A target dynamic scene filtering method, device, equipment and storage medium

Abstract

The present application discloses a method, apparatus, device, and medium for filtering a target dynamic scene, creating a dynamic scene template, and defining a learning rate. The dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template. Filtering is performed on the dynamic scene, and filtering operations are performed on the static shielding area and the objects in the static shielding area. The present invention adds dynamic filtering to object detection, filters common misdetection phenomena such as night breathing lights, dust, and flying insects, significantly reduces the occurrence of misdetection and missed detection in object detection, and improves the detection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a target dynamic scene filtering method, device, equipment and storage medium. Background Art

[0002] Currently, kitchen hygiene in catering establishments is a major concern for consumers, and food safety issues have frequently trended in recent years. In particular, incidents involving rats in restaurants pose a serious threat to consumer health and significantly damage the reputation of catering businesses. Since 2014, the China Food and Drug Administration, taking into account the specific characteristics of food safety regulation in the catering industry, has deployed local food and drug regulatory authorities to guide the "Open Kitchen and Stove" initiative in the catering industry. With the development and advancement of video analysis and artificial intelligence technologies, an increasing number of incidents can be detected and identified in real time through video surveillance. Therefore, the use of video analysis to detect rats in surveillance scenes, and the real-time recording and alerting of rat tracks are gradually becoming intelligent. However, existing technologies, such as the invention patent with publication number CN114419560A, suffer from errors in nighttime image capture. Therefore, reducing errors and false alarms in nighttime rat detection is a pressing technical issue in this field. Summary of the Invention

[0003] In view of this, the present invention aims to provide a method for filtering dynamic scenes, and also provides a corresponding device, equipment, and storage medium for filtering dynamic scenes, which can support sample images of different sizes and improve the accuracy of the model's detection at night. The specific scheme is as follows:

[0004] A first aspect of the present application provides a target dynamic scene filtering method, comprising:

[0005] Step S11: creating a dynamic scene template and defining a learning rate, wherein the dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template;

[0006] Step S12: performing filtering on the dynamic scene, and performing filtering operations on the static shielding area and the objects in the static shielding area.

[0007] Furthermore, the static shielding area counting template is responsible for defining the area of continuously changing objects under the lens; the dynamic shielding area counting template is responsible for filtering occasional objects or events under the lens; and the learning rate is used to adjust the counting amplitude under different templates and different states.

[0008] Furthermore, the specific process of performing filtering on the dynamic scene includes:

[0009] Step S121, initialize the dynamic scene, count the foreground template information of a fixed number of frames within a predetermined time. At this time, no reporting is performed. Starting from the first frame, traverse all pixel values of the foreground template, filter out the moving point coordinates, and increase the learning rate for the corresponding positions of the static shielding area counting template.

[0010] Step S122, perform dynamic update on the dynamic scene, and update it once every other time period.

[0011] Furthermore, preferably, it further includes step S123, reporting filtering. Calculate the ratio of the foreground points in the foreground frame to the foreground area, and determine the foreground frames with the ratio less than the flying insect threshold as false detections of flying insects and filter them out.

[0012] Furthermore, count a certain number of frames of images in the previous several times, retain the points with the moving frame count exceeding the statistical threshold, set the remaining points to zero, and then frame the non-zero points with a rectangular box. The area within the box is the static shielding area and subsequent moving frame measurements and counts are not performed.

[0013] Furthermore, the dynamic update performs the following operations:

[0014] Step S1221, traverse the foreground template for each frame to filter out moving points, and increase the learning rate for the corresponding positions of the dynamic shielding area counting template. For non-moving point positions, subtract 1 / 4 * learning rate. Statistically filter once every preset number of frames, and filter out the areas greater than zero as the temporary shielding area.

[0015] Step S1222, set shielding box fusion to avoid false detections of incomplete temporary shielding areas.

[0016] Step S1223, calculate the total area of the dynamic shielding box. When the area is greater than the comparison threshold * image area, it is determined that the change in the picture within the lens is too large. If the change is too large, re-initialize the dynamic scene template.

[0017] Furthermore, the step of setting shielding box fusion has the following specific process:

[0018] Filter out independent shielding boxes and shielding boxes with adjacent boxes. Traverse all temporary shielding boxes and calculate the distance between the current box and the other boxes.

[0019] Merge adjacent shielding boxes.

