Method and system for realizing intelligent analysis of camera shooting area based on deep learning

Abstract

The invention relates to the technical field of camera shooting intelligent analysis, and discloses a method for realizing intelligent analysis of a camera shooting area based on deep learning, and the method comprises the steps: collecting a target video of a to-be-analyzed camera shooting area, carrying out the framing of the target video, obtaining a plurality of frames of images, calculating the loss and feature point number of the images, and repairing the plurality of frames of images, obtaining a repaired image; querying a tracking target in the repaired image, dynamically marking the tracking target to obtain a dynamically marked target, and querying continuous associated frames of the dynamically marked target to dynamically update the dynamically marked target to obtain a dynamically updated target; performing feature association on the dynamic target according to the appearance feature of the tracking target to obtain a target association feature; and performing simulation tracking on the dynamic target based on the dynamically updated target and the target association characteristics, and performing real-time intelligent analysis on the to-be-analyzed target when a simulation tracking score meets a preset score value to obtain an analysis report. According to the invention, the intelligent analysis precision of the camera shooting area can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent camera analysis, and in particular to a method for realizing intelligent analysis of a camera area based on deep learning. Background Art

[0002] With the widespread use of surveillance cameras, a large amount of video data has been generated. How to effectively extract valuable information from this data has become an important research topic. Camera area intelligent analysis technology has emerged. It uses computer vision and image processing technology to detect, identify, track and analyze targets in the camera area, thereby realizing intelligent monitoring and management of the monitoring scene, and through real-time monitoring and analysis of the behavior of people and objects in the camera area, timely discover abnormal situations such as theft, robbery, fire, etc., and improve security prevention capabilities.

[0003] At present, the intelligent analysis method for the camera area is generally based on behavioral analysis technology. By classifying and identifying the detected target objects, determining their types and attributes, and then realizing intelligent analysis of the detected targets, however, when using this method, due to environmental influences, target occlusion often occurs, resulting in the loss of the detected target or the misalignment of the detected target, resulting in inaccurate analysis. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a method for intelligent analysis of the camera area based on deep learning, which can improve the accuracy of intelligent analysis of the camera area.

[0005] In a first aspect, the present invention provides a method for intelligently analyzing a camera area based on deep learning, comprising: Collecting a camera video of a camera area to be analyzed, identifying a dynamic target in the camera video, and extracting a target video in the camera video based on the dynamic target; Performing video framing on the target video to obtain multiple frame images, calculating pixel losses of the multiple frame images, identifying feature points of the multiple frame images, and calculating feature losses of the multiple frame images based on the feature points, and performing image restoration on the multiple frame images based on the pixel losses and the feature losses to obtain restored images; Querying a tracking target in the repaired image, dynamically marking the tracking target to obtain a dynamically marked target, querying continuous associated frames of the dynamically marked target, and dynamically updating the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target; Identifying appearance features of the tracking target, encoding the appearance features to obtain encoding features, and performing feature association on the dynamic target based on the encoding features to obtain target association features; Based on the dynamically updated target and the target association characteristics, a dynamic analysis strategy for the dynamic target is constructed, the dynamic analysis strategy is used to simulate tracking of the dynamic target, and a simulation score of the simulated tracking is calculated. When the simulation score meets a preset score value, the dynamic analysis strategy is used to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed to obtain a target analysis report.

[0006] In a possible implementation manner of the first aspect, calculating the pixel loss of the multiple frames of images includes: The pixel loss of the multi-frame images is calculated using the following formula: ; in represents pixel loss, Indicates the image length of multiple frames of images, Indicates the image width of the multi-frame image, n indicates the original pixel position of the multi-frame image, and m indicates the pixel position of the multi-frame image after framing. Represents the pixel loss of multiple frames of images, Represents the image edge loss of multiple frames.

[0007] In a possible implementation manner of the first aspect, calculating the feature loss of the multiple frames of images based on the number of feature points includes: Calculating characteristic values ​​of the multiple frames of images; Based on the feature value and the number of feature points, the feature loss of the loss image is calculated using the following formula: ; in, represents feature loss, n represents the number of feature points of the loss image, represents the feature matching function, Represents the matrix space of multiple frames of images, A represents the image features, Represents the eigenvalue.

