Low-computing-power ultra-high-definition panoramic monitoring image stitching method and device

Abstract

The present invention relates to the technical field of image monitoring, and discloses a low-computing-power ultra-high-definition panoramic monitoring image stitching method and device. The present invention uses a panoramic camera array composed of multiple camera units to collect images of a monitoring area, generates an image time series set of each camera unit, applies a pre-trained motion detection algorithm, analyzes the image detail change feature distribution of each image time series set, determines the image stitching priority according to the preset priority weights of the camera units and the image detail change feature distribution, and performs stitching processing on the monitoring images in each image time series set according to the stitching priority, fuses the data of the overlapping area, and directly calls the image resources of the non-overlapping area, and finally realizes panoramic detail monitoring. The method realizes efficient panoramic image stitching and detail monitoring in a low-computing-power environment, reduces the demand for computing and storage resources, and solves the problem of high equipment cost required for panoramic monitoring in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of image monitoring, and in particular to a low-computing-power ultra-high-definition panoramic monitoring image stitching method and device. Background Art

[0002] With the increasing demand for social security, surveillance systems are widely used in urban security, traffic management, public places and other fields. In order to cover a larger area and obtain more detailed information, panoramic surveillance systems have become an ideal choice. However, traditional high-resolution panoramic surveillance systems have very high requirements for computing and storage resources, which brings many challenges to practical applications.

[0003] In existing panoramic monitoring, the high-definition images obtained by each camera unit need to be stitched together, which requires high-computing power equipment to bear the computing load and storage pressure. However, the production, deployment and maintenance costs of high-computing power equipment are too high. Summary of the invention

[0004] The purpose of the present invention is to provide a low-computing power ultra-high-definition panoramic monitoring image stitching method and device, aiming to solve the problem of high cost of equipment required for panoramic monitoring in the prior art.

[0005] The present invention is implemented in this way. In a first aspect, the present invention provides a low-computing-power ultra-high-definition panoramic monitoring image stitching method, comprising:

[0006] A panoramic image acquisition is performed on the monitoring area through a panoramic camera array composed of a plurality of camera units to obtain a time series set of images corresponding to each camera unit; wherein the time series set of images includes a plurality of monitoring images acquired by the camera units arranged in chronological order, and the monitoring images acquired by the camera units have overlapping areas, and the overlapping areas between the monitoring images are the areas to be spliced;

[0007] Performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm to obtain a feature distribution of image detail changes for each of the image time series sets; wherein the feature distribution of image detail changes includes image detail change features corresponding to each monitoring image in the image time series set;

[0008] According to the distribution of image detail change characteristics corresponding to each monitoring image, performing image stitching priority analysis on each of the image time series sets to obtain the stitching priority of each monitoring image in each of the image time series sets;

[0009] According to the stitching priority of each monitoring image in each of the image time series sets, each monitoring image is stitched in turn to achieve panoramic detail monitoring of the monitoring area; wherein, when each monitoring image is stitched, image data is fused and stitched on the areas to be stitched between each monitoring image, and image resources are directly called on the remaining areas of each monitoring image to achieve image stitching processing of each monitoring image.

[0010] In a second aspect, the present invention provides a low-computing-power ultra-high-definition panoramic image and detail image matching device, which is used to implement a low-computing-power ultra-high-definition panoramic monitoring image stitching method as described in any one of the first aspects, including:

[0011] An image acquisition module, used to acquire panoramic images of the monitoring area through a panoramic camera array composed of a plurality of camera units, so as to obtain a time-series set of images corresponding to each camera unit; wherein the time-series set of images includes a plurality of monitoring images acquired by the camera units and arranged in chronological order;

[0012] A change monitoring module, used for performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm, so as to obtain a distribution of image detail change features of each of the image time series sets; wherein the image detail change feature distribution includes image detail change features corresponding to each monitoring image in the image time series set;

[0013] A splicing analysis module, used to perform image splicing priority analysis on each of the image time series sets according to the distribution of image detail change characteristics corresponding to each of the monitoring images, so as to obtain the splicing priority of each monitoring image in each of the image time series sets;

[0014] The splicing execution module is used to perform image splicing processing on each monitoring image in turn according to the splicing priority of each monitoring image in each image time series set, so as to realize panoramic and detailed monitoring of the monitoring area.

[0015] The present invention provides a low-computing-power ultra-high-definition panoramic monitoring image stitching method, which has the following beneficial effects:

[0016] The present invention uses a panoramic camera array composed of multiple camera units to collect images of the monitoring area, generates an image time series set of each camera unit, applies a pre-trained motion detection algorithm, analyzes the image detail change feature distribution of each image time series set, determines the image stitching priority according to the preset priority weights of the camera units and the image detail change feature distribution, and performs stitching processing on the monitoring images in each image time series set according to the stitching priority, merges the data of the overlapping areas, directly calls the image resources of the non-overlapping areas, and finally realizes panoramic detail monitoring. The method realizes efficient panoramic image stitching and detail monitoring in a low computing power environment, reduces the demand for computing and storage resources, and solves the problem of high equipment cost required for panoramic monitoring in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the steps of a low-computing-power ultra-high-definition panoramic monitoring image stitching method provided by an embodiment of the present invention;

[0018] Figure 2 It is a structural schematic diagram of a low-computing-power ultra-high-definition panoramic image and detail image matching device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0021] Reference Figure 1 , Figure 2 As shown, a preferred embodiment of the present invention is provided.