[0020] After the loop ends, splice the list of shielding boxes with adjacent boxes after merging with the list of independent shielding boxes as the dynamic shielding box.

[0021] In a second aspect, the present invention discloses a video transcoding device, including:

[0022] A configuration module for creating a dynamic scene template and defining a learning rate. The dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template;

[0023] A filtering module, configured to perform filtering on the dynamic scene, and perform filtering operations on the static shielding area and the objects in the static shielding area.

[0024] In a third aspect, the present invention discloses an electronic device, including:

[0025] A memory, configured to store a computer program;

[0026] A processor, configured to execute the computer program to implement the steps of the foregoing disclosed target dynamic scene filtering method.

[0027] In a fourth aspect, the present invention discloses a computer-readable storage medium, configured to store a computer program; wherein, when the computer program is executed by a processor, the steps of the foregoing disclosed target dynamic scene filtering method are implemented.

[0028] The present application discloses a target dynamic scene filtering method, device, equipment, and medium, which create a dynamic scene template and define a learning rate. The dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template, perform filtering on the dynamic scene, and perform filtering operations on the static shielding area and the objects in the static shielding area. The present invention adds dynamic filtering in target detection, filters common misdetection phenomena such as night breathing lights, dust, and flying insects, significantly reduces the occurrence of misdetection and missed detection in target detection, and improves the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0030] Figure 1 It is a flowchart of a target dynamic scene filtering method provided by the present application;

[0031] Figure 2 It is a flowchart of a specific target dynamic scene filtering method provided by the present application;

[0032] Figure 3 It is a flowchart of a specific preset ratio determination method provided by the present application;

[0033] Figure 4 It is a flowchart of a specific preset ratio determination method provided by the present application;

[0034] Figure 5A structural diagram of an object detection electronic device provided for this application. Specific implementation manners

[0035] 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 only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] Currently, the algorithms of the prior art mainly detect moving targets at night and then report through counting logic. It can realize various functions such as intrusion detection, mouse detection, and other living body detections. However, the actual application scenarios are relatively complex. The flashing lights at night, the dust in front of the lens, and flying insects will all cause false detections. The present invention mainly optimizes these false detection situations, and optimizes the prior art algorithms by adding a dynamic scene filtering module to improve the accuracy of night target detection.

[0037] Figure 1 A flowchart of a method for filtering dynamic scenes of an object provided for an embodiment of this application. When the algorithm starts, it is like a breathing light flashing gradually or causing changes in the surrounding environment and is easily detected by motion detection, so it needs to be filtered in advance.

[0038] See Figure 1 As shown, the method for filtering dynamic scenes of an object includes:

[0039] Step S11: Create a dynamic scene template and define a learning rate. The dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template;

[0040] In this embodiment, the dynamic scene template includes MLModel and MSModel templates. The MLModel is a static shielding area counting template, which is mainly responsible for defining the area of continuously changing items under the lens; the MSModel is a dynamic shielding area counting template, which is mainly responsible for filtering occasional items or events under the lens; the learning rate λ is used to adjust the counting amplitude in different templates and different states.

[0041] Step S12: Filter the dynamic scene and perform filtering operations on the static shielding area and the objects in the static shielding area.

[0042] When the dynamic scene template is initialized, it is a zero matrix of w*h, and the initial value of λ is 1. The dynamic scene filtering includes two states: initialization and dynamic update. More preferably, it can also include a third state of reporting and filtering. As shown in the appendix Figure 2 As shown, the specific process of filtering the dynamic scene includes:

[0043] Step S121: Initialize the dynamic scene, and count the foreground template information of a fixed number of frames within a predetermined time. At this time, no reporting is performed. Starting from the first frame, traverse all pixel values of the foreground template, filter out the moving point coordinates, and increase the learning rate for the corresponding position of the static shielding area counting template.

[0044] The initialization of the template mainly filters the continuously changing areas under the camera. Initializing the MLModel template requires counting the foreground template FGModel information of 1200 frames (25 frames * 60s). At this time, no reporting of dynamic targets is performed. Starting from the first frame, traverse all pixels pix[i][j] of the foreground template FGMoldel, filter out the moving point coordinates (x, y), and increase the learning rate λ for the corresponding position of the MLModel template. The requirement for a moving point in the foreground template FGModel is that the number of matches between this pixel point and the background frame template sample is less than or equal to the first threshold, and this threshold can be set to 5 or other values around 5.