[0008] In a possible implementation manner of the first aspect, performing image restoration on the multiple frames of images based on the pixel loss and the feature loss to obtain a restored image includes: Querying the loss image containing pixel loss and feature loss in the multiple frames of images, According to the feature loss and the pixel loss, the loss image is repaired using the following formula to obtain a repaired image: ; in, represents the repaired image, k represents the number of pixels before the loss image is processed, y represents the number of pixels after the multi-frame image is processed, a represents the feature value before the multi-frame image is processed, and b represents the feature value after the multi-frame image is processed. , Represents the feature compensation function of multiple frames of images, Represents the feature loss of multiple frames of images, Represents the pixel compensation function of multiple frame images.

[0009] In a possible implementation manner of the first aspect, dynamically marking the tracking target to obtain the dynamically marked target includes: Calculating the pixel value of the tracking target; Identifying a characteristic direction of the tracking target; Performing target positioning on the tracking target based on the pixel value and the characteristic direction to obtain a positioning target; A positioning identifier of the positioning target is constructed, and the tracking target is dynamically marked based on the positioning identifier to obtain a dynamically marked target.

[0010] In a possible implementation manner of the first aspect, dynamically updating the dynamically marked target based on the continuous associated frames to obtain the dynamically updated target includes: Based on the continuous associated frames, constructing the movement trajectory of the dynamically marked target; Based on the motion trajectory, analyzing the speed change of the dynamically marked target, and querying the position change of the dynamically marked target; The dynamically marked target is dynamically updated based on the speed change and the position change to obtain a dynamically updated target.

[0011] In a possible implementation manner of the first aspect, analyzing the speed change of the dynamically marked target based on the action trajectory includes: Based on the motion trajectory, query the motion-related frame number of the dynamically marked target; Based on the motion-related frame number, the speed change of the dynamic marking target is calculated using the following formula: ; in, Indicates speed change, Indicates the horizontal speed of the dynamic marker target. represents the vertical speed of the dynamic marker target, N represents the number of motion-related frames, t represents time, Indicates the horizontal position of the dynamic marking target at the i-th moment, Indicates Dynamically mark the horizontal position of the target at all times. Indicates the vertical position of the dynamic marking target at the i-th moment, Indicates Dynamically mark the target's vertical position at all times.

[0012] In a possible implementation manner of the first aspect, the identifying the appearance feature of the tracking target includes: identifying a feature descriptor of the tracking target; Based on the feature descriptor, feature matching is performed on the tracking target to obtain matching information; Querying feature points of the tracking target; The target information of the matching information is determined by using the feature points, and the appearance features of the tracking target are identified based on the target information.

[0013] In a possible implementation manner of the first aspect, calculating the simulation score of the simulation tracking includes: Extracting multiple frames of associated images of the dynamic target; The preconfigured depth model is used to track the corresponding dynamic targets in the multi-frame associated images, and the target tracking score is calculated using the following formula: ; Among them, SS represents the target tracking score, represents the number of false positives in the kth frame, represents the number of false negatives in the kth frame, represents the number of identity switching times in the kth frame, Indicates the number of true bounding boxes in the kth frame; Based on the target tracking score, a simulation score of the simulated tracking is determined.

[0014] In a second aspect, the present invention provides a system for intelligently analyzing a camera area based on deep learning, the system comprising: The target extraction module is used to collect the camera video of the camera area to be analyzed, identify the dynamic target in the camera video, and extract the target video in the camera video based on the dynamic target; An image restoration module, configured to perform video framing on the target video to obtain multiple frames of images, calculate pixel losses of the multiple frames of images, identify feature points of the multiple frames of images, calculate feature losses of the multiple frames of images based on the feature points, and perform image restoration on the multiple frames of images based on the pixel losses and the feature losses to obtain restored images; a target updating module, configured to query a tracking target in the repaired image, dynamically mark the tracking target to obtain a dynamically marked target, query continuous associated frames of the dynamically marked target, and dynamically update the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target; A target association module, used for identifying the appearance features of the tracking target, performing feature encoding on the appearance features to obtain encoding features, and performing feature association on the dynamic target based on the encoding features to obtain target association features; The intelligent analysis module is used to construct a dynamic analysis strategy for the dynamic target based on the dynamically updated target and the target association characteristics, use the dynamic analysis strategy to simulate tracking of the dynamic target, calculate a simulation score of the simulated tracking, and when the simulation score meets a preset score value, use the dynamic analysis strategy to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed, and obtain a target analysis report.