[0022] In a first aspect, the present invention provides a low-computing-power ultra-high-definition panoramic surveillance image stitching method, comprising:

[0023] S1: A panoramic image of the monitoring area is collected by a panoramic camera array composed of a plurality of camera units to obtain a time series set of images corresponding to each camera unit; wherein the time series set of images includes a plurality of monitoring images collected by the camera units arranged in chronological order, and the monitoring images collected by the camera units have overlapping areas, and the overlapping areas between the monitoring images are areas to be spliced;

[0024] S2: performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm to obtain image detail change feature distribution of each of the image time series sets; wherein the image detail change feature distribution includes image detail change features corresponding to each monitoring image in the image time series set;

[0025] S3: performing image stitching priority analysis on each of the image time series sets according to the image detail change feature distribution corresponding to each of the surveillance images, so as to obtain the stitching priority of each of the surveillance images in each of the image time series sets;

[0026] S4: According to the stitching priority of each surveillance image in each of the image time series sets, each surveillance image is stitched in turn to achieve panoramic detail monitoring of the surveillance area; wherein, when each surveillance image is stitched, image data is fused and stitched on the areas to be stitched between each surveillance image, and image resources are directly called on the remaining areas of each surveillance image to achieve image stitching processing of each surveillance image.

[0027] Specifically, in step S1 of the embodiment provided by the present invention, panoramic images of the monitoring area are captured by a panoramic camera array composed of several camera units to obtain an image time series set corresponding to each camera unit; the image time series set includes monitoring images captured by several camera units arranged in chronological order, and there are overlapping areas between the monitoring images captured by each camera unit, and the overlapping areas between each monitoring image are areas to be spliced.

[0028] It should be noted that the panoramic camera array is a device for capturing panoramic images of a specified area, and the method for capturing panoramic images is to set up multiple camera units, with each camera unit being responsible for a unit area of ​​the specified area, and the various unit areas are superimposed together to achieve panoramic capture of the specified area.

[0029] More specifically, in order to avoid omissions, there are overlapping areas between adjacent unit areas, and these overlapping areas are the boundary areas of the adjacent unit areas.

[0030] More specifically, in traditional technical methods, in order to comprehensively monitor multiple surveillance images captured by a panoramic camera array, all surveillance images need to be spliced ​​to achieve panoramic monitoring of a designated area. It is not difficult to see that in this method, a large amount of computing resources are required because a high-performance CPU or GPU must be used, resulting in a high application cost of the panoramic camera array, including production cost, operating cost, and maintenance cost.

[0031] In one embodiment of the technical solution provided by the present invention, the resolution of each surveillance image is first reduced through a priority-sorted stitching method, and then the change analysis of the downgraded surveillance image is performed to obtain a surveillance image with dynamic changes, which is used as a priority stitching object, and the surveillance images are stitched and detected according to the priority of the surveillance images to be stitched. In this way, there is no need to stitch all the surveillance images at one time, so a lower-performance CPU or GPU can be used to reduce costs.

[0032] Specifically, in step S2 of the embodiment provided by the present invention, the characteristics of image detail changes are monitored for each image time series set respectively according to a pre-trained motion detection algorithm to obtain the image detail change characteristic distribution of each image time series set; wherein, the image detail change characteristic distribution includes the image detail change characteristics corresponding to each monitored image in the image time series set.

[0033] More specifically, the motion detection algorithm first performs a resolution reduction operation on the surveillance images in the image time series set, and then performs feature monitoring of image detail changes on the surveillance images in the reduced resolution form, which can significantly reduce the computing resources required to identify dynamic features.

[0034] More specifically, the image detail change features represent areas where dynamic changes occur in the surveillance image. For these areas, the original high-definition version of the surveillance image needs to be mobilized in the subsequent process for specific detection and image stitching, so as to identify the dynamic changes in the designated area monitored by the panoramic camera array.

[0035] Specifically, in step S3 of the embodiment provided by the present invention, based on the preset priority weights of each camera unit and the distribution of each image detail change characteristic, an image stitching priority analysis is performed on each image time series set to obtain the stitching priority of each monitoring image in each image time series set.

[0036] More specifically, the panoramic camera array points to a designated area, and each camera unit corresponds to a portion of the designated area. The importance of each portion in the designated area varies, so each camera unit has a different preset priority weight.

[0037] More specifically, when performing image stitching on the surveillance images of the camera units, it is necessary to consider the preset priority weights of the various camera units, that is, the surveillance images acquired by the camera units with high preset priority weights are spliced ​​first.

[0038] It should be noted that the purpose of image stitching is that when dynamic changes appear simultaneously in two adjacent surveillance images, the surveillance images need to be stitched to comprehensively identify the dynamic changes. Especially when dynamic changes appear in the overlapping area between two adjacent surveillance images, the stitching processing of the areas to be stitched of the two surveillance images can make the dynamic change identification more accurate.

