area of ​​the viewpoint based on the local exposure features to obtain a boundary-optimized image with consistent brightness. S16, extract the color distribution feature map of the common area from the boundary optimization image, form cross-view texture data based on the color distribution feature map, repair the boundary optimization image based on the cross-view texture data, and obtain the fusion preparation image.

[0012] In step S11, images of the surrounding environment of the vehicle are acquired.

[0013] It is worth noting that the vehicle's surrounding environment image is continuously captured by four cameras installed in front of, behind, to the left and right of the vehicle. The images include environmental information such as road boundaries, adjacent vehicles, pedestrians, curbs and parking lines.

[0014] In step S12, pixel correction is performed on the vehicle surrounding environment image to obtain preliminarily corrected multi-view image data, including: taking each pair of the four original images in the vehicle surrounding environment image as a time-series comparison image pair, and calculating the timestamp difference of the time-series comparison image pair to obtain timestamp features; when the timestamp features are less than a preset overlap threshold, the time-series comparison image pair is regarded as a synchronous time frame image; based on the vehicle surrounding environment image, distortion parameter matching is performed on the synchronous time frame image to obtain a lens distortion feature set; based on the original coordinates of each pixel in the vehicle surrounding environment image and the lens distortion feature set, geometric distortion correction features are calculated; based on the geometric distortion correction features, pixel remapping processing is performed on the synchronous time frame image to obtain preliminarily corrected multi-view image data.

[0015] It is worth noting that the frame header time values ​​of the four original images of the vehicle's surrounding environment (front, rear, left, and right) are read, and these four original images are used as time-series comparison image pairs. Then, the difference between the frame header time values ​​of the two original images in the time-series comparison image pairs is calculated to obtain the timestamp feature.

[0016] If the timestamp feature is less than the preset overlap threshold of 0.05 seconds, the time-series comparison image pair is determined to be a synchronized time frame image. After obtaining the synchronized time frame images, all synchronized time frame image pairs are combined and merged to obtain the synchronized frame index feature.

[0017] After obtaining the synchronized time frame images, distortion parameter matching is performed on each image in the synchronized frame group, combining the pixel coordinate information already retained in the original images, to obtain lens distortion features. Specifically, the intrinsic parameter data and distortion coefficient data saved by the corresponding camera during the calibration phase are first read, and then these data are matched point-by-point with the pixel coordinates of the current synchronized frame image. The distortion coefficients include radial distortion coefficients and tangential distortion coefficients, and the pixel coordinate information includes the lateral and longitudinal positions of each pixel. For any pixel, the original pixel coordinates are first converted to a normalized coordinate plane based on the principal point position and focal length parameters. Then, the squared distance from that point to the imaging center is calculated. This distance value is then substituted into the radial distortion correction form and the tangential offset correction form to obtain the lateral and longitudinal offset of that pixel. This process is repeated for all pixels in the entire image to obtain the lens distortion feature set corresponding to the current synchronized frame group.

[0018] After obtaining the lens distortion feature set, geometric distortion correction features are calculated based on the original coordinates and offset of each pixel. Specifically, a position relocation calculation is performed on each pixel in the synchronized frame image, and the original pixel coordinates are superimposed with the corresponding horizontal and vertical offsets to obtain the corrected target coordinates. Then, a coordinate mapping table is generated based on the target coordinates of all pixels.

[0019] After obtaining the geometric distortion correction features, pixel-by-pixel remapping processing is performed on the four images (front, back, left, and right) in the synchronized frame group to obtain preliminary corrected images. Specifically, the pixel value in the original image is searched point-by-point according to the coordinate mapping table, and this pixel value is written to the target coordinate position. When the target position falls on a non-integer coordinate, the grayscale values ​​of its surrounding pixels are read, and the output value of the current target point is calculated using a distance-weighted method, followed by pixel filling. This yields the forward-view corrected image, the back-view corrected image, the left-view corrected image, and the right-view corrected image. After obtaining the four preliminary corrected images, the unified frame number, synchronization judgment result, overlapping area identifier, and the corresponding corrected front, back, left, and right image frames are combined and stored to finally obtain the preliminary corrected multi-view image data.

[0020] Distance-weighted calculation uses bilinear interpolation, which takes the four neighboring pixels around the target coordinates as the basis for calculation, calculates the horizontal and vertical distances from the target point to the four neighboring pixels respectively, and then determines the weight of each pixel by the inverse ratio of the distances.

[0021] The preset overlap threshold is determined based on the frame rate of the multi-view camera and the allowable positional offset of the target within the overlapping area. Specifically, the single frame period is first calculated based on the camera's frame rate. When the frame rate is 30 frames per second, the single frame period is 1 divided by 30 seconds, which is 33.3 milliseconds. Then, considering the additional latency introduced by software cache reading, timestamp writing, and channel transmission, to ensure that the displacement of the moving target within the overlapping area between two adjacent frames does not exceed one pixel, the maximum allowable time mismatch is controlled within one frame period based on common vehicle speeds and camera field of view. An additional 50% margin is reserved on this basis to cope with software processing jitter. Therefore, the upper limit of the allowable time mismatch is set to 1.5 times the single frame period, resulting in 33.3 multiplied by 1.5 equals 49.95 milliseconds, which is rounded down to 0.05 seconds as the preset overlap threshold.