[0045] It should be noted that the matching requirement is (abs(pix[i][j] - samples[i][j][k])) < R, where the optimal value of R (i.e., the second threshold) is 20. When calculating pix[i][j] - samples[i][j][k], it is the subtraction of numerical values, and the abs operation is to take the absolute value. pix[i][j] refers to the pixel value of the point with coordinates (i, j) in this image, and samples[i][j][k] is the background frame template, which can be set as a matrix samples[i][j][k] of w * h * numSample (a three-dimensional matrix, samples[i][j][k] represents the pixel value size of the point with coordinates (i, j) in the k-th sample of the background frame template, where k ∈ [0, numSample]). w and h are the height and width of the input image, and numSample is the number of samples in the background frame template sample set. The sample set includes at least one sample. Assuming numSample is equal to 20 and there is only one image for initialization, when initializing the background frame template, generally 20 points in the neighborhood of the target point in the image are randomly selected through the Vibe algorithm, and a sample set of the background frame template is established for each point. This method has the advantages of fast initialization speed, low memory consumption, and low resource occupancy.

[0046] In this embodiment, several frames of images in the previous several time periods are counted, and the points with the moving frame count exceeding the statistical threshold are retained, and the remaining points are set to zero. Then, the non-zero points are framed with a rectangular box, and the area inside the box is the static shielding area, and no moving frame measurement count is performed subsequently. In a specific embodiment, the images of about 200 frames in the previous 8 s are counted, and the points with the moving frame count exceeding 100 are retained, and the remaining points are set to zero. The non-zero points are framed with a rectangular box, and the area inside the box is the static shielding area, and no moving frame measurement count is performed subsequently.

[0047] Step S122, perform dynamic update on the dynamic scene, and update it once every other time period;

[0048] In this embodiment, the dynamic scene needs to be updated in real time, and the target reporting in the dynamic update stage and the dynamic scene filtering are carried out simultaneously. The dynamic scene filtering is updated once every other time period (such as 10 s, 15 s, etc.), which is mainly used for the optimization of the occasional situation of dust and luminous screens under night shots. Specifically, as shown in the appendix Figure 3 shown, the dynamic update performs the following operations:

[0049] Step S1221, traverse the foreground template for each frame to screen for moving points, and increase the learning rate at the corresponding position of the dynamic shielding area counting template, and subtract 1 / 4 * learning rate from the position of the non-moving points. Statistically screen the area greater than zero as a temporary shielding area every preset number of frames;

[0050] Specifically, traverse the foreground template FGMoldel for each frame to screen for moving points, and increase the learning rate λ at the corresponding position of the dynamic shielding area counting template MSModel. At this time, the value of the learning rate λ is set to 2, and the position of the non-moving points is subtracted by λ / 4. Statistically screen once every 250 frames (25 frames * 10 s), and screen the area with a value greater than 0 as a temporary shielding box. The area framed by the temporary shielding box is the temporary shielding area.

[0051] Step S1222, set the shielding box fusion to avoid false detection of incomplete temporary shielding areas;

[0052] As mentioned above, according to the phenomenon that the lens imaging moves slowly and the light and dark distribution is uneven in this kind of situation, the temporary shielding area often can only frame out part of the interfering object area, and the incomplete shielding area also causes false detection. Therefore, the step of shielding box fusion is added, and the specific process is as follows:

[0053] 1) Screen the independent shielding boxes and the shielding boxes with adjacent boxes, traverse all the temporary shielding boxes, and calculate the distance between the current box and the other boxes.

[0054] Create two lists respectively with a distance of 10 pix as the boundary. Those exceeding the boundary are stored in the independent shielding box list rect1, and otherwise are stored in the shielding box list rect2 with adjacent boxes.

[0055] 2) Merge adjacent shielding frames.

[0056] Loop through the list of shielding frames rect2 with adjacent frames, divide all the frames in the list into the current frame and the remaining frames. The current frame represents the shielding frame being traversed. At this time, all frames in the list except the current frame are the remaining frames. Merge the current frame with the frames in the remaining frames whose distance is less than 10 pix. The merging method is to take the union of the two frames. After merging, replace the current frame with the merged frame.