[0015] Compared with the prior art, the technical principle and beneficial effects of this solution are: The embodiment of the present invention can automatically analyze the camera area through the acquisition of the camera video of the camera area to be analyzed, quickly obtain valuable information, and perform video framing on the target video to obtain multiple frames of images, which can make it easier to understand the movements and changes of the target object in the video, such as the movement trajectory of an athlete in a sports video, or the driving trajectory of a car on a road, and identify the number of feature points of the multiple frames of images, and calculate the feature loss of the multiple frames of images based on the feature points to repair the image, thereby improving the quality of the generated image and enhancing the specific features of the image. Further, the embodiment of the present invention can improve the image quality by performing image repair on the multiple frames of images based on the pixel loss and the feature loss to obtain a repaired image, so that the target point detection in the image is more accurate. Furthermore, the embodiment of the present invention can locate the position of the target to be analyzed in the video in real time by querying the tracking target in the repaired image and dynamically marking the tracking target to obtain a dynamically marked target, thereby avoiding the loss of the analysis target, and dynamically updating the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target to understand the changes of the tracking target within a certain period of time, such as the speed change of a car within a certain period of time. The embodiments of the present invention provide a method and system for realizing intelligent analysis of a camera area based on deep learning, which can improve the accuracy of intelligent analysis of the camera area. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 A schematic diagram of a process for implementing an intelligent analysis method of a camera area based on deep learning provided by an embodiment of the present invention; Figure 2 A schematic diagram of a module of a system for intelligent analysis of a camera area based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0020] The embodiment of the present invention provides a method for intelligent analysis of a camera area based on deep learning, and the execution subject of the method for intelligent analysis of a camera area based on deep learning includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present invention. In other words, the method for intelligent analysis of a camera area based on deep learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0021] See also Figure 1 FIG. 1 is a flow chart of a method for realizing intelligent analysis of a camera area based on deep learning according to an embodiment of the present invention. Figure 1 A method for intelligent analysis of camera areas based on deep learning described in the specification includes: S1. Collecting a camera video of a camera area to be analyzed, identifying a dynamic target in the camera video, and extracting a target video in the camera video based on the dynamic target.

[0022] The embodiment of the present invention can automatically analyze the camera area by collecting the camera video of the camera area to be analyzed, and quickly obtain valuable information. The camera area to be analyzed can be a traffic management scene, an intelligent security scene, and an environmental monitoring scene.

[0023] Furthermore, the embodiment of the present invention can quickly locate the dynamic target by identifying the dynamic target in the camera video and extracting the target video in the camera video based on the dynamic target, so as to quickly obtain the video segment corresponding to the target to be analyzed, thereby improving the analysis efficiency.

[0024] Optionally, the dynamic target can be obtained by constructing a background model of the camera video, using the background model to mark the part of the camera video that is different from the background as a dynamic target, and then cutting the corresponding video according to the dynamic target to obtain a target video.

[0025] S2. Divide the target video into frames to obtain multiple frame images, calculate the pixel loss of the multiple frame images, identify the number of feature points of the multiple frame images, and calculate the feature loss of the multiple frame images based on the number of feature points, and perform image restoration on the multiple frame images based on the pixel loss and the feature loss to obtain a restored image.

[0026] The embodiment of the present invention performs video framing on the target video to obtain multiple frames of images, which can make it easier to understand the movements and changes of the target object in the video, such as the movement trajectory of an athlete in a sports video or the driving trajectory of a car on a road.

