Adaptive adjustment method and system for resolution of monitoring image

Abstract

The invention relates to the field of resolution adjustment, in particular to a monitoring image resolution self-adaptive adjustment method and a monitoring image resolution self-adaptive adjustment system. The method comprises the following steps: acquiring a preprocessed monitoring image according to a preset sampling interval; calculating a difference image of the reference image and the target image, and calculating a fluctuation area of the target image according to the difference image; performing connected domain extraction on the difference image to obtain an extraction result, and calculating the concentration degree of the target image according to the extraction result; and inputting the fluctuation area and the concentration degree of the target image into a preset network, outputting the adaptive resolution of the target image, and completing resolution adaptive adjustment. Through the technical scheme of the invention, the accuracy of a resolution self-adaption result can be improved.

Description

Technical Field

[0001] The present invention relates to the field of resolution adjustment. More specifically, the present invention relates to a method and system for adaptively adjusting the resolution of monitoring images. Background Art

[0002] With the continuous development of information technology, monitoring systems are increasingly used in various fields. From urban security monitoring to enterprise internal environment management and then to the public security field, monitoring images have become an indispensable part of daily life. The quality and resolution of monitoring images are one of the key factors for evaluating the performance of monitoring systems, affecting the clarity, recognition accuracy, and post-processing efficiency of images. Therefore, how to improve the resolution of monitoring images and make it adaptively adjusted according to the actual situation has become an important topic in current technical research and applications.

[0003] The Chinese patent application document with the publication number CN104125419A discloses an adaptive resolution implementation method based on a CMOS image sensor. Based on image processing technology, it determines whether there are dynamic targets in the image output by the CMOS image sensor. When there are dynamic targets in the field of view, the resolution of the CMOS image sensor is adjusted to the best according to the image recognition requirements; when there are no dynamic targets in the field of view, the resolution of the CMOS image sensor is automatically reduced.

[0004] In target recognition, the recognition accuracy varies with the change of resolution. The number of pixel points in the image is different at different resolutions, so the number of pixel points required to achieve the same recognition accuracy is also different. However, the above application document only considers the number of moving pixel points and does not consider the distribution of moving pixel points, resulting in inaccurate resolution adaptation results. Summary of the Invention

[0005] To solve the problem of inaccurate resolution adaptation results, the present invention proposes a method and system for adaptively adjusting the resolution of monitoring images.

[0006] In a first aspect, the present invention discloses a method for adaptively adjusting the resolution of monitoring images, including: obtaining a preprocessed monitoring image at a preset sampling interval; taking the monitoring image at any sampling moment as the target image, taking the monitoring image at the previous sampling moment adjacent to the target image as the reference image, calculating the difference image between the reference image and the target image, and calculating the fluctuation area of the target image according to the difference image; performing connected component extraction on the difference image to obtain an extraction result, and calculating the concentration degree of the target image according to the extraction result; inputting the fluctuation area and concentration degree of the target image into a preset network, and outputting the adapted resolution of the target image to complete the resolution adaptive adjustment.

[0007] Optimizes the image processing efficiency and resource utilization, while ensuring the monitoring effect and target recognition accuracy in different scenarios.

[0008] Preferably, the number of non-zero pixel points in the difference image is used as the fluctuation area of the target image.

[0009] Can effectively capture the dynamic changes in the monitoring scene, which helps to accurately identify the movement of objects or other potential events. When the fluctuation area is large, it indicates that the image has changed significantly, and the resolution can be adjusted accordingly for more precise monitoring; while when the fluctuation area is small, a lower resolution can be maintained, thus saving system resources.

[0010] Preferably, the fluctuation area also satisfies the relational expression: , represents the fluctuation area, represents the adapted resolution of the reference image, represents the th row and th column pixel value in the difference image, represents the vector function, represents the natural constant.

[0011] Can dynamically adjust the resource usage of the monitoring system, accurately adjust the resolution according to the actual change amount, thus avoiding unnecessary calculation and storage burdens while improving the monitoring clarity, maximizing the system's efficiency and resource utilization rate, and ensuring that the real-time monitoring process can both quickly respond to changes and effectively manage the system load.

[0012] Preferably, calculating the concentration degree of the target image according to the extraction result includes: clustering the positions of the connected components to obtain several clustering clusters; the concentration degree satisfies the relational expression: , represents the concentration degree, represents the clustering cluster and the sum of the areas of all connected components in it, represents the sum of the areas of all connected components in the difference image, represents the logarithmic function, represents the natural constant.

[0013] Can effectively identify whether the changed areas in the image are concentrated. If the concentrated changed areas are small, the system can more easily identify the target, so there is no need to overly increase the resolution; while when the changed areas are scattered, a more precise resolution adjustment is required to ensure that all targets can be identified.