[0022] In step S13, the boundary feature map of the multi-view image data is extracted, and a graded curvature analysis is performed on the boundary feature map to obtain curvature difference features. This includes: weighted summation of the red channel values, green channel values, and blue channel values ​​of each pixel in the multi-view image data to obtain a single-channel grayscale image; performing first-order difference operations on the single-channel grayscale image in the horizontal and vertical directions to obtain the gradient magnitude and gradient direction of each pixel; and aggregating the single-channel grayscale image based on the gradient magnitude and gradient direction to obtain a boundary feature map; classifying the boundary feature map into levels according to a preset resolution classification rule to obtain resolution level features; and extracting continuous boundary points sequentially from the boundary feature map based on the resolution level features, calculating the directional change between the continuous boundary points and the ratio between the directional change and the boundary arc length to obtain the local curvature value of the corresponding grid, and subtracting and averaging the local curvature values ​​to obtain curvature difference features.

[0023] It is worth noting that the red, green, and blue channel values ​​of each pixel in each color image of the multi-view image data are weighted and summed to obtain a single-channel grayscale value, where the weight of the red channel is 0.28, the weight of the green channel is 0.59, and the weight of the blue channel is 0.13. While generating the single-channel grayscale image, the red, green, and blue channel data from the corresponding multi-view image data are retained. The weight of each channel is determined by photographing a 24-color chart under standard lighting conditions, calculating the correlation coefficient between each channel's pixel value and the standard value, and then dividing the correlation coefficient of the corresponding color by the sum of the correlation coefficients of the three colors to obtain the weight of the corresponding color.

[0024] After grayscale conversion, first-order difference operations are performed on the grayscale image in both the horizontal and vertical directions to obtain the grayscale change value of each pixel in the horizontal and vertical directions. Then, the gradient magnitude and gradient direction are calculated based on these values. Subsequently, a two-level filtering process is applied to the gradient magnitude using a low threshold of 50 and a high threshold of 150. If the gradient magnitude of a pixel is greater than or equal to 150, the pixel is directly marked as a strong edge point. If the gradient magnitude of a pixel is greater than 50 but less than 150, the pixel is marked as a candidate edge point. If the gradient magnitude of a pixel is less than 50, the pixel is determined to be a non-edge point and is discarded. Finally, the remaining edge points are aggregated according to connected regions, merging interconnected edge points into edge regions, ultimately obtaining the boundary feature map within the overlapping regions.

[0025] The weights for the red, green, and blue channels are determined by dividing the correlation coefficients of each color channel in the historical samples by the sum of the correlation coefficients of the three color channels. Subsequently, regions with a resolution of 1920 x 1080 in the boundary feature map are classified as high-level, regions with a resolution of 1280 x 720 as medium-level, and regions with a resolution of 640 x 480 as low-level.

[0026] It should be noted that the resolution levels are divided based on the camera's imaging distance. High-level resolution corresponds to the near-field area within 5 meters of the vehicle, medium-level resolution corresponds to the mid-field area between 5 and 15 meters, and low-level resolution corresponds to the far-field area beyond 15 meters. The adjustment coefficients are determined based on the correlation between curvature correction and boundary fit improvement in historical samples. Curvature correction has the greatest impact on the improvement in high-level areas, so the coefficient is set to 0.90; medium-level resolution is set to 0.75; and low-level resolution is set to 0.60. The specific values ​​were obtained by conducting a grid curvature correction experiment on 2000 frames of samples, statistically analyzing the reduction ratio of boundary pixel position error after correction for each level of area, and taking the average reduction ratio of samples for each level as the adjustment coefficient. The experiment showed that the average reduction ratio was 90% for high-level areas, 75% for medium-level areas, and 60% for low-level areas, so the corresponding values ​​were set to 0.90, 0.75, and 0.60.

[0027] The adjustment coefficients for high-level, medium-level, and low-level areas are directly set according to the correlation coefficient between the curvature correction amount and the improvement in boundary fit in historical samples, and the coefficients are controlled within the range of 0 to 1.

[0028] After obtaining the resolution level features, five consecutive boundary points are extracted sequentially along the boundary line within each grid of the boundary feature map. First, the directional change between adjacent boundary points is calculated, and then the ratio between the directional change and the boundary arc length is calculated. This ratio is used as the local curvature value at that location. Subsequently, the local curvature values ​​of the corresponding grids of the two viewpoint images within the same overlapping area are subtracted to obtain the curvature difference value. Finally, all curvature differences within a grid are averaged to generate the curvature difference feature of that grid.

[0029] In step S14, the step of constructing a curvature mapping table based on the curvature difference features and performing pixel resampling on the multi-view image data according to the curvature mapping table to obtain a curvature-corrected image includes: normalizing the curvature difference features to obtain a normalized curvature difference value; multiplying the normalized curvature difference value by a preset adjustment coefficient to obtain a curvature correction value; then writing the grid number of the curvature correction value in the boundary feature map and the curvature correction value into the mapping table to obtain a curvature mapping table; and performing pixel resampling on the multi-view image data based on the curvature mapping table to obtain a curvature-corrected image.