[0027] Optionally, the target video may be segmented into frames using a segmentation tool, such as an image processing library in Python.

[0028] Furthermore, the embodiment of the present invention can help the user determine the extent of image loss during the processing by calculating the pixel loss of the multiple frames of images.

[0029] As an embodiment of the present invention, the calculating the pixel loss of the multiple frames of images includes: calculating the pixel loss of the multiple frames of images using the following formula: ; in represents pixel loss, Indicates the image length of multiple frames of images, Indicates the image width of the multi-frame image, n indicates the original pixel position of the multi-frame image, and m indicates the pixel position of the multi-frame image after framing. Represents the pixel loss of multiple frames of images, Represents the image edge loss of multiple frames.

[0030] The embodiment of the present invention can repair the image by identifying the number of feature points of the multi-frame images and calculating the feature loss of the multi-frame images based on the number of feature points, thereby improving the quality of the generated image and enhancing the specific features of the image.

[0031] As an embodiment of the present invention, the calculating the feature loss of the multi-frame images based on the number of feature points includes: calculating the feature values ​​of the multi-frame images, and calculating the feature loss of the loss image based on the feature values ​​and the number of feature points using the following formula: ; in, represents feature loss, n represents the number of feature points of the loss image, represents the feature matching function, Represents the matrix space of multiple frames of images, A represents the image features, Represents the eigenvalue.

[0032] Optionally, the number of feature points of the multi-frame images can be identified through an edge detection algorithm, and the process of calculating the eigenvalues ​​of the multi-frame images is: identifying the image features of the multi-frame images, constructing a matrix space of the multi-frame images, mapping the image features to the matrix space, and obtaining the eigenvalues.

[0033] Furthermore, the embodiment of the present invention performs image restoration on the multiple frames of images based on the pixel loss and the feature loss to obtain a restored image, which can improve image quality and make the detection of target points in the image more accurate.

[0034] As an embodiment of the present invention, performing image restoration on the multiple frames of images based on the pixel loss and the feature loss to obtain the restored images includes: Querying the loss image containing pixel loss and feature loss in the multiple frames of images, According to the feature loss and the pixel loss, the loss image is repaired using the following formula to obtain a repaired image: ; in, represents the repaired image, k represents the number of pixels before the loss image is processed, y represents the number of pixels after the multi-frame image is processed, a represents the feature value before the multi-frame image is processed, and b represents the feature value after the multi-frame image is processed. , Represents the feature compensation function of multiple frames of images, Represents the feature loss of multiple frames of images, Represents the pixel compensation function of multiple frame images.

[0035] S3, querying the tracking target in the repaired image, and dynamically marking the tracking target to obtain a dynamically marked target, querying the continuous associated frames of the dynamically marked target, and dynamically updating the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target.

[0036] The embodiment of the present invention queries the tracking target in the repaired image and dynamically marks the tracking target to obtain the dynamically marked target, which can locate the position of the target to be analyzed in the video in real time and avoid loss of the analysis target.

[0037] As an embodiment of the present invention, dynamically marking the tracking target to obtain the dynamically marked target includes: calculating the pixel value of the tracking target, identifying the characteristic direction of the tracking target, locating the tracking target based on the pixel value and the characteristic direction to obtain the located target, constructing a positioning identifier of the located target, and dynamically marking the tracking target based on the positioning identifier to obtain the dynamically marked target.

[0038] The characteristic direction refers to the main direction or orientation of the characteristic point in the image or data, and the positioning identifier refers to a special code used to mark the target.

[0039] Optionally, the pixel value of the tracking target can be calculated by a gradient function, the feature direction of the tracking target is obtained by identifying a local feature descriptor of the tracking target, and the target positioning of the tracking target based on the pixel value and the feature direction can be obtained by inputting parameter values ​​corresponding to the pixel value and the feature direction into a preconfigured Harris corner detection algorithm and calculating using the Harris corner detection algorithm.

[0040] Furthermore, the embodiment of the present invention can understand the motion trajectory and speed of the determined target of the dynamically marked target by querying the continuous associated frames of the dynamically marked target, thereby achieving tracking of the target.