[0039] It should be noted that the object of image stitching is the surveillance image of the same time node, and the stitching result is used to feedback the specific dynamic changes of each part in the specified area at this time node.

[0040] Specifically, in step S4 of the embodiment provided by the present invention, image stitching processing is performed on each monitoring image in turn according to the stitching priority of each monitoring image in each image time series set to achieve panoramic detail monitoring of the monitoring area; wherein, when each monitoring image is subjected to image stitching processing, image data fusion stitching is performed on the areas to be stitched between each monitoring image, and image resources are directly called for the remaining areas of each monitoring image to achieve image stitching processing of each monitoring image.

[0041] More specifically, when the surveillance images are stitched together, the original high-definition version of the surveillance images is used. That is to say, the dynamic features of the downgraded surveillance images are first collected, and the areas with dynamic features are selected to perform detailed identification of the original high-definition resolution surveillance images to determine the specific content corresponding to the dynamic features. This method can effectively reduce computing resources and thus reduce equipment costs.

[0042] The present invention provides a low-computing-power ultra-high-definition panoramic monitoring image stitching method, which has the following beneficial effects:

[0043] The present invention uses a panoramic camera array composed of multiple camera units to collect images of the monitoring area, generates an image time series set of each camera unit, applies a pre-trained motion detection algorithm, analyzes the image detail change feature distribution of each image time series set, determines the image stitching priority according to the preset priority weights of the camera units and the image detail change feature distribution, and performs stitching processing on the monitoring images in each image time series set according to the stitching priority, merges the data of the overlapping areas, directly calls the image resources of the non-overlapping areas, and finally realizes panoramic detail monitoring. The method realizes efficient panoramic image stitching and detail monitoring in a low computing power environment, reduces the demand for computing and storage resources, and solves the problem of high equipment cost required for panoramic monitoring in the prior art.

[0044] Preferably, the step of collecting panoramic images of the monitoring area by a panoramic camera array composed of a plurality of camera units to obtain a time series set of images corresponding to each camera unit includes:

[0045] S11: setting the panoramic camera array according to the camera performance of the panoramic camera array and the area information of the monitoring area, so that the panoramic camera array is in an initial state of performing camera monitoring on the monitoring area;

[0046] S12: configuring startup parameters for the panoramic camera array in an initial state, so as to activate each camera unit in the panoramic camera array to enter a calibration mode through the startup parameters;

[0047] S13: each camera unit in the calibration mode collects a calibration image of the monitoring area, and performs angle correction and clock synchronization on each camera unit according to the collected calibration image, so as to complete the calibration work of each camera unit and switch to the formal working mode;

[0048] S14: Each camera unit in the formal working mode continuously acquires monitoring images respectively to obtain a number of monitoring images acquired by the camera unit and arranged in chronological order. The number of monitoring images arranged in chronological order together constitute an image time series set of the camera unit; wherein the number of monitoring images in the image time series set continuously increases with time, and each of the monitoring images has a time mark for describing its acquisition time.

[0049] Specifically, the panoramic camera array is reasonably set according to the camera performance of the panoramic camera array (such as resolution, viewing angle range, frame rate, etc.) and the specific information of the monitoring area to ensure that the camera array can cover the entire monitoring area and is in the best monitoring position and angle, and enters the initial state of camera monitoring of the monitoring area.

[0050] More specifically, the startup parameters of the panoramic camera array in the initial state are configured, and these parameters include exposure time, white balance, focal length, etc. After the startup parameter configuration, each camera unit in the panoramic camera array is activated to enter the calibration mode.

[0051] More specifically, each camera unit in the calibration mode starts to collect calibration images of the monitored area, and performs perspective correction on each camera unit based on the collected calibration images to ensure that the perspectives of all camera units are consistent. Then, clock synchronization is performed to ensure that the time stamps of each camera unit are consistent to ensure the synchronization of image timing.

[0052] More specifically, after completing the angle correction and clock synchronization, each camera unit switches to the formal working mode. In the formal working mode, each camera unit begins to continuously collect monitoring images. Each camera unit continuously collects monitoring images to form a series of monitoring images arranged in chronological order. These monitoring images are arranged in chronological order to form an image time series set of the camera unit. The number of monitoring images in the image time series set continues to increase with the passage of time. Each monitoring image is timestamped to describe its acquisition time.

[0053] It can be understood that by reasonably setting the initial state of the panoramic camera array, ensuring that the camera unit can cover the entire monitoring area without blind spots, starting parameter configuration and the use of calibration mode, ensuring the consistency of viewing angle and time of each camera unit, and greatly improving the image stitching quality. In the formal working mode, the camera unit continuously collects monitoring images to ensure the continuity and integrity of the monitoring images. Each monitoring image is time-stamped to ensure the accuracy of the image time series set, which is helpful for subsequent image stitching and analysis. The number of image time series sets continues to increase with time, which can dynamically reflect the real-time situation of the monitoring area and provide complete monitoring records.