[0014] Preferably, calculating the concentration degree of the target image based on the extraction result further includes: dividing the difference image into grids, and calculating the dependence matrix of any grid; regarding the grid containing the connected domain as a special grid, and taking the number of special grids in the 8-neighborhood direction of any grid as the dependence value; the concentration degree satisfies the relational expression: , represents the concentration degree, represents the grid dependence value.

[0015] By carefully dividing the image area and considering the mutual dependence between neighborhoods, the calculation of the concentration degree becomes more accurate. If the distribution of the fluctuation area is relatively concentrated, the dependence value is higher, and the concentration degree is correspondingly larger; otherwise, the concentration degree is lower.

[0016] Preferably, the training process of the preset network is as follows: taking the historical monitoring image as the input information, and taking the true value of the adapted resolution of the historical monitoring image as the label to obtain a set of training data, inputting the training data into the preset network to obtain the predicted value of the adapted resolution; calculating the loss value of the preset network through the predicted value of the adapted resolution and the label, and using the cross-entropy loss function as the loss function; iteratively updating the parameters of the preset network, and stopping the update when the preset network reaches the set maximum number of training times or the loss value is less than the set loss value, so as to obtain the trained preset network.

[0017] Preferably, obtaining the true value of the adapted resolution includes: taking the same historical monitoring image at different resolutions as the model input, inputting it into the target detection model, outputting the detection accuracy rate, and taking the resolution corresponding to the model with the highest detection accuracy rate as the true value of the adapted resolution of the historical monitoring image.

[0018] In a second aspect, the present invention discloses a monitoring image resolution adaptive adjustment system, including: a processor; and a memory, where the memory stores computer instructions, and when the computer instructions are run by the processor, the system executes the above-mentioned monitoring image resolution adaptive adjustment method.

[0019] Advantages of the present invention: The present invention can not only dynamically adjust the resolution according to the change of the image content, ensure the image clarity while reducing the consumption of computing resources, but also gradually optimize the resolution adaptation through the training process of the deep learning network, and can adaptively adjust the quality of the monitoring image according to different environmental conditions. Through the method of the present invention, the monitoring system can respond more efficiently to the changing monitoring scenarios, improve the accuracy of image analysis, and ensure the efficient operation of the system under the condition of limited network bandwidth and storage resources, with strong practicability and forward-looking. Description of the Drawings

[0020] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understandable. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts, where: Figure 1 It is a flowchart of a method for adaptively adjusting the resolution of a monitoring image according to an embodiment of the present invention. Detailed implementation manners

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, rather than all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] It should be understood that when terms such as "first" and "second" are used in the claims, specifications, and drawings of the present invention, they are only used to distinguish different objects and not to describe a specific order. The terms "including" and "comprising" used in the specifications and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0023] The present invention provides a method for adaptively adjusting the resolution of a monitoring image. As Figure 1 shown, a method for adaptively adjusting the resolution of a monitoring image includes steps S1 - S4, which are specifically described below.

[0024] S1, obtain the preprocessed monitoring image at a preset sampling interval.

[0025] In one embodiment, at a preset sampling interval, the monitoring image captured by the monitoring camera is regularly obtained, and image preprocessing technology is used to improve the image quality. First, the histogram equalization technology is applied to adjust the brightness distribution of the image, enhance the details in the low-contrast area, and improve the global contrast of the image; at the same time, the contrast stretching method is used to further enhance the brightness difference in different regions of the image to more clearly display the target object. Then, the wavelet decomposition technology is combined to perform multi-scale analysis on the image, extract the feature information of different frequency levels, and use the filtering algorithm to remove the noise in the image, reducing the degradation of the image quality caused by environmental interference or transmission problems. Finally, through wavelet reconstruction and enhancement processing, the details and clarity of the image are optimized to ensure that the monitoring image is more stable and clear, providing more accurate data support for subsequent target detection and analysis.

[0026] S2. Calculate the difference image between the reference image and the target image, and calculate the fluctuation area of the target image based on the difference image.

[0027] It should be noted that when there is no object movement in the surveillance image captured by the surveillance camera, it means that there is no obvious change or fluctuation in the picture. At this time, the lowest resolution can be used for image acquisition to reduce the occupation of storage space and the consumption of transmission bandwidth, and ensure the effective utilization of system resources. On the contrary, when object movement or fluctuations are detected in the image, by analyzing the fluctuation area of the moving area, the degree of change in the scene can be evaluated in real time. According to the size of the fluctuation area, the system automatically adjusts the image resolution. If the fluctuation area is large, the resolution is increased to capture more details and accurate information; if the fluctuation area is small, the lower resolution is maintained to reduce unnecessary resource waste.

[0028] In one embodiment, the surveillance image at any sampling moment is used as the target image, and the surveillance image at the previous sampling moment adjacent to the target image is used as the reference image. Calculate the difference image between the reference image and the target image, and use the number of non-zero pixel points in the difference image as the fluctuation area of the target image.