[0041] The associated frames refer to the frames in which the dynamic marked targets appear continuously.

[0042] Furthermore, the embodiment of the present invention dynamically updates the dynamically marked target based on the continuous associated frames, and obtains the dynamically updated target, so as to understand the change of the tracked target within a certain period of time, such as the speed change of a car within a certain period of time.

[0043] As an embodiment of the present invention, dynamically updating the dynamically marked target based on the continuously associated frames to obtain the dynamically updated target includes: constructing a movement trajectory of the dynamically marked target based on the continuously associated frames, analyzing the speed change of the dynamically marked target based on the movement trajectory, and querying the position change of the dynamically marked target, and dynamically updating the dynamically marked target based on the speed change and the position change to obtain the dynamically updated target.

[0044] Optionally, the movement trajectory of the dynamically marked target is constructed based on the continuously associated frames by detecting the position and bounding box of the dynamically marked target in each frame of the continuously associated frames to associate the position of the dynamically marked target in the continuous frames, and the movement trajectory of the dynamically marked target is constructed according to the position of the target in the continuous frames.

[0045] Optionally, the dynamically marked target is dynamically updated based on the speed change and the position change to obtain the dynamically updated target: using a sensor to obtain the speed change and the position change information in real time to determine the movement trend of the dynamically marked target, and dynamically updating the dynamically marked target based on the movement trend.

[0046] Further, in another optional embodiment of the present invention, analyzing the speed change of the dynamically marked target based on the action trajectory includes: querying the motion-related frame number of the dynamically marked target based on the action trajectory, and calculating the speed change of the dynamically marked target based on the motion-related frame number using the following formula: ; in, Indicates speed change, Indicates the horizontal speed of the dynamic marker target. represents the vertical speed of the dynamic marker target, N represents the number of motion-related frames, t represents time, Indicates the horizontal position of the dynamic marking target at the i-th moment, Indicates Dynamically mark the horizontal position of the target at all times. Indicates the vertical position of the dynamic marking target at the i-th moment, Indicates Dynamically mark the target's vertical position at all times.

[0047] S4. Identify the appearance features of the tracking target, perform feature encoding on the appearance features to obtain encoding features, and perform feature association on the dynamic target based on the encoding features to obtain target association features.

[0048] As an embodiment of the present invention, the identifying the appearance features of the tracking target includes: identifying a feature descriptor of the tracking target, performing feature matching on the tracking target based on the feature descriptor to obtain matching information, querying feature points of the tracking target, using the feature points to determine target information of the matching information, and identifying the appearance features of the tracking target based on the target information.

[0049] The feature descriptor refers to a special code, which is a method used in the computer field to describe image features.

[0050] Optionally, the feature descriptor for identifying the tracking target is calculated by a descriptor calculation function in a computer vision library (such as opencv), and the process of obtaining matching information by performing feature matching on the tracking target based on the feature descriptor is as follows: converting the feature descriptor into a vector form, and inputting it into a preconfigured deep learning model, using the fully connected layer in the deep learning model to perform matching classification to obtain matching information, and the appearance features of the tracking target identified based on the target information can be obtained by compiling information based on the target information through feature engineering.

[0051] Furthermore, the embodiment of the present invention can convert the appearance features into a suitable coding form by encoding the appearance features to obtain the coded features, which can be better understood and processed by the machine learning model, thereby improving the prediction accuracy of the model. The features in the data are converted into a form suitable for the machine learning model.

[0052] Optionally, the appearance features are feature encoded to obtain the encoded features through label encoding.

[0053] Furthermore, the embodiment of the present invention can more accurately identify the dynamic target by performing feature association on the dynamic target based on the coding feature to obtain the target association feature, thereby achieving continuous tracking of the dynamic target.

[0054] Optionally, the process of performing feature association on the dynamic target based on the coding feature to obtain the target association feature is: extracting the coding feature from the dynamic target, inputting the coding feature into a fully connected layer in a preconfigured deep learning model to use the coding feature to match the features of other targets, performing association analysis based on the feature matching results, determining the degree of association between the targets, and generating the target association feature based on the association analysis results.