[0057] 3) After the loop ends, splice the merged list of shielding frames rect2 with adjacent frames and the list of independent shielding frames rect1 as the dynamic shielding frame.

[0058] In step S1223, calculate the total area of the dynamic shielding frame. When the area is greater than 1 / 3 of the image plane, determine whether the picture in the camera lens changes too much. If it changes too much, re-initialize the dynamic scene template, including initializing both the MLModel and MSModel.

[0059] Abnormal update: Calculate the total area of the dynamic shielding frame. When the area is greater than 1 / 3 of the image plane (which can be understood as a comparison threshold), determine whether the picture in the camera lens changes too much. If it changes too much, generally it is due to the change of the camera position or the situation such as turning the light on and off. When it cannot be processed only by dynamic update at this time, re-initialize the dynamic scene template. If the change is not significant, it may indicate a mouse, dust, or a monitor with a changing picture. At this time, there is no need to re-initialize the template.

[0060] As can be seen from the above, the present invention can already optimize the accuracy of night detection through target dynamic scene filtering. However, in addition, there are often flying insects in camera monitoring. The flying insect targets are small and the moving areas are uncertain. Therefore, neither the static nor the dynamic shielding areas can effectively filter out false detections of flying insects.

[0061] Therefore, in this embodiment, more preferably, the present invention proposes reporting and filtering.

[0062] In step S123, for reporting and filtering, calculate the ratio of the foreground points in the foreground frame to the foreground area, and determine the foreground frames with the ratio less than the flying insect threshold as false detections of flying insects and filter them.

[0063] Reporting and filtering mainly aims at false detections of flying insects. In this embodiment, according to the characteristic that the foreground points in the foreground frame detected by flying insect movement are basically a thin line, calculate the ratio p of the foreground points in the foreground frame to the foreground area. Determine the foreground frames with p less than 0.1 as false detections of flying insects and filter them.

[0064] It can be seen that there will be many false detections when only using motion detection algorithms for object detection in the present invention. A dynamic scene filtering method is added, which can filter common false detection phenomena such as night breathing lights, dust, and flying insects, significantly reduce the occurrence of false detections of mice, and ensure the accuracy of mouse detection.

[0065] The present invention mainly uses machine learning algorithms to solve the common mouse detection problem in the retail industry. On the basis of the existing technology, a dynamic scene filtering method is added, which greatly improves the detection rate of mouse infestation judgment and avoids false detections of night lights at the same time. At the same time, the night mouse infestation judgment method adopted by the present invention can avoid false judgments and missed judgments of mouse detection in most cases, and well solves the health and safety problems in the retail industry.

[0066] See Figure 4 As shown, the embodiment of the present application also correspondingly discloses a device for a target dynamic scene filtering method, including:

[0067] A configuration module for creating a dynamic scene template and defining a learning rate, where the dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template;

[0068] A filtering module for performing filtering on the dynamic scene and performing filtering operations on the static shielding area and the objects in the static shielding area.

[0069] It can be seen that the embodiment of the present application also provides a device for a target dynamic scene filtering method, which realizes the filtering optimization of night object detection.

[0070] Furthermore, the embodiment of the present application also provides an electronic device. Figure 5 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation on the scope of use of the present application.

[0071] Figure 5 It is a schematic structural diagram of an electronic device 20 provided by the embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the target dynamic scene filtering method and the related steps in the target dynamic scene filtering method disclosed in any of the foregoing embodiments.

[0072] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and no specific limitation is imposed on it here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application requirements, and no specific limitation is imposed here.

[0073] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, a random access memory, a magnetic disk, an optical disc, etc. The resources stored thereon can include an operating system 221, a computer program 222, data 223, etc., and the storage method can be temporary storage or permanent storage.

[0074] Among them, the operating system 221 is used to manage and control each hardware device on the electronic device 20 and the computer program 222, so as to realize the operation and processing of the massive data 223 in the memory 22 by the processor 21. It can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the target dynamic scene filtering method and the target dynamic scene filtering method executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks. The data 223 can include sample images collected by the electronic device 20.

[0075] Furthermore, an embodiment of this application also discloses a storage medium in which a computer program is stored. When the computer program is loaded and executed by a processor, the steps of the target dynamic scene filtering method and the target dynamic scene filtering method disclosed in any of the foregoing embodiments are realized.