[0030] It's worth noting that the maximum and minimum curvature difference values ​​of all grids within the current boundary feature map are then statistically analyzed. Next, the minimum curvature difference value of any grid is subtracted from the minimum value, and then divided by the difference between the maximum and minimum values. This maps all curvature difference values ​​to a uniform range of zero to one. The corresponding adjustment coefficients are then written according to the resolution level. High-resolution regions are more sensitive to boundary changes, so the adjustment coefficient is set to 0.90; medium-resolution regions are set to 0.75; and low-resolution regions are set to 0.60. The normalized curvature difference values ​​are then multiplied by the adjustment coefficients to obtain the curvature correction value for each grid. The grid number, curvature difference value, adjustment coefficient, and curvature correction value are then written into a table in regional order to form a curvature mapping table.

[0031] Subsequently, curvature correction values ​​are read grid-by-grid from the curvature map table, and the boundary points within the grid are offset along the boundary normal direction. Then, position rearrangement and pixel resampling are performed on the pixels on both sides of the boundary. During resampling, for target positions falling on non-integer coordinates, the color values ​​of the four surrounding adjacent pixels are read, weighted by distance to obtain new output pixel values, and then sequentially filled into the target coordinates. After the above processing, the curvature-corrected image is obtained.

[0032] In step S15, the overlapping viewing area is determined based on the curvature-corrected image, the local exposure features of the overlapping viewing area are extracted, and the brightness of the overlapping viewing area is equalized based on the local exposure features to obtain a boundary-optimized image with consistent brightness. This includes: taking each pair of the four original images in the curvature-corrected image as an adjacent viewing angle combination, extracting the stitching boundary position of the adjacent viewing angle combination in the curvature-corrected image, calculating the effective coverage width based on the stitching boundary position, and if the effective coverage width is greater than a preset coverage threshold, then the adjacent viewing angle combination is included in the overlapping viewing area; the overlapping viewing area is divided according to a preset window size, and the mean brightness feature, standard deviation brightness feature, and main peak position feature of the gray-level histogram for each window are statistically analyzed. Based on the mean brightness feature and the main peak position feature of the gray-level histogram, the target brightness value and brightness compensation amount for each window are determined, and combined with the standard deviation brightness feature, the brightness adjustment value for each window is generated. The brightness adjustment value is written to the corresponding window position to obtain an equalized overlapping region. A preset pixel width is extended to both sides along the stitching boundary of the equalized overlapping region to form a boundary band region. The brightness difference between adjacent pixels within the boundary band region is calculated to obtain a brightness gradient feature. Boundary candidate points are then selected and merged based on the brightness gradient feature to obtain a boundary pixel set. A corresponding weight is assigned to each pixel in the boundary pixel set based on its distance to the center line of the stitching boundary to obtain a pixel weight distribution feature. Based on the pixel weight distribution feature, a shallow edge correction is performed on the equalized overlapping region to obtain a gradient-adjusted boundary region. The gradient-adjusted boundary region replaces the corresponding region in the equalized overlapping region to obtain a boundary-optimized image with consistent brightness. The shallow edge correction refers to performing distance-weighted smoothing on pixel brightness within the transition band formed by extending to both sides along the center line of the stitching boundary, resulting in a gradient effect in brightness changes, similar to a soft transition of a light-colored edge.

[0033] It is worth noting that the viewpoint identifiers and pixel size information of the four views (front, back, left, and right) of the curvature-corrected image are read. Then, a correspondence between adjacent viewpoints is established according to the camera installation position: front and left views are matched as a front-left combination, front and right views as a front-right combination, back and left views as a back-left combination, and back and right views as a back-right combination. The stitching boundary position of each group of adjacent viewpoint images is extracted. Then, based on the effective coverage width between the boundary positions of the two image paths, the effective coverage width is compared with a preset coverage threshold. When the effective coverage width is greater than the preset coverage threshold, the corresponding area is determined as the viewpoint overlap area. For example, the effective column range of the front view image in this candidate area is columns 118 to 276, and the effective column range of the left view image at the corresponding position is columns 132 to 268. Therefore, the cross-coverage range of the two images is columns 132 to 268, corresponding to an effective coverage width of 136 pixels. The effective coverage width is then compared with a preset coverage threshold, which can be set to 120 pixels. When the effective coverage width of 136 pixels is greater than 120 pixels, the overlapping coverage area is determined as the view overlap area.

[0034] The preset coverage threshold was determined through statistical experiments on the coverage width of overlapping areas. Specifically, under standard test conditions, on a dry asphalt road surface with an ambient illuminance of 300 to 500 lux, using the same vehicle model and the same camera model, 2000 valid samples were collected. The effective coverage width of the overlapping area between adjacent viewpoints in each frame was measured, yielding a mean of 138 pixels and a standard deviation of 9 pixels. To ensure that the vast majority of overlapping areas are included, a conservative strategy of subtracting twice the standard deviation from the mean was adopted, setting the preset coverage threshold to 120 pixels. At this value, approximately 97.5% of the overlapping areas in historical samples can be correctly identified.

