High-resolution image generation method and high-speed camera integrating the method

By using subpixel motion information and optical flow characteristics to fusion images on high-speed cameras, the problems of image quality reduction and equipment diversity during high-resolution image generation in the prior art are solved, and the efficient generation of high-resolution images under high-speed conditions is achieved. Effect.

CN111932453BActive Publication Date: 2024-11-01HEFEI JUNDA HI TECH INFORMATION TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010700086.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-07-20
Publication Date
2024-11-01
Estimated Expiration
2040-07-20

AI Technical Summary

Technical Problem

The prior art often results in degradation of image quality or requires multiple cameras or multiple acquisitions when acquiring high-resolution images, and the image super-resolution reconstruction method based on deep learning models cannot be applied to high-speed cameras for high-speed processing.

Method used

By acquiring real-time image frames on a high-speed camera, using sub-pixel motion information to fuse, high-resolution images are generated, and motion estimation method based on optical flow characteristics is used and iterative reverse projection method is used to synthesize images in combination with regularization method, and in Adjust camera parameters if necessary to increase the acquisition frame rate to suit the moving target.

Benefits of technology

It realizes the generation of high-resolution images without reducing image quality, reduces image acquisition time, avoids the need for multiple cameras or multiple shots, and forms a high-speed camera that directly outputs high-resolution images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111932453B_ABST
    Figure CN111932453B_ABST
Patent Text Reader

Abstract

The present invention discloses a high-resolution image generation method and a high-speed camera integrating the method, comprising obtaining a group of synthetic base images and the motion displacement of each pair of image frames in the synthetic base images, judging whether the motion displacement of any two image frames is less than a pixel unit, and if so, fusing the image frames in the group of synthetic base images to obtain a high-resolution image, otherwise obtaining a high-resolution image corresponding to the group of synthetic base images by a backup method; performing a quality inspection on the output high-resolution image, and obtaining the next group of synthetic base images to continue super-resolution image synthesis if the quality is qualified. The present invention obtains sub-pixel motion information of a target image based on the high-speed acquisition capability of a high-speed camera, fills the sub-pixel motion information into the synthesized high-resolution image, improves the pixel information of the corresponding position in the high-resolution image, and realizes the generation of a high-resolution image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a high-resolution image generation method and a high-speed camera integrating the method. Background Art

[0002] In the prior art, high-resolution images can be obtained by upsampling the original image or by stitching images after multi-position acquisition. However, the former will reduce the quality of the enlarged image, and the latter requires multiple cameras or multiple acquisitions and accurate camera arrangement; both are post-image processing after acquisition, which requires additional processing time. As shown in the figure below, the upsampling method leads to reduced image quality, and the image stitching method requires more equipment or operation processing time.

[0003] Other image super-resolution reconstruction methods based on deep learning models cannot be applied to images of different scenes or to high-speed cameras that require high-speed processing because they require training data sets that match the target and a large amount of computing space. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a high-speed acquisition capability based on a high-speed camera to obtain sub-pixel motion information of the target image, which is used to improve the pixel information of the corresponding position in the high-resolution image, and synthesize a high-resolution image without reducing the image quality.

[0005] Specifically, the high-resolution image generation method of the present invention is applied to a high-speed camera and comprises the following steps:

[0006] (1) obtaining image frames collected in real time, and taking every n image frames as a group of synthetic base images, where n is a preset synthetic base number;

[0007] (2) Obtain the motion displacement of each pair of image frames in a set of synthetic base images, and determine whether the motion displacement of any two image frames is less than one pixel unit. If so, proceed to step (4); otherwise, proceed to step (3);

[0008] (3) issuing a warning message that the acquisition frame rate needs to be increased to obtain sub-pixel displacement features, and obtaining a high-resolution image corresponding to a set of synthetic base images by an alternative method;

[0009] (4) A high-resolution image is obtained by fusing sub-pixel displacement information between image frames in a set of synthetic base images;

[0010] (5) Performing a quality check on the output high-resolution image; if the quality is unsatisfactory, adjusting the camera parameters, re-capturing the image frames, and returning to step (1); if the quality is satisfactory, returning to step (1) to obtain the next set of synthetic base images.

[0011] As a further optimization of the above scheme, the step (1) is specifically to determine whether the total number of image frames that have been acquired has reached n based on the currently acquired image frame, and when the total number has reached n, the current image frame and the previous n-1 image frames are used as a group of synthetic base images.