[0076] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts between the various embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description of the method part for related parts.

[0077] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the said element.

[0078] The above has introduced in detail the target dynamic scene filtering method, the device, equipment and storage medium thereof provided by the present invention. Specific examples are used in this text to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for filtering a target dynamic scene, characterized in that Including: Step S11: Create a dynamic scene template and define a learning rate. The dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template; Step S12: Filter the dynamic scene and perform filtering operations on the static shielding area and the objects in the static shielding area; Among them, the static shielding area counting template is responsible for defining the area of continuously changing items under the camera; the dynamic shielding area counting template is responsible for filtering occasional items or events under the camera; the learning rate is used to adjust the counting amplitude in different templates and different states; The specific process of filtering the dynamic scene includes: Step S121, Initialize the dynamic scene, count the foreground template information of a fixed number of frames within a predetermined time. At this time, no reporting is performed. Starting from the first frame, traverse all pixel values of the foreground template, filter out the moving point coordinates, and increase the learning rate for the corresponding position of the static shielding area counting template; Step S122, Dynamically update the dynamic scene, and update it once every other time period; Count a certain number of frames of images in the previous period of time, retain the points where the moving frame count exceeds the statistical threshold, set the remaining points to zero, and then frame the non-zero points with a rectangular box. The area within the box is the static shielding area, and no moving frame measurement count will be performed subsequently.

2. The target dynamic scene filtering method according to claim 1, wherein It also includes step S123, Report filtering, calculate the ratio of the foreground points in the foreground box to the foreground area, and judge the foreground box with a ratio less than the fly threshold as a false detection of flies and perform filtering.

3. The target dynamic scene filtering method according to claim 1, characterized in that: The dynamic update performs the following operations: Step S1221, Traverse the foreground template for each frame to filter out moving points, and increase the learning rate for the corresponding position of the dynamic shielding area counting template. For non-moving point positions, subtract 1 / 4 * the learning rate. Statistically filter once every preset number of frames, and filter out the areas greater than zero as the temporary shielding area; Step S1222, Set shielding box fusion to avoid false detection of incomplete temporary shielding areas; Step S1223, Calculate the total area of the dynamic shielding box. When the area is greater than the comparison threshold * the image area, judge that the change in the picture within the camera is too large. If the change is too large, re-initialize the dynamic scene template.

4. The target dynamic scene filtering method according to claim 3, characterized in that, The steps of setting shielding box fusion are as follows: Filter out independent shielding boxes and shielding boxes with adjacent boxes, traverse all temporary shielding boxes, and calculate the distance between the current box and the other boxes; Merge adjacent shielding boxes; the adjacent shielding boxes are boxes with a distance less than 10pix from the current shielding box; After the loop ends, splice the list of shielding boxes with adjacent boxes after merging and the list of independent shielding boxes as the dynamic shielding box.

5. A target dynamic scene filtering device, characterized in that: Including: Configuration module, used to create a dynamic scene template and define a learning rate. The dynamic scene template includes a static shielding area counting template and a dynamic shielding area counting template; Filtering module, used to filter the dynamic scene and perform filtering operations on the static shielding area and the objects in the static shielding area; Among them, the static shielding area counting template is responsible for defining the area of continuously changing items under the camera; the dynamic shielding area counting template is responsible for filtering occasional items or events under the camera; the learning rate is used to adjust the counting amplitude in different templates and different states; The specific process of performing filtering on the dynamic scene includes: initializing the dynamic scene, counting the foreground template information of a fixed number of frames within a predetermined time, and not reporting at this time. Starting from the first frame, traverse all pixel values of the foreground template, screen out the moving point coordinates, and increase the learning rate for the corresponding position of the static shielding area counting template; perform dynamic update on the dynamic scene, and update it once every other time period; Count several frames of images in the previous several time periods, retain the points where the moving frame count exceeds the statistical threshold, set the remaining points to zero, and then frame out the non-zero points with a rectangular box. The area within the box is the static shielding area, and no moving frame measurement count will be performed subsequently.

6. An electronic device, characterized in that: It includes: A memory for storing a computer program; A processor for executing the computer program to implement the steps of the target dynamic scene filtering method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, For storing a computer program; wherein, when the computer program is executed by the processor, the steps of the target dynamic scene filtering method according to any one of claims 1 to 4 are implemented.

Images