[0035] Subsequently, the overlapping area is divided into blocks of 8×8 pixels, and each window is scanned sequentially from left to right and from top to bottom. The brightness values ​​of all pixels within each window are summed and divided by the total number of pixels in the window to obtain the mean brightness feature of that window. The dispersion is calculated based on the difference between the brightness values ​​of each pixel within the window and the mean brightness, to obtain the standard deviation brightness feature of that window. Then, the frequency of the distribution of brightness values ​​within the window in the range of 0 to 255 is statistically analyzed to determine the position of the brightness interval with the highest frequency, thus obtaining the main peak position feature of the grayscale histogram for that window. After completing the statistical analysis of all windows, the mean brightness feature, standard deviation brightness feature, and main peak position feature of the grayscale histogram for each window are arranged according to the row and column order of the original window to obtain the local exposure feature map. For example, the average brightness of an 8×8 window in the front left overlapping area near the parking line is 180, the standard deviation of brightness is 12, and the main peak of the grayscale histogram is 175; the average brightness of an 8×8 window in the corresponding position of the rear right overlapping area is 140, the standard deviation of brightness is 10, and the main peak of the grayscale histogram is 138, thus obtaining the local exposure characteristics of the two corresponding windows.

[0036] The system reads the mean brightness, standard deviation brightness, and main peak position of the grayscale histogram from two perspective images in terms of local exposure features. Then, it averages the mean brightness values ​​of the two images at the same window location to obtain the target brightness value for that window. Next, it subtracts the original mean brightness values ​​of the two windows from the target brightness value to obtain the brightness compensation amount for the corresponding window. Finally, it allocates the brightness compensation amount based on the standard deviation brightness feature, generating brightness adjustment values ​​for each window, and writes these values ​​to the corresponding window positions. After generating brightness adjustment values ​​for all windows, it rewrites the adjusted pixel brightness of each window back into the overlapping area image according to the original row and column order of the windows, resulting in an equalized overlapping area.

[0037] For example, if the average brightness of a window in the current left overlapping region is 180, the standard deviation of brightness is 12, and the main peak position of the grayscale histogram is 175, and the average brightness of the corresponding window in the subsequent right overlapping region is 140, the standard deviation of brightness is 10, and the main peak position of the grayscale histogram is 138, then the average brightness of the two windows is first averaged to obtain a target brightness value of 160. Then, the brightness compensation amount of the front left window is calculated as -20, and the brightness compensation amount of the subsequent right window is calculated as +20. Based on this, the brightness adjustment value of the corresponding window is generated, and finally, the equalized window region corresponding to that position is obtained.

[0038] After obtaining the equalization window region, the pixel width is extended to the left and right sides along the splicing boundary in the equalization overlap region to form a boundary band region. Then, the brightness value of each pixel in the boundary band region is read line by line according to the boundary normal direction, and the brightness difference between adjacent pixels is calculated to obtain the brightness gradient feature of each position in the boundary band. Based on the brightness gradient feature, continuous abrupt change positions are selected, and pixels with brightness differences greater than a preset brightness difference threshold are marked as boundary candidate points. Adjacent boundary candidate points are merged according to the connectivity relationship to obtain the boundary pixel set.

[0039] After extracting the boundary pixel set, the distance from each pixel on both sides of the boundary to the boundary center line is calculated based on the stitched boundary center line. A corresponding weight is assigned to each pixel based on this distance, resulting in a pixel weight distribution feature. The pixel weight at the boundary center line is set to one, and the weight decreases by a preset step size for each additional pixel on either side. Then, the brightness difference value corresponding to each pixel in the boundary pixel set is read, and this difference is multiplied by its corresponding weight to obtain the brightness adjustment value for each boundary pixel. This brightness adjustment value is then superimposed onto the corresponding pixel position in the equalization overlap region to obtain the gradient-adjusted boundary region. Finally, the gradient-adjusted boundary region is merged with the unadjusted out-of-boundary region according to their original spatial positions, outputting a boundary-optimized image with consistent brightness.

[0040] For example, adjacent equalization window regions are stitched together to form an equalization overlap region. A boundary band is formed by extending five pixels to the left and right sides along the stitching boundary. The brightness values ​​within the boundary band are calculated row by row, resulting in brightness differences of 12, 15, and 18 at some locations. Pixels with brightness differences greater than a preset brightness difference threshold of 10 are marked as boundary candidate points, and consecutively distributed candidate points are merged into a boundary pixel set. Then, the distance from each pixel on both sides of the boundary centerline to the centerline is calculated. The centerline position weight is set to 1, the position weight at a distance of one pixel is set to 0.8, and the position weight at a distance of two pixels is set to 0.6. The corresponding brightness difference is then multiplied by the weight to obtain the brightness adjustment value for each pixel. Finally, the brightness adjustment value is written back to the corresponding pixel position to form a boundary-optimized image with consistent brightness.

[0041] The preset pixel width was determined through boundary transition compensation experiments. Specifically, in multiple overlapping area samples with completed brightness equalization, 2, 3, 4, 5, 6, and 7 pixels were set as candidate single-sided expansion widths, respectively, using the center line of the stitching boundary as a reference. Then, the same gradient correction was performed on the boundary area under each candidate width, and the average brightness difference of the corresponding pixels on both sides of the stitching boundary was calculated. The candidate width with the smallest average brightness difference was selected as the preset pixel width. According to experimental statistics, when the single-sided expansion width is 5 pixels, the average brightness difference of the corresponding pixels on both sides of the stitching boundary is the smallest. Therefore, the preset pixel width was set to 5 pixels.