[0012] As a further optimization of the above solution, in step (2), the motion displacement of each pair of image frames in a set of synthetic base images is obtained by using a motion estimation method based on optical flow features.

[0013] As a further optimization of the above solution, the motion estimation method based on optical flow features is a Gaussian pyramid LK optical flow method.

[0014] As a further optimization of the above scheme, the step (4) uses an iterative back-projection method and a regularization method to fuse the sub-pixel displacement information between image frames to obtain a high-resolution image. The specific steps include:

[0015] An image frame at a preset position is used as a reference frame, and affine transformation parameters are obtained based on sub-pixel displacement information between the non-reference frame and the reference frame;

[0016] The reference frame is interpolated with a preset super-resolution reconstruction multiple to obtain an initial high-resolution image x, and a simulated low-resolution image frame yi of the initial high-resolution image x is obtained according to the degradation model, where i is the number of low-resolution image frames:

[0017] A regularization term is introduced to calculate the error between the simulated low-resolution image frame yi and the corresponding acquired image frame;

[0018] The error is back-projected onto the high-resolution image x, and the error is iteratively calculated to correct the high-resolution image x until the error value converges to meet the preset conditions, thus obtaining the high-resolution image x.

[0019] As a further optimization of the above solution, the preset super-resolution reconstruction factor is less than or equal to a preset synthesis base n.

[0020] As a further optimization of the above solution, the alternative method in step (3) adopts high-resolution image fitting of the acquired image frames based on an interpolation method.

[0021] As a further optimization of the above solution, the interpolation is based on judging the edge area and non-edge area of ​​the image frame, and different interpolation methods are used to interpolate different areas.

[0022] High-speed cameras based on the above high-resolution image generation method include:

[0023] A synthetic base image acquisition module is used to acquire image frames collected in real time, and take every n image frames as a group of synthetic base images, where n is a preset synthetic base number;

[0024] The super-resolution image synthesis condition judgment module obtains the motion displacement of each image frame in a set of synthetic base images, and judges whether the motion displacement of any two image frames is less than one pixel unit. If so, it enters the main super-resolution synthesis module, otherwise it enters the backup super-resolution synthesis module;

[0025] A backup super-resolution synthesis module is used to issue a warning prompt message that the acquisition frame rate needs to be increased to obtain sub-pixel displacement features, and to obtain a high-resolution image corresponding to a set of synthetic base images by a backup method;

[0026] The main super-resolution synthesis module is used to obtain a high-resolution image by fusing the sub-pixel displacement information between image frames in a set of synthetic base images;

[0027] The synthesis result feedback module is used to check the quality of the output high-resolution image. If the quality is unqualified, the camera parameters are adjusted to re-acquire the image frame and return it to the synthesis base image acquisition module. If the quality is qualified, it returns to the synthesis base image acquisition module to obtain the next set of synthesis base images.

[0028] As a further optimization of the above scheme, the super-resolution image synthesis condition judgment module includes an optical flow method sub-pixel motion estimation unit, which is used to obtain the motion displacement of two image frames in a set of synthetic base images using a motion estimation method based on optical flow features. The high-resolution image generation method of the present invention and the high-speed camera integrating the method have the following beneficial effects:

[0029] 1. The present invention is based on the high-speed acquisition capability of a high-speed camera, obtains sub-pixel motion information of a target image, fills the sub-pixel motion information into a synthesized high-resolution image, improves the pixel information of a corresponding position in the high-resolution image, and realizes the generation of a high-resolution image, thereby avoiding the need to use multiple cameras or multiple shooting scenes and the cost of post-processing, reducing the time required to obtain the final image, and forming a high-speed camera that directly outputs a high-resolution image.

[0030] 2. The present invention achieves generation of high-resolution images without reducing image quality by adjusting the parameters of the camera, adjusting the super-resolution synthesis reconstruction multiple to be less than or equal to the synthesis base n, and setting the high-speed camera acquisition frame rate to be adapted to the moving speed of the moving target. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flowchart of a method for generating a high-resolution image according to the present invention;

[0032] Figure 2 It is a flowchart of a method for obtaining the motion displacement of two image frames in the high-resolution image generation method of the present invention;

[0033] Figure 3 The structural block diagram of the high-speed camera integrated with the high-resolution image generation method of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further described below in conjunction with specific embodiments and drawings.