[0054] Preferably, the step of performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm to obtain the image detail change feature distribution of each of the image time series sets comprises:

[0055] S21: performing adaptive resolution reduction processing on each of the image time series sets to obtain a low-resolution form of the monitoring image in the image time series set;

[0056] S22: extracting large-amplitude dynamic features from the low-resolution surveillance images to obtain large-amplitude dynamic features of each low-resolution surveillance image, and dividing the surveillance images into regions according to the large-amplitude dynamic features to obtain large-amplitude dynamic regions of the surveillance images and other regions to be measured;

[0057] S23: performing a predetermined standard resolution enhancement on the remaining areas to be tested, so as to enhance the remaining areas to be tested in the monitoring image from a low-resolution form to a medium-resolution form;

[0058] S24: extracting small-amplitude dynamic features from the remaining areas to be tested in the medium-resolution form, so as to obtain small-amplitude dynamic features of the remaining areas to be tested in the monitoring image with the medium-resolution form;

[0059] S25: combining the large-amplitude dynamic features and the small-amplitude dynamic features of each monitoring image in the image time series set to obtain the image detail change feature distribution of each monitoring image in the image time series set.

[0060] Specifically, adaptive resolution reduction processing is performed on the monitoring images in each image time series set to generate low-resolution monitoring images. This step can be completed by selecting an appropriate resolution reduction algorithm to ensure that the main dynamic information is retained while reducing the image resolution.

[0061] More specifically, large-scale dynamic features are extracted from low-resolution surveillance images, and a pre-trained motion detection algorithm is used to identify and extract significant motion change areas in the image. These large-scale dynamic areas usually contain obvious moving objects or drastic changes.

[0062] More specifically, the monitoring image is divided into regions based on the extracted large-amplitude dynamic features, and the image is divided into large-amplitude dynamic regions and remaining regions to be tested. The large-amplitude dynamic regions are those parts containing significant motion, while the remaining regions to be tested are relatively static or little-changing parts.

[0063] More specifically, the remaining areas to be tested are subjected to a predetermined standard of resolution enhancement, which is to enhance these areas from a low-resolution form to a medium-resolution form. This step can be achieved through an interpolation algorithm or other image enhancement techniques, so as to analyze the subtle changes in these areas in more detail.

[0064] More specifically, small-amplitude dynamic features are extracted from the remaining areas to be tested in the medium-resolution form, and a motion detection algorithm is used to identify subtle dynamic changes in these areas, thereby obtaining small-amplitude dynamic features of the remaining areas to be tested in the medium-resolution monitoring image.

[0065] More specifically, the large-amplitude dynamic features and small-amplitude dynamic features of each surveillance image are combined to obtain the image detail change feature distribution. These feature combinations can fully describe the dynamic changes in the surveillance image, including significant motion and subtle changes.

[0066] It can be understood that through adaptive resolution reduction processing, the amount of calculation and resource usage are effectively reduced, so that the motion detection algorithm can run more efficiently. The processing of low-resolution images ensures the accurate extraction of large-scale dynamic features and can quickly identify large-scale significant movements. The combination of region division and resolution enhancement technology can analyze the dynamic features of different regions in detail, so as to more accurately detect subtle changes. The extraction of small-amplitude dynamic features in the form of medium resolution ensures the comprehensive capture of image detail changes and improves the dynamic detection accuracy of monitoring images. The combination of large-amplitude and small-amplitude dynamic features makes the final image detail change feature distribution more comprehensive and accurate, which can better reflect the dynamic changes in monitoring images.

[0067] Preferably, large-amplitude dynamic features are extracted from the low-resolution monitoring image, and small-amplitude dynamic features are extracted from the remaining areas to be measured in the medium-resolution form, and the dynamic feature extraction method adopted is a dynamic background separation extraction method;

[0068] The steps of the dynamic background separation and extraction method include:

[0069] S201: Continuously analyzing the image time series set, establishing a background model according to the analysis result, and updating the background model in real time through an adaptive update mechanism;

[0070] S202: marking the monitoring image to be subjected to dynamic feature extraction as an analysis object, and extracting pixel differences of adjacent frames of the analysis object according to the background model to obtain a set of changed pixels;

[0071] S203: Filtering the set of changed pixels according to a predetermined range and a predetermined time duration to obtain dynamic features;

[0072] The step of filtering the pixel set of the changed pixels according to the predetermined range area and the predetermined time duration to obtain the dynamic feature includes:

[0073] S2031: Identifying pixel position parameters of each changing pixel in the changing pixel set to obtain a discrete feature distribution of the changing pixels; wherein the discrete feature distribution of the changing pixels includes a discrete degree feature between each changing pixel and other pixels, and the discrete degree feature is used to describe the degree of distance between the changing pixels;

[0074] S2032: performing region generation of adjacent pixels for each changing pixel according to the discrete feature distribution of the changing pixels, so as to obtain a dynamic feature composed of each changing pixel within a preset adjacent range, dividing each dynamic feature into a qualified region and an unqualified region according to the area occupied by the dynamic feature, and filtering the changing pixels in the unqualified region;

[0075] S2033: Analyzing the pixel time duration of the changed pixel sets of each monitoring image within the continuous time period to obtain the pixel duration of each changed pixel in the changed pixel set of the monitoring image at the current moment;

[0076] S2034: Perform stability evaluation on the pixel duration of each changing pixel according to a preset duration, divide each changing pixel into stable pixels and flickering pixels according to the result of the stability evaluation, and filter the flickering pixels.