[0042] A preset brightness difference threshold is used to filter brightness abrupt change points near the splicing boundary, and its value is determined through brightness abrupt change recognition experiments. Specifically, manually labeled brightness abrupt change points and non-brightness abrupt change points are first extracted from the overlapping area samples that have completed brightness equalization, and then the brightness difference between each point and its adjacent pixels is calculated. Subsequently, 6, 8, 10, 12, and 14 are set as candidate brightness difference thresholds, and the detection rate of abrupt change points and the false detection rate of non-brightness abrupt change points under each candidate threshold are calculated. Then, the ratio of the detection rate to the false detection rate corresponding to each candidate threshold is calculated, and the candidate value with the smallest ratio is selected as the preset brightness difference threshold. According to experimental statistics, the ratio of the detection rate to the false detection rate is the smallest when the candidate brightness difference threshold is 10. Therefore, the preset brightness difference threshold is set to 10.

[0043] The preset step size is determined based on the single-sided transition width and the weight reduction rule. Specifically, the pixel weight at the center line of the splicing boundary is first set to 1, and the pixel weight is required to decrease gradually with increasing pixel distance within the single-sided transition width, decaying to 0 at the edge of the transition zone. Then, the initial weight at the center line of the boundary is divided by the single-sided transition width to obtain the weight reduction amount between adjacent pixel positions, and this reduction amount is used as the preset step size. For example, when the single-sided transition width is set to 5 pixels, the preset step size is calculated as 0.20 by dividing 1 by 5. At this time, the weight at the center line of the boundary is 1, the weight at the position 1 pixel away from the center line is 0.80, the weight at the position 2 pixels away from the center line is 0.60, and so on, decreasing sequentially until the weight at the edge of the transition zone is 0.

[0044] In step S16, the step of extracting the color distribution feature map of the common area from the boundary optimization image, forming cross-view texture data based on the color distribution feature map, and repairing the boundary optimization image based on the cross-view texture data to obtain a fusion preparation image includes: extracting the common area from the boundary optimization image according to the adjacent view combination, dividing the common area into windows, and statistically analyzing the distribution information of the red, green, and blue channels of each window to obtain a color distribution feature map; determining the color deviation value between each window based on the color distribution feature map, and identifying the window whose color deviation value is less than a preset color threshold as... Color-corresponding windows; read the original pixel arrangement information of the color-corresponding windows to obtain cross-view texture data; based on the cross-view texture data, calculate the texture matching degree of each window, merge windows with texture matching degrees lower than a preset matching threshold according to their positional relationship to obtain fracture region features, extract neighboring pixels around the fracture region in the fracture region features, and determine the corresponding weights according to the distance of the neighboring pixels to the repair position in the fracture region features to generate a repair pixel set; according to the original spatial position of the fracture region features, write the repair pixel set back to the corresponding region in the boundary optimization image to obtain a fusion preparation image.

[0045] It's worth noting that, based on adjacent viewpoints (front left, front right, back left, and back right), the common area on both sides of the stitching boundary is extracted from the boundary-optimized image with consistent brightness. This common area is then divided into 8x8 pixel blocks. For each block, the mean, variance, and number of pixels in each brightness range for the red, green, and blue channels are calculated to form a color histogram for that block. After all windows are statistically analyzed, they are arranged in their original row and column order to obtain the color distribution feature map.

[0046] After obtaining the color distribution feature map, a window in the common area of ​​the previous viewpoint is used as the reference window. Based on the center coordinates and stitching boundary position of the reference window, a search range is set along the normal and tangent directions of the stitching boundary in the common area of ​​the other viewpoint to extract several candidate windows. Then, the average values ​​of the red, green, and blue channels of each candidate window are read, and the inter-window color deviation value between them and the reference window is calculated. The candidate window with the smallest color deviation value that is less than a preset color threshold is determined as the color-corresponding window. After completing the window correspondence, the pixel color values ​​arranged in row and column order in the two color-corresponding windows are read, the horizontal and vertical position relationships of each pixel in the window are retained, and the color change sequence in the horizontal direction and the color change sequence in the vertical direction between adjacent pixels are calculated. At the same time, the main edge direction information of the edge pixels in the window is extracted. Then, the correspondence between the two windows and their respective texture descriptions are written into the same data unit to obtain the cross-view texture data at that position.

[0047] The preset color threshold was determined through a candidate threshold screening experiment using multiple scene samples. Specifically, overlapping window pairs from adjacent front left, front right, back left, and back right views were first selected from historical data. The historical data simultaneously covered dry asphalt pavement, concrete pavement, grass areas, parking space marking areas, traffic sign areas, and areas with vehicles of different colors to ensure the integrity of the color distribution range. Then, each window pair was manually labeled to classify them into window pairs with the same texture area and window pairs with different texture areas. Subsequently, the mean difference of the red channel, the mean difference of the green channel, and the mean difference of the blue channel for each window pair were calculated, and the three were added together to obtain the color deviation value between windows. Then, nine, eleven, thirteen, fifteen, and seventeen were set as candidate color thresholds, and the retention rate of window pairs with the same texture area and the mismatch rate of window pairs with different texture areas were calculated for each candidate threshold. The difference between the retention rate and the mismatch rate was calculated as an indicator. The retention rate refers to the proportion of the number of windows correctly identified as color-corresponding to all window pairs with the same texture area, and the mismatch rate refers to the proportion of the number of windows incorrectly identified as color-corresponding to all window pairs with different texture areas. Experimental results show that when the candidate color threshold is thirteen, the comprehensive judgment index reaches its maximum. At this time, it can retain most of the window pairs with the same texture region and effectively suppress the mismatch of window pairs with different texture regions. Therefore, the preset color threshold is set to thirteen.