[0035] The present invention provides a high-resolution image generation method, which is applied to a high-speed camera and comprises the following steps:

[0036] First, obtain the user's setting data for high-speed camera parameters, including the camera's acquisition frame rate, acquisition resolution, synthesis base, super-resolution reconstruction magnification, exposure time, and acquisition output frame number, among which the super-resolution reconstruction magnification is less than or equal to the synthesis base value.

[0037] The super-resolution reconstruction magnification is less than or equal to the synthetic base value, so that sufficient sub-pixel position points can be provided under the reconstruction magnification requirement to ensure the high quality of the high-resolution image after super-resolution reconstruction.

[0038] (1) Based on the camera parameters, image frames are acquired in real time, and every n image frames are used as a group of synthetic base images, where n is a preset synthetic base number. Preferably, based on the currently acquired image frame, it is determined whether the total number of acquired image frames has reached n. When it reaches n, the current image frame and the previous n-1 image frames are used as a group of synthetic base images; that is, the first frame to the nth frame are used as a group of synthetic base images, and the second frame to the n+1th frame are used as the next group of synthetic base images, and so on, to ensure the continuity of the synthetic base, so that the processed image sequence maintains the original acquisition frame rate output, thereby ensuring the consistency of the camera parameters and the continuity of the output video image.

[0039] (2 In this embodiment, a high-speed camera is used to capture image frames. The time interval between adjacent image frames of a group of synthetic base images is very short, and the difference between the images is small. Therefore, a sub-pixel motion estimation method based on optical flow features is used to obtain the motion displacement of two image frames in a group of synthetic base images. Preferably, a Gaussian pyramid LK optical flow method is used to obtain the motion displacement of two image frames in a group of synthetic base images. The steps include:

[0040] Perform multi-level downsampling on the two image frames to obtain motion displacement to obtain multiple upper-layer reduced image frames, form two pyramids, and perform motion estimation on the image frames of the same layer of the two pyramids;

[0041] First, start from the top of the pyramid:

[0042] The first constraint formula of sub-pixel displacement is obtained under the condition that the brightness I(x, y, t) of the same pixel (x, y) in two image frames on the same layer is constant:

[0043] I x u+I y v+I t =0 (1);

[0044] The first constraint condition is improved, and the second constraint formula of sub-pixel displacement is obtained by taking the constant brightness gradient of the same pixel (x, y) in two image frames as the second constraint condition:

[0045]

[0046] The above equation (2) is solved in a preset size window, and the equation is obtained under the condition that the motion velocity vectors of all pixels in the small spatial area are constant:

[0047]

[0048] Take W(x) as the rectangular window function,

[0049] Let W=diag[w(X1-X0),w(X2-X0),...,w(X N -X0)], solve equation (3) to obtain the motion displacement vector parameters in, A T WA is a non-singular matrix;

[0050] Then, the motion displacement vector parameters of the top layer are used as the initial values ​​of the next layer, and the motion displacement vector parameters of the two pixel frames of the next layer are calculated based on the motion estimation method of the two pixel frames of the top layer. The above steps are repeated until the motion displacement vector parameters of the two pixel frames of the bottom layer are obtained, which are the motion displacement between the two acquired image frames.

[0051] Since the high-resolution image is acquired based on multiple continuous image frames in this embodiment, the common area of ​​the images in the same group is not cropped, and only the super-resolution reconstruction of the target is performed. Therefore, if the movement speed of the target is too fast and the movement of the target exceeds one pixel between adjacent frames, super-resolution reconstruction of the target under the same background cannot be achieved.

[0052] Therefore, it is necessary to judge based on the motion displacement between any two acquired image frames whether the motion displacement of any two image frames is less than one pixel unit. If so, it means that in a set of synthetic base images, the motion displacement of the moving target is less than one pixel unit, and the process goes to step (4) to obtain high-resolution information of each pixel based on the different sub-pixel displacement information in the same pixel point on multiple image frames. Otherwise, in a set of synthetic base images, the motion displacement of the moving target is greater than or equal to one pixel, and the sub-pixel displacement information between the image frames cannot be used for super-resolution reconstruction. At this time, the process goes to step (3) to adopt an alternative method, that is, to obtain a high-resolution image based on an interpolation method.