[0055] S5. Based on the dynamically updated target and the target association characteristics, a dynamic analysis strategy for the dynamic target is constructed, the dynamic analysis strategy is used to simulate tracking of the dynamic target, and a simulation score of the simulated tracking is calculated. When the simulation score meets a preset score value, the dynamic analysis strategy is used to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed, and a target analysis report is obtained.

[0056] The embodiment of the present invention can realize real-time tracking and monitoring of the target by constructing the dynamic analysis strategy of the dynamic target based on the dynamic update target and the target association characteristics through real-time analysis of the dynamic target and its associated characteristics. For example, in a school sports meet, the athlete's movement status, position and other information can be tracked, abnormal situations can be discovered in time and corresponding measures can be taken. Among them, the dynamic analysis strategy refers to a strategy that uses specific methods and means to track, monitor, analyze and evaluate the target in real time based on the dynamic update target and the target association characteristics.

[0057] Optionally, the process of constructing a dynamic analysis strategy for the dynamic target based on the dynamically updated target and the target-associated characteristics is: determining the object to be analyzed based on the dynamically updated target and the target-associated characteristics, querying the feature information of the object to be analyzed to construct relationship information between the dynamic target and the target-associated characteristics, and performing real-time analysis using a preconfigured deep learning model based on the relationship information.

[0058] Furthermore, the embodiment of the present invention can determine the performance of the strategy in different situations by using the dynamic analysis strategy to simulate tracking of the dynamic target, and then adjust the strategy to improve its performance.

[0059] Furthermore, the embodiment of the present invention can analyze the effect and accuracy of the simulated tracking by calculating the simulated score of the simulated tracking, and then find out the shortcomings in the tracking process.

[0060] As an embodiment of the present invention, the calculation of the simulation score of the simulated tracking includes: extracting multiple frame associated images of the dynamic target, tracking the corresponding dynamic target in the multiple frame associated images using a preconfigured depth model, and calculating the target tracking score using the following formula: ; Among them, SS represents the target tracking score, represents the number of false positives in the kth frame, represents the number of false negatives in the kth frame, represents the number of identity switching times in the kth frame, Indicates the number of true bounding boxes in the kth frame; Based on the target tracking score, a simulation score of the simulated tracking is determined.

[0061] Among them, the number of false positives refers to the number of samples that are mistakenly judged as positive (i.e., the presence of a certain feature or situation) in a detection or classification task; the number of false negatives refers to the situation where a certain situation or feature actually exists but is mistakenly judged as not existing. For example, in machine learning, a deep learning model mistakenly predicts a sample that should belong to a certain category as not belonging to that category; the number of identity switches refers to the total number of identity switches that occur during the tracking process (the number of target losses); the number of real bounding boxes refers to the precise annotation of the actual position and range of a target object in an image or video in computer vision.

[0062] It should be noted that when the simulation score meets the preset score value, it means that the target tracking score is not less than 0.9 (it can also be set according to the actual application scenario), which means that the accuracy of the model analysis is relatively high.

[0063] Furthermore, the embodiment of the present invention utilizes the dynamic analysis strategy to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed, and obtains a target analysis report that can intelligently analyze the situation in the camera area and track the target to be analyzed in real time to deal with sudden situations or anomalies.

[0064] like Figure 2 The figure shows a functional module diagram of a system for intelligent analysis of the camera area based on deep learning according to the present invention.

[0065] The intelligent analysis system 200 for realizing the camera area based on deep learning described in the present invention can be installed in an electronic device. According to the functions implemented, the intelligent analysis system for realizing the camera area based on deep learning can include a target extraction module 201, an image repair module 202, a target update module 203, a target association module 204 and an intelligent analysis module 205. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, which are stored in the memory of the electronic device.