[0048] After obtaining the cross-view texture data, the horizontal color change sequences, vertical color change sequences, and edge direction information of the two color-corresponding windows are matched and calculated to obtain the texture matching degree of the window. Specifically, the color change sequences of the two windows in the horizontal direction are read item by item according to their numbers, and the absolute difference of the color change values ​​at corresponding positions is taken. Then, all absolute differences are averaged to obtain the horizontal color difference value. Then, the absolute difference of the color change sequences in the vertical direction is taken item by item in the same way and averaged to obtain the vertical color difference value. For the edge direction information, the angle values ​​corresponding to the main edge directions of the two windows are read, the absolute value of the angle between them is calculated, and the angle difference greater than 180 degrees is converted into its supplementary angle to obtain the edge direction difference value. Subsequently, the horizontal color difference value, vertical color difference value, and edge direction difference value are normalized to the range of zero to one, wherein the horizontal color difference value and vertical color difference value are normalized according to the maximum possible difference of the color change sequence, and the edge direction difference value is normalized according to the range of zero to ninety degrees; then the normalized horizontal color difference value, the normalized vertical color difference value, and the normalized edge direction difference value are weighted and summed according to preset weights to obtain the texture difference value. Finally, the texture difference value is subtracted from one to obtain the texture matching degree of the window.

[0049] The preset weights for the normalized horizontal color difference value, the normalized vertical color difference value, and the normalized edge direction difference value are determined based on the proportion of their correlation coefficients with the texture continuity annotation results. Specifically, firstly, color windows manually labeled as texture continuity and texture breakage are extracted from historical samples. The horizontal color difference value, vertical color difference value, and edge direction difference value of each window are calculated separately. Then, the correlation coefficients between the above three parameters and the texture continuity annotation results are calculated separately. For example, if the correlation coefficient corresponding to the normalized horizontal color difference value is 0.76, the correlation coefficient corresponding to the normalized vertical color difference value is 0.63, and the correlation coefficient corresponding to the normalized edge direction difference value is 0.51, then the sum of the three correlation coefficients is 1.90. Then, dividing 0.76 by 1.90 yields a normalized horizontal color difference value weight of 0.40; dividing 0.63 by 1.90 yields a normalized vertical color difference value weight of 0.33; and dividing 0.51 by 1.90 yields a normalized edge direction difference value weight of 0.27. Therefore, the texture difference value can be obtained by weighting and summing the three normalized difference values ​​according to 0.40, 0.33, and 0.27 respectively.

[0050] After obtaining the texture matching score, the texture matching score of each window is compared with a preset matching threshold one by one. Windows with a score below the threshold are marked as low-matching windows. Then, adjacent low-matching windows are merged according to row and column relationships to obtain continuous low-matching regions. Subsequently, the boundary range and coordinate position of each continuous region are extracted to form the fracture region features.

[0051] A preset matching threshold is used to distinguish between texture continuity windows and texture breakage windows, and its value is determined through a candidate threshold screening experiment. Specifically, firstly, windows manually labeled as texture continuity and windows manually labeled as texture breakage are extracted from historical samples, and the texture matching degree of each window is calculated. Then, 0.65, 0.68, 0.71, 0.74, and 0.77 are set as candidate matching thresholds, and the retention rate of texture continuity windows and the misclassification rate of texture breakage windows are statistically analyzed under each candidate threshold. The candidate matching threshold with the largest difference between the retention rate and the misclassification rate is selected as the preset matching threshold. According to the experimental statistics, the difference between the retention rate and the misclassification rate is the largest when the candidate matching threshold is 0.71. Therefore, the preset matching threshold is set to 0.71.

[0052] After obtaining the fracture region, all pixels within the fracture region are marked as pixels to be repaired, and adjacent pixels outside the fracture region are marked as effective boundary pixels. Then, for each location to be repaired, the nearest effective boundary pixels are searched along the fracture boundary normal direction to both sides of the fracture region, with that location as the center. Adjacent effective boundary pixels are also searched along the fracture boundary tangent direction, forming a repair sampling set. After determining the repair sampling set, the distance from each effective boundary pixel to the current location to be repaired is calculated, and the reciprocal of the distance is used as the basic weight; the smaller the distance, the larger the corresponding weight. Simultaneously, the basic weight is adjusted based on the consistency between the effective boundary pixels and the fracture boundary normal direction, so that effective pixels located in the boundary extension direction have a higher participation ratio. Then, the red, green, and blue channel values ​​of each effective boundary pixel are multiplied by their corresponding weights, summed, and divided by the sum of all weights to obtain the red, green, and blue channel repair values ​​for the current location to be repaired, thus generating repair pixels and forming a repair pixel set.

[0053] For example, suppose the current location to be repaired is at column 215, row 84, and this location has already been marked as a pixel inside the fractured area. Then, the original pixel value at this location is not read for repair. First, search along the normal direction of the fracture boundary to find the nearest valid boundary pixel outside the fractured area, obtaining three valid pixels at column 214, row 84, column 216, row 84, and column 215, row 83. Each of these pixels is 1 pixel away from the current location to be repaired, and each has a base weight of 1. Then, search along the tangent direction of the boundary to find column 214, row 83, which is √2 pixels away from the current location to be repaired, with a base weight of 1 divided by √2, approximately 0.71. Then, sum the red, green, and blue channel values ​​of the above four valid pixels according to their respective weights, and divide each sum by the total weight to obtain the repair pixel value for column 215, row 84. Repeat the above process for the remaining locations to be repaired within the fractured area to form the repair pixel set.