[0053] (3) Specifically, when the motion displacement of the moving target in a set of synthetic base images is greater than or equal to one pixel, an alarm prompt message is issued to increase the acquisition frame rate to obtain sub-pixel displacement features. At this time, it indicates that the moving speed of the moving target makes the displacement that occurs within the acquisition time of a set of synthetic base images exceed one pixel unit, so the camera acquisition frame rate needs to be increased to adapt to the increased motion speed of the moving target. Specifically, assuming that the acquisition frame rate parameter of the camera is MFPS, the projection speed of the target motion in the image is V pixel / s, and the number of synthetic bases per group is n, it is required that V≤M / n;

[0054] After the alarm prompt information is issued, a high-resolution image corresponding to a set of synthetic base images is obtained by an alternative method; the alternative method uses an interpolation method to perform high-resolution image fitting of the acquired image frame. The interpolation adopts different interpolation methods based on the judgment of the edge area and non-edge area of ​​the image frame to interpolate different areas. The specific steps include:

[0055] Using the image frame at the preset position as a reference frame;

[0056] For the pixel to be interpolated on the reference frame, obtain the adjacent low-resolution pixels in 8 directions of the pixel to be interpolated, calculate the standard deviation between the pixel to be interpolated and the adjacent low-resolution pixels in the 8 directions, and take the adjacent low-resolution pixels whose difference between the pixel to be interpolated and the adjacent low-resolution pixels is less than the standard deviation as edge pixels, and take the adjacent low-resolution pixels whose difference between the pixel to be interpolated and the adjacent low-resolution pixels is greater than the standard deviation as non-edge pixels;

[0057] For the pixels in the edge pixel area, the grayscale values ​​of the adjacent low-resolution pixels in the four directions of the pixel to be interpolated are obtained, and interpolation is performed in the direction where the difference between the grayscale values ​​of the pixel to be interpolated and the adjacent low-resolution pixels is the smallest. The interpolated grayscale value is the weighted sum of the grayscale values ​​of the pixel to be interpolated and the adjacent low-resolution pixels;

[0058] For pixels in non-edge pixel areas, interpolation is performed using a bilinear interpolation algorithm.

[0059] (4) When the motion displacement of the moving object in a set of synthetic base images is less than one pixel, a high-resolution image is obtained by fusing the sub-pixel displacement information between image frames in the set of synthetic base images;

[0060] Preferably, an iterative back-projection method and a regularization method are used to fuse sub-pixel displacement information between image frames to obtain a high-resolution image, and the specific steps include:

[0061] Based on the motion displacement estimation results between the two image frames obtained above, an image frame at a preset position is used as a reference frame, and affine transformation parameters are obtained based on sub-pixel displacement information between the non-reference frame and the reference frame;

[0062] The reference frame is interpolated with a preset super-resolution reconstruction multiple to obtain the initial high-resolution image x, and the simulated low-resolution image frame yi of the initial high-resolution image x is obtained according to the degradation model, where i is the number of low-resolution image frames:

[0063] y i =DB i M i x+n i , where D is the downsampling matrix, B i is the fuzzy matrix, Mi is the affine transformation matrix between the reference frame and the non-reference frame, n i is the noise vector matrix;

[0064] The regularization term is introduced to calculate the error between the simulated low-resolution image frame yi and the corresponding acquired image frame:

[0065]

[0066] in, is the regularization term, x is the super-resolution reconstructed image, D is the downsampling matrix, B i is the fuzzy matrix, Mi is the affine transformation matrix between the reference frame and the non-reference frame, n i is the noise vector matrix, λ is the regularization parameter, Dx in Λ(x) represents the brightness gradient of the image, ||.||1 is the 1-norm, N i is the number of image components; among them, there are three common blur matrices: Gaussian blur function, linear motion blur function and defocus blur function.

[0067] The error is back-projected onto the high-resolution image x, and the error is iteratively calculated to correct the high-resolution image x until the error value converges to meet the preset conditions, thus obtaining the high-resolution image x.

[0068] (5) Performing a quality check on the output high-resolution image; if the quality is unsatisfactory, adjusting the camera parameters, re-capturing the image frames, and returning to step (1); if the quality is satisfactory, returning to step (1) to obtain the next set of synthetic base images.