[0066] In the embodiment of the present invention, the functions of each module / unit are as follows: The target extraction module 201 is used to collect the camera video of the camera area to be analyzed, identify the dynamic target in the camera video, and extract the target video in the camera video based on the dynamic target; The image restoration module 202 is used to perform video framing on the target video to obtain multiple frames of images, calculate pixel losses of the multiple frames of images, identify feature points of the multiple frames of images, calculate feature losses of the multiple frames of images based on the feature points, and perform image restoration on the multiple frames of images based on the pixel losses and the feature losses to obtain restored images; The target updating module 203 is used to query the tracking target in the repaired image, dynamically mark the tracking target to obtain a dynamically marked target, query the continuous associated frames of the dynamically marked target, and dynamically update the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target; The target association module 204 is used to identify the appearance features of the tracking target, perform feature encoding on the appearance features to obtain encoding features, and perform feature association on the dynamic target based on the encoding features to obtain target association features; The intelligent analysis module 205 is used to construct a dynamic analysis strategy for the dynamic target based on the dynamically updated target and the target association characteristics, use the dynamic analysis strategy to simulate tracking of the dynamic target, calculate the simulation score of the simulated tracking, and when the simulation score meets the preset score value, use the dynamic analysis strategy to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed, and obtain a target analysis report.

[0067] In detail, the modules in the system 200 for realizing intelligent analysis of the camera area based on deep learning in the embodiment of the present invention are used in the same manner as described above. Figure 1 The technical means described in the article are similar to the method for realizing intelligent analysis of the camera area based on deep learning, and can produce the same technical effects, so they will not be repeated here.

[0068] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0069] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for intelligent analysis of camera areas based on deep learning, characterized in that: The method comprises: Collecting a camera video of a camera area to be analyzed, identifying a dynamic target in the camera video, and extracting a target video in the camera video based on the dynamic target; Performing video framing on the target video to obtain multiple frame images, calculating pixel losses of the multiple frame images, identifying feature points of the multiple frame images, and calculating feature losses of the multiple frame images based on the feature points, and performing image restoration on the multiple frame images based on the pixel losses and the feature losses to obtain restored images; Querying a tracking target in the repaired image, dynamically marking the tracking target to obtain a dynamically marked target, querying continuous associated frames of the dynamically marked target, and dynamically updating the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target; Identifying appearance features of the tracking target, encoding the appearance features to obtain encoding features, and performing feature association on the dynamic target based on the encoding features to obtain target association features; Based on the dynamically updated target and the target association characteristics, a dynamic analysis strategy for the dynamic target is constructed, the dynamic analysis strategy is used to simulate tracking of the dynamic target, and a simulation score of the simulated tracking is calculated. When the simulation score meets a preset score value, the dynamic analysis strategy is used to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed to obtain a target analysis report.

2. The method according to claim 1, characterized in that The calculating pixel loss of the multiple frames of images comprises: The pixel loss of the multi-frame images is calculated using the following formula: ； in represents pixel loss, Indicates the image length of multiple frames of images, Indicates the image width of the multi-frame image, n indicates the original pixel position of the multi-frame image, and m indicates the pixel position of the multi-frame image after framing. Represents the pixel loss of multiple frames of images, Represents the image edge loss of multiple frames.

3. The method according to claim 1, characterized in that The calculating the feature loss of the multiple frames of images based on the number of feature points includes: Calculating characteristic values ​​of the multiple frames of images; Based on the feature value and the number of feature points, the feature loss of the loss image is calculated using the following formula: ； in, represents feature loss, n represents the number of feature points of the loss image, represents the feature matching function, Represents the matrix space of multiple frames of images, Represents image features, Represents the eigenvalue.

4. The method according to claim 1, characterized in that: The performing image restoration on the multiple frames of images based on the pixel loss and the feature loss to obtain a restored image includes: Querying the loss image including pixel loss and feature loss in the multiple frames of images; According to the feature loss and the pixel loss, the loss image is repaired using the following formula to obtain a repaired image: ； in, represents the repaired image, k represents the number of pixels before the loss image is processed, y represents the number of pixels after the multi-frame image is processed, a represents the feature value before the multi-frame image is processed, and b represents the feature value after the multi-frame image is processed. , Represents the feature compensation function of multiple frames of images, Represents the feature loss of multiple frames of images, Represents the pixel compensation function of multiple frame images.