[0029] The number of non-zero pixel points in the difference image is the fluctuation area of the target image, which reflects the changes or object movements in the image. This calculation method can effectively capture the dynamic changes in the surveillance scene, helping to accurately identify object movements or other potential events. When the fluctuation area is large, it indicates that significant changes have occurred in the image, and the resolution can be adjusted accordingly for more precise monitoring; when the fluctuation area is small, the lower resolution can be maintained to save system resources.

[0030] In one embodiment, the fluctuation area also satisfies the relationship: , represents the fluctuation area, represents the adaptive resolution of the reference image, represents the th column pixel value of the th row in the difference image,

[0031] When is 0, the value of the vector function is 0. When is not 0, the value of the vector function is 1.

[0032] The higher the resolution of the reference image, the finer the calculation of the fluctuation area, which can more accurately reflect the subtle changes in the image. After the pixel values in the difference image are processed by the vector function, the sensitivity of the image change is further enhanced, ensuring that the calculation of the fluctuation area can accurately respond to the movement in the scene. This method can dynamically adjust the resource usage of the monitoring system, accurately adjust the resolution according to the actual change amount, thereby improving the monitoring clarity while avoiding unnecessary calculation and storage burdens, maximizing the system's efficiency and resource utilization rate, and ensuring that the real-time monitoring process can quickly respond to changes and effectively manage the system load.

[0033] Exemplarily, obtain the monitoring image at time and the monitoring image at time If all pixel points in the difference image between the monitoring image at time and the monitoring image at time are 0, it means that there is no fluctuation in the monitoring image at time Then, the monitoring image at time captured at the resolution of the monitoring image at time is the accurate monitoring image. Retain the monitoring image at time and capture the monitoring image at time at the resolution of the monitoring image at time If there is a fluctuation in the monitoring image at time that is, there are pixel points not equal to 0 in the difference image between the monitoring image at time and the monitoring image at time

[0034] S3. Perform connected component extraction on the difference image to obtain the extraction result, and calculate the concentration degree of the target image according to the extraction result.

[0035] It should be noted that the adjustment of the resolution of the monitored image not only depends on the size of the fluctuation area, but is also closely related to the distribution of the fluctuation positions. When the fluctuations are concentrated in a specific area, it indicates that the changing targets are relatively concentrated. In this case, the recognition of the targets is usually easier because the concentrated changes can more prominently highlight the characteristics of the target objects. Therefore, when the fluctuation areas are equal, if the fluctuation distribution is concentrated, the system can identify the targets with a relatively low resolution; while when the fluctuation distribution is relatively dispersed, the resolution needs to be increased to capture more details in order to accurately identify the targets. Therefore, it is necessary to calculate the concentration degree of the fluctuations in the monitored image.

[0036] In one embodiment, connected component extraction is performed on the difference image to obtain an extraction result, and position clustering of the connected components is performed to obtain several clustering clusters.

[0037] The concentration degree satisfies the relational expression: , represents the concentration degree, represents the clustering cluster the sum of the areas of all connected components in represents the sum of the areas of all connected components in the difference image, represents the logarithmic function, represents the natural constant.

[0038] It can effectively identify whether the changing areas in the image are concentrated. If the concentrated changing area is small, the system can relatively easily identify the targets, so there is no need to overly increase the resolution; while when the changing areas are dispersed, more precise adjustment of the resolution is required to ensure that all targets can be identified.

[0039] In one embodiment, calculating the concentration degree of the target image according to the extraction result further includes: dividing the difference image into grids, and calculating the dependence matrix of any grid; taking the grids containing connected components as special grids, and taking the number of special grids in the 8-neighborhood direction of any grid as the dependence value.

[0040] The concentration degree satisfies the relational expression: , represents the concentration degree, represents the grid the dependence value of.

[0041] By carefully dividing the image area and considering the mutual dependence between neighborhoods, the calculation of the concentration degree becomes more accurate. If the fluctuation area distribution is relatively concentrated, the dependence value is higher and the concentration degree is correspondingly larger; otherwise, the concentration degree is lower.

[0042] This calculation method can effectively evaluate the distribution characteristics of the changing regions in the image. Specifically, regions with a higher concentration can be processed at a lower resolution because the targets are concentrated and easy to identify; while regions with a lower concentration may require a higher resolution to capture details.

[0043] S4. Input the fluctuation area and concentration degree of the target image into the preset network, and output the adaptive resolution of the target image to complete the resolution adaptive adjustment.