[0069] In this step, the quality of the high-resolution image is inspected. Common methods include subjective method and objective method. In this embodiment, the objective method is selected to measure the quality of the high-resolution image by the peak signal-to-noise ratio index of the high-resolution image.

[0070] If the quality is not up to standard, the camera parameters are readjusted, including the acquisition frame rate, synthesis base or magnification, where the magnification is not allowed to be greater than the synthesis base. The image frame is acquired again and the image frame counting is started. When the number of acquired frames reaches the preset synthesis count n, the current image frame and the previous n-1 image frames are used as a group of synthesis base images to obtain a high-resolution image. If the quality is qualified, the next group of synthesis base images is acquired.

[0071] Based on the above-mentioned high-resolution image generation method, the present invention also provides a high-speed camera integrating the method. Based on the high-speed acquisition capability of the high-speed camera, it is possible to perform multiple acquisitions of a moving target in a very short time, capture the sub-pixel motion information of the target, and simultaneously meet the output requirements of high resolution and high image quality, avoiding the need to use multiple cameras or multiple shooting scene requirements and post-processing costs, reducing the time to obtain the final image, and forming a high-speed camera that directly outputs high-resolution images.

[0072] At the same time, when the acquisition frame rate is too low and the target movement speed is too fast, resulting in the inability to obtain the sub-pixel motion information of the target, a backup method is set. Users can choose to use the backup method to obtain high-resolution images, and then consider whether to change the acquisition frame rate parameters after seeing the processing effect, and re-acquire image frames for super-resolution image synthesis.

[0073] Specifically, the high-speed camera includes:

[0074] A synthetic base image acquisition module is used to acquire image frames collected in real time, and take every n image frames as a group of synthetic base images, where n is a preset synthetic base number;

[0075] The super-resolution image synthesis condition judgment module obtains the motion displacement of each image frame in a set of synthetic base images, and judges whether the motion displacement of any two image frames is less than one pixel unit. If so, it enters the main super-resolution synthesis module, otherwise it enters the backup super-resolution synthesis module;

[0076] A backup super-resolution synthesis module is used to issue a warning prompt message that the acquisition frame rate needs to be increased to obtain sub-pixel displacement features, and to obtain a high-resolution image corresponding to a set of synthetic base images by a backup method;

[0077] The main super-resolution synthesis module is used to obtain a high-resolution image by fusing the sub-pixel displacement information between image frames in a set of synthetic base images;

[0078] The synthesis result feedback module is used to check the quality of the output high-resolution image. If the quality is unqualified, the camera parameters are adjusted to re-acquire the image frame and return it to the synthesis base image acquisition module. If the quality is qualified, it returns to the synthesis base image acquisition module to obtain the next set of synthesis base images.

[0079] The super-resolution image synthesis condition judgment module mentioned above includes an optical flow method sub-pixel motion estimation unit, which is used to obtain the motion displacement of each pair of image frames in a set of synthetic base images by using a motion estimation method based on optical flow features.

[0080] The present invention is not limited to the above-mentioned specific implementation modes. Various changes made by ordinary technicians in this field based on the above-mentioned concepts without creative work are all within the protection scope of the present invention.

Claims

1. A high-resolution image generation method, applied to a high-speed camera, characterized in that: The steps include: (1) Obtaining the setting data of high-speed camera parameters, including the camera's acquisition frame rate, acquisition resolution, synthesis base, super-resolution reconstruction magnification, exposure time, and acquisition output frame number, where the super-resolution reconstruction magnification is less than or equal to the synthesis base value; obtaining real-time acquired image frames, taking every n image frames as a group of synthesis base images, where n is a preset synthesis base; (2) Obtain the motion displacement of each pair of image frames in a set of synthetic base images, and determine whether the motion displacement of any two image frames is less than one pixel unit. If so, proceed to step (4); otherwise, proceed to step (3); (3) issuing an alarm prompt message that the acquisition frame rate needs to be increased to obtain sub-pixel displacement features, and obtaining a high-resolution image corresponding to a set of synthetic base images by an alternative method, wherein the increased acquisition frame rate satisfies: the acquisition frame rate parameter of the camera is M FPS, the projection speed of the target motion in the image is V pixel / s, then V ≤ M / n; (4) Obtain a high-resolution image based on the sub-pixel displacement information between image frames in a set of synthetic base images; (5) Performing a quality check on the output high-resolution image. If the quality is unqualified, adjusting the camera parameters, re-capturing the image frames, and returning to step (1). If the quality is qualified, returning to step (1) to obtain the next set of synthetic base images. The camera parameters adjusted include the acquisition frame rate, the synthetic base, or the magnification, wherein the magnification shall not be greater than the synthetic base.