5. The method according to claim 1, characterized in that The dynamically marking the tracking target to obtain the dynamically marked target includes: Calculating the pixel value of the tracking target; Identifying a characteristic direction of the tracking target; Performing target positioning on the tracking target based on the pixel value and the characteristic direction to obtain a positioning target; A positioning identifier of the positioning target is constructed, and the tracking target is dynamically marked based on the positioning identifier to obtain a dynamically marked target.

6. The method according to claim 1, characterized in that The dynamically updating the dynamically marked target based on the continuous associated frames to obtain the dynamically updated target includes: Based on the continuous associated frames, constructing the movement trajectory of the dynamically marked target; Based on the motion trajectory, analyzing the speed change of the dynamically marked target, and querying the position change of the dynamically marked target; The dynamically marked target is dynamically updated based on the speed change and the position change to obtain a dynamically updated target.

7. The method according to claim 6, characterized in that The analyzing the speed change of the dynamically marked target based on the action trajectory includes: Based on the motion trajectory, query the motion-related frame number of the dynamically marked target; Based on the motion-related frame number, the speed change of the dynamic marking target is calculated using the following formula: ； in, Indicates speed change, Indicates the horizontal speed of the dynamic marker target. represents the vertical speed of the dynamic marker target, N represents the number of motion-related frames, t represents time, Indicates the horizontal position of the dynamic marking target at the i-th moment, Indicates Dynamically mark the horizontal position of the target at all times. Indicates the vertical position of the dynamic marking target at the i-th moment, Indicates Dynamically mark the target's vertical position at all times.

8. The method according to claim 1, characterized in that The identifying the appearance features of the tracking target includes: identifying a feature descriptor of the tracking target; Based on the feature descriptor, feature matching is performed on the tracking target to obtain matching information; Querying feature points of the tracking target; The target information of the matching information is determined by using the feature points, and the appearance features of the tracking target are identified based on the target information.

9. The method according to claim 1, characterized in that: The calculating of the simulation score of the simulation tracking comprises: Extracting multiple frames of associated images of the dynamic target; The preconfigured depth model is used to track the corresponding dynamic targets in the multi-frame associated images, and the target tracking score is calculated using the following formula: ； Among them, SS represents the target tracking score, represents the number of false positives in the kth frame, represents the number of false negatives in the kth frame, represents the number of identity switching times in the kth frame, Indicates the number of true bounding boxes in the kth frame; Based on the target tracking score, a simulation score of the simulated tracking is determined.

10. A system for intelligent analysis of camera areas based on deep learning, characterized in that: The system comprises: The target extraction module is used to collect the camera video of the camera area to be analyzed, identify the dynamic target in the camera video, and extract the target video in the camera video based on the dynamic target; An image restoration module, configured to perform video framing on the target video to obtain multiple frames of images, calculate pixel losses of the multiple frames of images, identify feature points of the multiple frames of images, calculate feature losses of the multiple frames of images based on the feature points, and perform image restoration on the multiple frames of images based on the pixel losses and the feature losses to obtain restored images; a target updating module, configured to query a tracking target in the repaired image, dynamically mark the tracking target to obtain a dynamically marked target, query continuous associated frames of the dynamically marked target, and dynamically update the dynamically marked target based on the continuous associated frames to obtain a dynamically updated target; A target association module, used for identifying the appearance features of the tracking target, performing feature encoding on the appearance features to obtain encoding features, and performing feature association on the dynamic target based on the encoding features to obtain target association features; The intelligent analysis module is used to construct a dynamic analysis strategy for the dynamic target based on the dynamically updated target and the target association characteristics, use the dynamic analysis strategy to simulate tracking of the dynamic target, calculate a simulation score of the simulated tracking, and when the simulation score meets a preset score value, use the dynamic analysis strategy to perform real-time intelligent analysis on the target to be analyzed in the camera area to be analyzed, and obtain a target analysis report.

Images