[0077] Specifically, the image time series set is continuously analyzed to establish an initial background model, which is constructed based on the average value or other statistical features of the images over a period of time. The background model is updated in real time through an adaptive update mechanism. This update mechanism can dynamically adjust the background model according to the newly collected image data to adapt to environmental changes, such as changes in lighting and movement of background objects.

[0078] More specifically, the surveillance image to be subjected to dynamic feature extraction is marked as an analysis object, and the pixel differences of adjacent frames of the analysis object are extracted according to the background model. Specifically, by calculating the pixel differences between the current frame and the background model, the changed pixels in the image can be identified, and these changed pixels form a set of changed pixels.

[0079] More specifically, the set of changed pixels is filtered, and the filtering process includes two aspects: according to the set spatial range, only the changed pixels within the predetermined range are retained to exclude noise and irrelevant changes; according to the set time duration, only the pixels that have changed in multiple consecutive frames are retained, and occasional noise pixels are further filtered out. After filtering, a pixel set that truly reflects dynamic characteristics is obtained.

[0080] It can be understood that by establishing and adaptively updating the background model, the system can adapt to environmental changes in real time, improve the accuracy and reliability of dynamic feature extraction, and pixel difference extraction combined with the background model can accurately identify the changing areas in the image. Even at low resolution, it can effectively detect large-scale dynamic features. Through continuous filtering of predetermined range areas and predetermined time, noise and irrelevant changes are effectively eliminated to ensure that the extracted dynamic features have high authenticity and stability. Large-scale dynamic features are extracted in low-resolution form, and small-scale dynamic features are extracted in medium-resolution form. It can take into account large-scale significant movements and subtle changes, provide multi-level dynamic analysis results, and combine different resolutions and dynamic feature extraction methods. The final generated dynamic feature set can fully reflect the detailed changes in the monitoring image and improve the overall monitoring effect.

[0081] More specifically, the position parameters of each changing pixel in the changing pixel set are identified to obtain the spatial distribution of these pixels. The discrete feature distribution describes the distance relationship between changing pixels. By understanding the distribution of changing pixels, it is possible to better identify which pixels are part of the same dynamic feature and improve the accuracy of detection.

[0082] More specifically, based on the discrete feature distribution of the changing pixels, the adjacent areas of the changing pixels (i.e., the dynamic feature areas) are generated, and the dynamic features are divided into qualified areas and unqualified areas according to the size of the area occupied by the dynamic features. Then, the changing pixels in the unqualified areas are filtered out. This step helps to exclude those isolated changing pixels that may be noise and retain pixels with actual dynamic features, thereby improving the accuracy and reliability of detection.

[0083] More specifically, dynamic features smaller than 40 pixels are judged as unqualified areas.

[0084] More specifically, the time continuity analysis is performed on the set of changing pixels in the monitoring images within a continuous time period, and the duration of each changing pixel at the current moment is calculated. By analyzing the time continuity of the pixels, those pixels that appear continuously in multiple frames can be identified, thereby further improving the stability recognition ability of dynamic features.

[0085] More specifically, according to a preset duration threshold, the duration of each changing pixel is evaluated for stability, the changing pixels are divided into stable pixels and flickering pixels, and the flickering pixels are filtered out. The stability evaluation can exclude pixels with a short duration that may be temporary interference, thereby ensuring that the dynamic features finally identified are more stable and reliable.

[0086] More specifically, a changing pixel with a duration of less than 2 milliseconds is determined to be a flickering pixel.

[0087] It can be understood that through these steps, the system can effectively filter out noise and unstable changing pixels, and only retain stable pixels that truly have dynamic features. In this way, the final dynamic features are more accurate and reliable, suitable for various monitoring and image analysis applications, and improve the overall effect and efficiency of image processing.

[0088] Preferably, according to the distribution of image detail change characteristics corresponding to each surveillance image, the step of performing image stitching priority analysis on each of the image time series sets to obtain the stitching priority of each surveillance image in each of the image time series sets includes:

[0089] S31: analyzing the area occupied by the dynamic features of the monitoring image according to the distribution of the image detail change characteristics to obtain a dynamic area parameter in the monitoring image, and converting the dynamic area parameter into a splicing weight to obtain a first splicing weight;

[0090] S32: performing position correlation analysis on the monitoring image between the position of the dynamic feature and the overlapping area between each monitoring image according to the distribution of the image detail change feature, so as to obtain the dynamic cross-region parameter in the monitoring image, and performing splicing weight conversion processing on the dynamic cross-region parameter, so as to obtain the second splicing weight;

[0091] S33: performing weighted fusion analysis on the first splicing weight and the second splicing weight of each monitoring image to obtain the splicing priority of each monitoring image in each image time series set.