[0054] The preset maximum weight is used to limit the upper limit of the weight of the nearest effective boundary pixel outside the fracture region under the regularized distance weight function. Specifically, the inverse of the distance from the effective boundary pixel outside the fracture region to the current position to be repaired is first used as the base weight, and the one-sided transition width is used as the regularized distance parameter, so that the weight when the distance is 1 pixel is normalized to 1; on this basis, when the regularized distance approaches 0, its maximum weight is determined according to the ratio of the one-sided transition width plus 1 to the one-sided transition width. For example, when the one-sided transition width is set to 5 pixels, the preset maximum weight is calculated by adding 1 and dividing by 5, which gives 1.2, so the preset maximum weight is set to 1.2.

[0055] Based on the original spatial location of the fractured area in the fractured area features, the original pixels within the fractured area are replaced point by point with the corresponding repair pixels, while keeping the pixel values ​​of other locations outside the fractured area unchanged; then the replaced area is written back to the corresponding position in the boundary optimization image with consistent brightness to obtain the fusion preparation image, and output to the DA smart interconnected screen.

[0056] In summary, this invention discloses a multi-view panoramic image fusion and display method based on a DA smart interconnected screen. The method includes acquiring images of the vehicle's surrounding environment, performing pixel correction on synchronous time frames to obtain multi-view images, converting them to grayscale images, gradient aggregation to generate boundary feature maps, and extracting curvature difference features according to resolution. Based on the curvature differences, correction values ​​are calculated to construct a mapping table, and the images are resampled to obtain a curvature correction map. Overlapping view areas are identified and brightness is balanced to obtain an optimized boundary image. Color features are extracted to form a cross-view texture restoration image, thus solving the problem of mismatched boundary connections in multi-view images.

[0057] Reference Figure 2 The second embodiment of the present invention provides a multi-view panoramic image fusion display system based on a DA smart interconnected screen, comprising: a data acquisition module for acquiring images of the vehicle's surrounding environment; a preliminary correction module for performing pixel correction on the images of the vehicle's surrounding environment to obtain preliminarily corrected multi-view image data; a curvature analysis module for extracting boundary feature maps of the multi-view image data and performing hierarchical curvature analysis on the boundary feature maps to obtain curvature difference features; a curvature correction module for performing pixel resampling on the multi-view image data based on the curvature difference features to obtain a curvature-corrected image; a boundary optimization module for determining the view overlap region based on the curvature-corrected image, extracting the local exposure features of the view overlap region, and performing brightness equalization on the view overlap region based on the local exposure features to obtain a boundary-optimized image with consistent brightness; and an image fusion module for extracting color distribution feature maps of common areas from the boundary-optimized image, forming cross-view texture data based on the color distribution feature maps, and repairing the boundary-optimized image based on the cross-view texture data to obtain a fusion preparation image.

[0058] It should be noted that the multi-view panoramic image fusion display system based on DA smart interconnected screen provided in this embodiment of the invention is used to execute all the process steps of the multi-view panoramic image fusion display method based on DA smart interconnected screen in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0059] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0060] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for multi-view panoramic image fusion display based on a DA smart interconnected screen, characterized in that, include: Acquire images of the vehicle's surrounding environment; perform pixel correction on the images of the vehicle's surrounding environment to obtain preliminarily corrected multi-view image data; extract boundary feature maps from the multi-view image data, and perform hierarchical curvature analysis on the boundary feature maps to obtain curvature difference features; Based on the curvature difference features, pixel resampling is performed on the multi-view image data to obtain a curvature-corrected image; based on the curvature-corrected image, the view overlap region is determined, the local exposure features of the view overlap region are extracted, and the brightness of the view overlap region is equalized based on the local exposure features to obtain a boundary-optimized image with consistent brightness; the color distribution feature map of the common area is extracted from the boundary-optimized image, and cross-view texture data is formed based on the color distribution feature map; the boundary-optimized image is repaired based on the cross-view texture data to obtain a fusion preparation image.

2. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 1, characterized in that, The step of performing pixel correction on the vehicle surrounding environment image to obtain preliminarily corrected multi-view image data includes: taking each pair of four original images from the vehicle surrounding environment image as a time-series comparison image pair, and calculating the timestamp difference of the time-series comparison image pairs to obtain timestamp features; when the timestamp features are less than a preset overlap threshold, the time-series comparison image pairs are considered as synchronous time frame images; based on the vehicle surrounding environment image, performing distortion parameter matching on the synchronous time frame images to obtain a lens distortion feature set; calculating geometric distortion correction features based on the original coordinates of each pixel in the vehicle surrounding environment image combined with the lens distortion feature set; and performing pixel remapping processing on the synchronous time frame images based on the geometric distortion correction features to obtain preliminarily corrected multi-view image data.

3. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 1, characterized in that, The step of extracting the boundary feature map of the multi-view image data includes: weighted summing of the red channel values, green channel values, and blue channel values ​​of each pixel in the multi-view image data to obtain a single-channel grayscale image; performing first-order difference operations on the single-channel grayscale image in the horizontal and vertical directions to obtain the gradient magnitude and gradient direction of each pixel; and aggregating the single-channel grayscale image based on the gradient magnitude and gradient direction to obtain the boundary feature map.

4. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 3, characterized in that, The step of performing hierarchical curvature analysis on the boundary feature map to obtain curvature difference features includes: classifying the boundary feature map into levels according to a preset resolution classification rule to obtain resolution level features; extracting continuous boundary points from the boundary feature map in sequence based on the resolution level features; calculating the directional changes between the continuous boundary points and the ratio between the directional changes and the boundary arc length to obtain the local curvature values ​​of the corresponding grid; and subtracting and averaging the local curvature values ​​to obtain curvature difference features.

5. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 1, characterized in that, The step of resampling pixels of the multi-view image data based on the curvature difference features to obtain a curvature-corrected image includes: normalizing the curvature difference features to obtain a normalized curvature difference value; multiplying the normalized curvature difference value by a preset adjustment coefficient to obtain a curvature correction value; writing the grid number of the curvature correction value in the boundary feature map and the curvature correction value into a mapping table to obtain a curvature mapping table; and resampling pixels of the multi-view image data based on the curvature mapping table to obtain a curvature-corrected image.

6. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 1, characterized in that, The step of determining the overlapping view area based on the curvature-corrected image and extracting the local exposure features of the overlapping view area includes: taking each pair of the four original images in the curvature-corrected image as an adjacent view combination, and extracting the stitching boundary position of the adjacent view combination in the curvature-corrected image; calculating the effective coverage width based on the stitching boundary position; if the effective coverage width is greater than a preset coverage threshold, then including the adjacent view combination in the overlapping view area; dividing the overlapping view area according to a preset window size, and statistically analyzing the mean brightness feature, standard deviation brightness feature, and main peak position feature of the grayscale histogram of each window as the local exposure features.

7. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 6, characterized in that, The step of performing brightness equalization on the overlapping region based on the local exposure features to obtain a boundary-optimized image with consistent brightness includes: determining the target brightness value and brightness compensation amount for each window based on the mean brightness feature and the main peak position feature of the grayscale histogram in the local exposure features, and generating a brightness adjustment value for each window in combination with the brightness standard deviation feature; writing the brightness adjustment value into the corresponding window position to obtain an equalized overlapping region; extending a preset pixel width along the stitching boundary in the equalized overlapping region to both sides to form a boundary band region; calculating the brightness difference between adjacent pixels in the boundary band region to obtain a brightness gradient feature, and filtering and merging boundary candidate points according to the brightness gradient feature to obtain a boundary pixel set; assigning corresponding weights to each pixel in the boundary pixel set according to the distance to the center line of the stitching boundary to obtain a pixel weight distribution feature; performing shallow edge correction on the equalized overlapping region based on the pixel weight distribution feature to obtain a gradient-adjusted boundary region; replacing the corresponding region in the equalized overlapping region with the gradient-adjusted boundary region to obtain a boundary-optimized image with consistent brightness.

8. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 1, characterized in that, The step of extracting color distribution feature maps of common areas from the boundary optimization image and forming cross-view texture data based on the color distribution feature maps includes: extracting common areas from the boundary optimization image according to adjacent view combinations; dividing the common areas into windows and statistically analyzing the distribution information of the red, green, and blue channels of each window to obtain a color distribution feature map; determining the color deviation value between each window based on the color distribution feature map; identifying windows with color deviation values ​​less than a preset color threshold as color-corresponding windows; and reading the original pixel arrangement information of the color-corresponding windows to obtain cross-view texture data.

9. The multi-view panoramic image fusion display method based on DA smart interconnected screen according to claim 8, characterized in that, The step of repairing the boundary optimization image based on the cross-view texture data to obtain a fusion preparation image includes: calculating the texture matching degree of each window based on the cross-view texture data; merging windows with texture matching degrees lower than a preset matching threshold according to their positional relationship to obtain fracture region features; extracting neighboring pixels around the fracture region in the fracture region features, and determining corresponding weights based on the distance of the neighboring pixels to the repair position in the fracture region features to generate a repair pixel set; and writing the repair pixel set back to the corresponding region in the boundary optimization image according to the original spatial position of the fracture region features to obtain a fusion preparation image.

10. A multi-view panoramic image fusion display system based on a DA smart interconnected screen, characterized in that, include: The data acquisition module is used to acquire images of the vehicle's surrounding environment; the preliminary correction module is used to perform pixel correction on the images of the vehicle's surrounding environment to obtain preliminary corrected multi-view image data; the curvature analysis module is used to extract the boundary feature map of the multi-view image data and perform hierarchical curvature analysis on the boundary feature map to obtain curvature difference features. The curvature correction module is used to resample the multi-view image data based on the curvature difference features to obtain a curvature-corrected image; the boundary optimization module is used to determine the view overlap region based on the curvature-corrected image, extract the local exposure features of the view overlap region, and perform brightness equalization on the view overlap region based on the local exposure features to obtain a boundary optimization image with consistent brightness; the image fusion module is used to extract the color distribution feature map of the common area from the boundary optimization image, form cross-view texture data based on the color distribution feature map, and repair the boundary optimization image based on the cross-view texture data to obtain a fusion preparation image.

Images