[0044] In one embodiment, the training process of the preset network is as follows: Use the historical surveillance images as input information, and use the true value of the adaptive resolution of the historical surveillance images as labels to obtain a set of training data. Input the training data into the preset network to obtain the predicted value of the adaptive resolution; Calculate the loss value of the preset network through the predicted value of the adaptive resolution and the labels. The loss function uses the cross-entropy loss function; Iteratively update the parameters of the preset network. When the preset network reaches the set maximum number of training times or the loss value is less than the set loss value, stop the update to obtain the trained preset network.

[0045] The true value of the adaptive resolution includes: Use the same historical surveillance image at different resolutions as the model input, input it into the target detection model, and output the detection accuracy. Use the resolution corresponding to the model with the highest detection accuracy as the true value of the adaptive resolution of the historical surveillance image.

[0046] The true value of the adaptive resolution can also be set by those skilled in the art.

[0047] Thus, a trained preset network can be obtained. Input the fluctuation area and concentration degree of the real-time captured surveillance image into the preset network, and output the adaptive resolution of the real-time captured surveillance image, thereby completing the resolution adaptive adjustment.

[0048] The embodiment of the present invention also discloses a surveillance image resolution adaptive adjustment system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, it implements a surveillance image resolution adaptive adjustment method according to the present invention.

[0049] The above system also includes other components well known to those skilled in the art such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be described in detail here.

[0050] In the present invention, the aforementioned memory may be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory (RRAM), a dynamic random access memory (DRAM), a static random access memory (SRAM), an enhanced dynamic random access memory (EDRAM), a high-bandwidth memory (HBM), a hybrid memory cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application program, a module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

[0051] Although this specification has shown and described multiple embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many variations, changes, and alternative approaches will occur to those skilled in the art without departing from the spirit and scope of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

[0052] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A method for adaptively adjusting monitoring image resolution, characterized in that: include: Acquire the preprocessed surveillance image according to the preset sampling interval; The monitoring image at any sampling moment is taken as the target image, the monitoring image at the previous sampling moment adjacent to the target image is taken as the reference image, the difference image between the reference image and the target image is calculated, and the fluctuation area of ​​the target image is calculated according to the difference image; Performing connected domain extraction on the difference image to obtain extraction results, and calculating the concentration degree of the target image according to the extraction results; The fluctuation area and concentration of the target image are input into the preset network, and the adapted resolution of the target image is output to complete the adaptive resolution adjustment.

2. The method for adaptively adjusting monitoring image resolution according to claim 1, characterized in that: The number of non-zero pixels in the differential image is taken as the fluctuation area of ​​the target image.

3. The method for adaptively adjusting monitoring image resolution according to claim 1, characterized in that: The fluctuation area also satisfies the relationship: , represents the fluctuation area, represents the adapted resolution of the reference image, Indicates the difference image Line The pixel value of the column, represents the vector function, Represents a natural constant.

4. The method for adaptively adjusting monitoring image resolution according to claim 1, characterized in that: Calculating the concentration degree of the target image according to the extraction result includes: Cluster the locations of connected domains to obtain several clusters; The concentration degree satisfies the relationship: , Indicates the degree of concentration, Represents clusters The sum of the areas of all connected domains in represents the sum of the areas of all connected domains in the difference image, represents the logarithmic function, Represents a natural constant.

5. The method for adaptively adjusting monitoring image resolution according to claim 1, characterized in that: The step of calculating the concentration level of the target image according to the extraction result further comprises: Grid the difference image and calculate the dependency matrix of any grid; The grid containing the connected domain is regarded as a special grid, and the number of special grids of any grid in the 8-neighborhood direction is regarded as the dependent value; The concentration degree satisfies the relationship: , Indicates the degree of concentration, Representation Grid The dependent value of .

6. The method for adaptively adjusting monitoring image resolution according to claim 1, characterized in that: The training process of the preset network is: The historical monitoring images are used as input information, and the true values ​​of the adapted resolutions of the historical monitoring images are used as labels to obtain a set of training data, and the training data are input into a preset network to obtain the predicted values ​​of the adapted resolutions; The loss value of the preset network is calculated by adapting the resolution prediction value and label, and the loss function uses the cross entropy loss function; Iteratively update the parameters of the preset network. When the preset network reaches the set maximum number of training times or the loss value is less than the set loss value, stop updating to obtain a trained preset network.

7. The method for adaptively adjusting monitoring image resolution according to claim 6, characterized in that: Obtaining the true value of the adapted resolution includes: The same historical monitoring image at different resolutions is used as model input and input into the target detection model. The detection accuracy is output, and the resolution corresponding to the model input with the highest detection accuracy is used as the true value of the adapted resolution of the historical monitoring image.

8. A monitoring image resolution adaptive adjustment system, characterized in that: include: Processor; and A memory storing computer instructions, wherein when the computer instructions are executed by a processor, the system executes a monitoring image resolution adaptive adjustment method according to any one of claims 1 to 7.

Images