[0092] Specifically, by analyzing the distribution of image detail change features, the area of ​​the dynamic region in the surveillance image (i.e., the area occupied by the changing part of the image) is evaluated. The dynamic region area is an indicator to measure the importance of dynamic features in the image. According to the dynamic region area, it is transformed to obtain a "first stitching weight", which is a measure of the priority of the image when stitching. Images with larger dynamic regions may contain more significant change features, and these images should have higher priorities when stitching. Through this step, areas with more dynamic features can be processed first, improving the quality of the stitched image.

[0093] More specifically, according to the distribution of image detail change characteristics, the positional relationship between the dynamic features in the surveillance image and the overlapping areas of other images is analyzed. The dynamic features of the overlapping areas may affect the stitching effect, so they need special processing. According to the dynamic features of the overlapping areas, the "second stitching weight", that is, the dynamic cross-region parameter, is converted. The dynamic features of the overlapping areas help determine the stitching order and ensure that the stitched image has a better transition effect in the overlapping parts. This step can effectively identify those parts that need additional processing during the stitching process to avoid unnatural boundaries during stitching.

[0094] More specifically, the first stitching weight and the second stitching weight are weightedly fused and analyzed to obtain the final stitching priority of each surveillance image. By fusing different weight parameters, the size, position, and cross-region characteristics of the dynamic area can be comprehensively considered to obtain a more reasonable stitching order. After weighted fusion, the stitching priority obtained can balance the dynamic characteristics and overlapping areas of each surveillance image, thereby improving the quality of the stitched image and ensuring natural image transitions and smooth boundaries.

[0095] Preferably, according to the stitching priority of each surveillance image in each of the image time series sets, the steps of sequentially performing image stitching processing on each surveillance image to achieve panoramic detail monitoring of the surveillance area include:

[0096] S41: performing image division of the same time on each monitoring image in each of the image time series sets, so as to divide each monitoring image in each of the image time series sets into a plurality of monitoring image sequences; wherein the monitoring image sequences include the monitoring images in each of the image time series sets at the same time;

[0097] S42: analyzing the splicing execution order of each monitoring image sequence according to the splicing priority of each monitoring image in each image time series set, so as to obtain the splicing execution order of each monitoring image sequence;

[0098] S43: analyzing specific images to be spliced ​​on the surveillance image sequence having the highest priority splicing execution order, so as to determine that a number of surveillance images in the surveillance image sequence are images to be spliced;

[0099] S44: dividing each of the images to be stitched into overlapping areas and non-overlapping areas, marking the overlapping areas in the images to be stitched as areas to be stitched, and marking the non-overlapping areas as direct call areas;

[0100] S45: fusing and stitching the image data of the area to be stitched, and directly calling the image resources of the direct calling area to obtain an image stitching result, thereby realizing image stitching processing of each of the images to be stitched;

[0101] S46: after completing the image stitching process for each of the images to be stitched, taking each of the surveillance images in the surveillance image sequence that has not been stitched as a new surveillance image sequence, and returning to the step of stitching priority of each of the surveillance images in each of the image time series sets;

[0102] S47: Repeat the above steps until all monitoring images are stitched together to achieve panoramic and detailed monitoring of the monitoring area.

[0103] Specifically, the monitoring images in each image time series set are divided into images of the same time, and these images are divided into a number of monitoring image sequences, each sequence containing the monitoring images in the image time series set of the same time.

[0104] More specifically, according to the stitching priority of each surveillance image, the stitching execution order of the surveillance image sequence is analyzed to determine the stitching execution order of each surveillance image sequence, and the surveillance image sequence with the highest priority stitching execution order is analyzed to determine that several surveillance images in the sequence are images to be stitched.

[0105] More specifically, the image to be stitched is divided into overlapping areas and non-overlapping areas, the overlapping areas are marked as areas to be stitched, the non-overlapping areas are marked as direct call areas, the image data of the areas to be stitched are fused and stitched, and the image resources of the direct call areas are directly called to obtain the image stitching result.

[0106] More specifically, after completing the stitching process of the images to be stitched, the surveillance images in the surveillance image sequence that have not been stitched are taken as a new surveillance image sequence, and the process returns to the stitching priority analysis step, and the cycle is repeated until the stitching process of all surveillance images is completed.

[0107] It can be understood that panoramic and detailed monitoring of the monitoring area can be achieved, more comprehensive monitoring images can be provided, and the monitoring effect can be improved. Through the analysis of stitching priority, the stitching order can be optimized, the efficiency of image stitching can be improved, the processing time can be reduced, and the overlapping area and non-overlapping area can be divided and processed to ensure the accuracy and effectiveness of image stitching, avoid repeated processing and waste of resources, save computing resources for the processing of the directly called area, and improve the efficiency of the overall stitching processing. Through the loop mechanism of processing the unstitched image after each stitching and returning to the analysis step, the stitching strategy is dynamically adjusted to ensure the optimization of the final effect. The loop mechanism and the concept of stitching priority of this method make it have good scalability and can adapt to the processing needs of monitoring images of different scales and complexities.