2. A high-resolution image generation method according to claim 1, characterized in that: Specifically, the step (1) comprises: determining, based on the currently acquired image frame, whether the total number of acquired image frames reaches n; when the total number reaches n, taking the current image frame and the previous n-1 image frames as a group of synthetic base images.

3. A high-resolution image generation method according to claim 1, characterized in that: In the step (2), the motion displacement of each pair of image frames in a set of synthetic base images is obtained by using a motion estimation method based on optical flow features.

4. A high-resolution image generation method according to claim 3, characterized in that: The motion estimation method based on optical flow features is Gaussian pyramid LK optical flow method.

5. The high-resolution image generation method according to claim 1, characterized in that: The step (4) uses an iterative back-projection method and a regularization method to fuse the sub-pixel displacement information between image frames to obtain a high-resolution image. The specific steps include: An image frame at a preset position is used as a reference frame, and affine transformation parameters are obtained based on sub-pixel displacement information between the non-reference frame and the reference frame; The reference frame is interpolated with a preset super-resolution reconstruction multiple to obtain the initial high-resolution image x, and the simulated low-resolution image frame y of the initial high-resolution image x is obtained according to the degradation model i ,i is the number of low-resolution image frames; Introduce the regularization term to calculate the simulated low-resolution image frame y i and the error between the corresponding acquired image frame; The error is back-projected onto the high-resolution image x, and the error is iteratively calculated to correct the high-resolution image x until the error value converges to meet the preset conditions, thus obtaining the high-resolution image x.

6. A high-resolution image generation method according to claim 1, characterized in that: The alternative method in step (3) uses an interpolation method to perform high-resolution image fitting of the acquired image frames.

7. A high-resolution image generation method according to claim 6, characterized in that: The interpolation is based on judging the edge area and the non-edge area of ​​the image frame, and using different interpolation methods to interpolate different areas.

8. A high-speed camera based on the high-resolution image generation method according to any one of claims 1 to 7, characterized in that: include: The synthetic base image acquisition module is used to obtain the setting data of the high-speed camera parameters, including the camera's acquisition frame rate, acquisition resolution, synthetic base, super-resolution reconstruction magnification, exposure time and acquisition output frame number, where the super-resolution reconstruction magnification is less than or equal to the synthetic base value; it is also used to obtain the image frames acquired in real time, taking every n image frames as a group of synthetic base images, where n is a preset synthetic base; The super-resolution image synthesis condition judgment module obtains the motion displacement of each image frame in a set of synthetic base images, and judges whether the motion displacement of any two image frames is less than one pixel unit. If so, it enters the main super-resolution synthesis module, otherwise it enters the backup super-resolution synthesis module; The standby super-resolution synthesis module is used to issue a warning prompt message that the acquisition frame rate needs to be increased to obtain sub-pixel displacement features, and to obtain a high-resolution image corresponding to a set of synthetic base images by a standby method, and the increased acquisition frame rate satisfies: the acquisition frame rate parameter of the camera is M FPS, the projection speed of the target movement in the image is V pixel / s, then V≤M / n; The main super-resolution synthesis module is used to obtain a high-resolution image by fusing the sub-pixel displacement information between image frames in a set of synthetic base images; The synthesis result feedback module is used to check the quality of the output high-resolution image. If the quality is unqualified, the camera parameters are adjusted to re-acquire the image frame and return it to the synthesis base image acquisition module. If the quality is qualified, it returns to the synthesis base image acquisition module to obtain the next set of synthesis base images. The adjusted camera parameters include the acquisition frame rate, synthesis base or magnification, and the magnification shall not be greater than the synthesis base.

9. The high-speed camera according to claim 8, characterized in that: The super-resolution image synthesis condition judgment module includes an optical flow method sub-pixel motion estimation unit, which is used to obtain the motion displacement of two image frames in a group of synthetic base images by using a motion estimation method based on optical flow features.