[0108] Reference Figure 2 As shown, in a second aspect, the present invention provides a low-computing-power ultra-high-definition panoramic image and detail image matching device, which is used to implement a low-computing-power ultra-high-definition panoramic monitoring image stitching method as described in any one of the first aspects, including:

[0109] An image acquisition module, used to acquire panoramic images of the monitoring area through a panoramic camera array composed of a plurality of camera units, so as to obtain a time-series set of images corresponding to each camera unit; wherein the time-series set of images includes a plurality of monitoring images acquired by the camera units and arranged in chronological order;

[0110] A change monitoring module, used for performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm, so as to obtain a distribution of image detail change features of each of the image time series sets; wherein the image detail change feature distribution includes image detail change features corresponding to each monitoring image in the image time series set;

[0111] A splicing analysis module, used to perform image splicing priority analysis on each of the image time series sets according to the distribution of image detail change characteristics corresponding to each of the monitoring images, so as to obtain the splicing priority of each monitoring image in each of the image time series sets;

[0112] The splicing execution module is used to perform image splicing processing on each monitoring image in turn according to the splicing priority of each monitoring image in each image time series set, so as to realize panoramic and detailed monitoring of the monitoring area.

[0113] In this embodiment, for the specific implementation of each module in the above-mentioned device embodiment, please refer to the above-mentioned method embodiment, which will not be described in detail here.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A low-computing ultra-high-definition panoramic monitoring image stitching method, characterized in that: include: A panoramic image acquisition is performed on the monitoring area through a panoramic camera array composed of a plurality of camera units to obtain a time series set of images corresponding to each camera unit; wherein the time series set of images includes a plurality of monitoring images acquired by the camera units arranged in chronological order, and the monitoring images acquired by the camera units have overlapping areas, and the overlapping areas between the monitoring images are the areas to be spliced; Performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm to obtain a feature distribution of image detail changes for each of the image time series sets; wherein the feature distribution of image detail changes includes image detail change features corresponding to each monitoring image in the image time series set; According to the distribution of image detail change characteristics corresponding to each monitoring image, performing image stitching priority analysis on each of the image time series sets to obtain the stitching priority of each monitoring image in each of the image time series sets; According to the stitching priority of each monitoring image in each of the image time series sets, each monitoring image is stitched in turn to achieve panoramic detail monitoring of the monitoring area; wherein, when each monitoring image is stitched, image data is fused and stitched on the areas to be stitched between each monitoring image, and image resources are directly called on the remaining areas of each monitoring image to achieve image stitching processing of each monitoring image.

2. The low-computing-power ultra-high-definition panoramic monitoring image stitching method according to claim 1, characterized in that: The steps of collecting panoramic images of the monitoring area by a panoramic camera array composed of a plurality of camera units to obtain a time series set of images corresponding to each camera unit include: The panoramic camera array is set according to the camera performance of the panoramic camera array and the area information of the monitoring area, so that the panoramic camera array is in an initial state of performing camera monitoring on the monitoring area; configuring startup parameters for the panoramic camera array in an initial state, so as to activate each camera unit in the panoramic camera array to enter a calibration mode through the startup parameters; Each camera unit in the calibration mode collects calibration images of the monitoring area, and performs viewing angle correction and clock synchronization on each camera unit according to the collected calibration images, so as to complete the calibration work of each camera unit and switch to the formal working mode; Each camera unit in the formal working mode continuously acquires monitoring images to obtain a number of monitoring images acquired by the camera unit and arranged in chronological order. The number of monitoring images arranged in chronological order together constitute an image time series set of the camera unit; wherein the number of monitoring images in the image time series set continuously increases with time, and each of the monitoring images has a time mark for describing its acquisition time.

3. The low-computing-power ultra-high-definition panoramic monitoring image stitching method according to claim 1, characterized in that: The steps of monitoring the characteristics of image detail changes of each of the image time series sets according to a pre-trained motion detection algorithm to obtain the distribution of image detail change characteristics of each of the image time series sets include: Adaptively reducing the resolution of each of the image time series sets to obtain a low-resolution form of the monitoring image in the image time series set; Extracting large-amplitude dynamic features from low-resolution surveillance images to obtain large-amplitude dynamic features of each low-resolution surveillance image, and dividing the surveillance image into regions according to the large-amplitude dynamic features to obtain large-amplitude dynamic regions of the surveillance image and other regions to be measured; Performing a predetermined standard resolution enhancement on the remaining areas to be tested, so as to enhance the remaining areas to be tested in the monitoring image from a low-resolution form to a medium-resolution form; Extracting small-amplitude dynamic features from the remaining areas to be tested in the medium-resolution form to obtain small-amplitude dynamic features of the remaining areas to be tested in the monitoring image with the medium-resolution; The large-amplitude dynamic features and the small-amplitude dynamic features of each monitoring image in the image time series set are combined to obtain the image detail change feature distribution of each monitoring image in the image time series set.

4. The low-computing-power ultra-high-definition panoramic monitoring image stitching method according to claim 3, characterized in that: Extracting large-scale dynamic features from the low-resolution monitoring image, and extracting small-scale dynamic features from the remaining areas to be tested in the medium-resolution form, the dynamic feature extraction method used is a dynamic background separation extraction method; The steps of the dynamic background separation and extraction method include: Continuously analyzing the image time series set, establishing a background model according to the analysis results, and updating the background model in real time through an adaptive update mechanism; Marking the monitoring image to be subjected to dynamic feature extraction as an analysis object, and extracting pixel differences of adjacent frames of the analysis object according to the background model to obtain a set of changed pixels; Filtering the set of changed pixels according to a predetermined range and a predetermined time duration to obtain dynamic features; The step of filtering the pixel set of the changed pixels according to the predetermined range area and the predetermined time duration to obtain the dynamic feature includes: Identifying pixel position parameters of each changing pixel in the changing pixel set to obtain a discrete feature distribution of the changing pixels; wherein the discrete feature distribution of the changing pixels includes a discrete degree feature between each changing pixel and other pixels, and the discrete degree feature is used to describe the degree of distance between the changing pixels; Generate an area of ​​adjacent pixels for each changing pixel according to the discrete characteristic distribution of the changing pixels to obtain a dynamic feature composed of each changing pixel within a preset adjacent range, divide each dynamic feature into a qualified area and an unqualified area according to the area occupied by the dynamic feature, and filter the changing pixels in the unqualified area; Performing pixel time continuity analysis on the changed pixel sets of each monitoring image within a continuous time period to obtain the pixel duration of each changed pixel in the changed pixel set of the monitoring image at the current moment; The stability of the pixel duration of each changing pixel is evaluated according to the preset duration, each changing pixel is divided into a stable pixel and a flickering pixel according to the result of the stability evaluation, and the flickering pixel is filtered.

5. The low-computing-power ultra-high-definition panoramic monitoring image stitching method according to claim 1, characterized in that: The step of performing image stitching priority analysis on each of the image time series sets according to the image detail change feature distribution corresponding to each of the surveillance images to obtain the stitching priority of each of the surveillance images in each of the image time series sets includes: Analyzing the area occupied by the dynamic features of the monitoring image according to the distribution of changes in the image details to obtain a dynamic area parameter in the monitoring image, and converting the dynamic area parameter into a splicing weight to obtain a first splicing weight; Performing a position correlation analysis on the monitoring image between the position of the dynamic feature and the overlapping area between each monitoring image according to the image detail change distribution to obtain a dynamic cross-region parameter in the monitoring image, and performing a splicing weight conversion process on the dynamic cross-region parameter to obtain a second splicing weight; A weighted fusion analysis is performed on the first splicing weight and the second splicing weight of each monitoring image to obtain the splicing priority of each monitoring image in each image time series set.

6. The low-computing-power ultra-high-definition panoramic monitoring image stitching method according to claim 1, characterized in that: The steps of sequentially performing image stitching processing on each monitoring image according to the stitching priority of each monitoring image in each image time series set to achieve panoramic detail monitoring of the monitoring area include: Performing image division of the same time on each monitoring image in each of the image time series sets, so as to divide each monitoring image in each of the image time series sets into a plurality of monitoring image sequences; wherein the monitoring image sequences include the monitoring images in each of the image time series sets at the same time; According to the splicing priority of each monitoring image in each image time series set, analyzing the splicing execution order of each monitoring image sequence to obtain the splicing execution order of each monitoring image sequence; Analyzing specific images to be spliced ​​on a surveillance image sequence having the highest priority splicing execution order to determine that a number of surveillance images in the surveillance image sequence are images to be spliced; Dividing each of the images to be stitched into overlapping areas and non-overlapping areas, marking the overlapping areas in the images to be stitched as areas to be stitched, and marking the non-overlapping areas as direct call areas; Performing fusion stitching of image data on the area to be stitched, and directly calling image resources on the direct calling area to obtain an image stitching result, thereby realizing image stitching processing on each of the images to be stitched; After completing the image stitching process for each of the images to be stitched, taking each of the surveillance images in the surveillance image sequence that has not been stitched as a new surveillance image sequence, and returning to the step of stitching priority of each of the surveillance images in each of the image time series sets; The above steps are repeated in a cycle until all monitoring images are stitched together to achieve panoramic and detailed monitoring of the monitoring area.

7. A low-computing ultra-high-definition panoramic image and detail image matching device, characterized in that: A low-computing-power ultra-high-definition panoramic monitoring image stitching method for implementing any one of claims 1 to 6 comprises: An image acquisition module, used to acquire panoramic images of the monitoring area through a panoramic camera array composed of a plurality of camera units, so as to obtain a time-series set of images corresponding to each camera unit; wherein the time-series set of images includes a plurality of monitoring images acquired by the camera units and arranged in chronological order; A change monitoring module, used for performing feature monitoring of image detail changes on each of the image time series sets according to a pre-trained motion detection algorithm, so as to obtain a distribution of image detail change features of each of the image time series sets; wherein the image detail change feature distribution includes image detail change features corresponding to each monitoring image in the image time series set; A splicing analysis module, used to perform image splicing priority analysis on each of the image time series sets according to the distribution of image detail change characteristics corresponding to each of the monitoring images, so as to obtain the splicing priority of each monitoring image in each of the image time series sets; The splicing execution module is used to perform image splicing processing on each monitoring image in turn according to the splicing priority of each monitoring image in each image time series set, so as to realize panoramic and detailed monitoring of the monitoring area.

Images