A data collection and analysis method and system for law enforcement recorders

Abstract

The present invention discloses a data acquisition and analysis method and system for law enforcement recorders. The law enforcement recorder acquires video in real time, and extracts an image set after preprocessing. The image recognition model identifies the scene in combination with positioning information. Images are acquired at preset intervals, people are identified and marked, and the fuzziness of the person information is calculated. If the fuzziness does not meet the standard, the defective area is identified, and the GAN model and reference features are used to supplement the features and identify the person again. The person and scene information are integrated and transmitted to the central scheduling platform in real time. Through the present invention, the law enforcement recorder can achieve high adaptability to the environment, effectively increase the intelligence of the law enforcement recording process, the accuracy of data, and the efficiency of data acquisition.

Description

Technical Field

[0001] The present invention relates to the field of video data processing, and more specifically, to a data acquisition and analysis method and system for law enforcement recorders. Background Art

[0002] With the advancement of science and technology and the development of society, law enforcement video plays an increasingly important role in judicial practice. However, traditional law enforcement video analysis methods mostly rely on manual viewing and analysis, which is inefficient and prone to errors.

[0003] Moreover, restricted by traditional technology, the data collection process of law enforcement recorders often results in inaccurate recognition and deviations between scene recognition information and actual conditions, which greatly reduces the efficiency of law enforcement by central dispatch. Summary of the invention

[0004] The present invention overcomes the defects of the prior art and proposes a data collection and analysis method and system for law enforcement recorders.

[0005] The first aspect of the present invention provides a data collection and analysis method for a law enforcement recorder, comprising:

[0006] Obtain law enforcement video data in real time based on law enforcement recorders and pre-process the law enforcement video data;

[0007] Extracting an image set based on the law enforcement video data to obtain a real-time image set, performing scene recognition based on an image recognition model, and generating real-time scene information in combination with positioning information;

[0008] Based on the preset acquisition interval, the real-time image set is acquired, the person region is marked and the person is recognized on the acquired image, the corresponding person recognition result is obtained, the difference of the person recognition results at different acquisition intervals is determined, and the fuzziness of the person information is calculated;

[0009] Determine whether the blurriness meets expectations. If not, identify defective areas based on the captured images, obtain reference personnel features from the system database, build a GAN-based generative model, train the model based on reference personnel features, import the captured images into the generative model, and generate and supplement features based on defective area simulation features to form complete captured image features. Feature supplementation is performed on captured images at different acquisition intervals.

[0010] Based on the complete captured image features, secondary person recognition is performed on the captured images at different capture intervals, and information is extracted from the obtained person recognition results to form real-time person information;

[0011] The real-time character information and real-time scene information are integrated to obtain scene recognition data, and the scene recognition data is transmitted to the central dispatching platform in real time.

[0012] In this solution, the law enforcement video data is obtained in real time based on the law enforcement recorder, and the law enforcement video data is preprocessed, specifically:

[0013] Collect corresponding law enforcement video data in real time through law enforcement recorders;

[0014] The law enforcement video data is segmented and preprocessed for data enhancement.

[0015] In this solution, the image set is extracted based on the law enforcement video data to obtain a real-time image set, scene recognition is performed based on the image recognition model, and real-time scene information is generated in combination with positioning information, specifically:

[0016] Extract key frames from law enforcement video data to form a real-time image set;

[0017] Construct an image recognition model based on SVM, import the real-time image set into the image recognition model to perform scene area extraction and scene recognition, and obtain the scene recognition result;

[0018] Obtain positioning information through law enforcement recorders, and extract building information and environmental information within a preset range based on the networked map model using the positioning information as a reference point to obtain positioning scene information;

[0019] The scene recognition results and the positioning scene information are used to perform scene matching and scene information confirmation to form real-time scene information.

[0020] In this solution, based on the preset acquisition interval, the real-time image set is acquired, the person region is marked and the person is recognized on the acquired image, the corresponding person recognition result is obtained, the difference of the person recognition results at different acquisition intervals is judged and the fuzziness of the person information is calculated, specifically:

[0021] Based on a preset acquisition interval, the real-time image set is acquired to obtain acquired images at different acquisition intervals;

[0022] Importing the collected image into the image recognition model to identify the person area and extract the person features, and obtain the person area information and person feature data;

[0023] Associating the character area information with the scene information of the corresponding picture, importing the character feature data into the image recognition model for character recognition, and obtaining the character recognition result;

[0024] Determine the differences in person recognition results at different acquisition intervals and calculate the ambiguity of person information.

[0025] In this solution, the fuzziness is judged to be in line with expectations. If not, defective areas are identified based on the acquired images, reference personnel features are obtained from the system database, a generation model based on GAN is constructed, model training is performed based on reference personnel features, the acquired images are imported into the generation model, and feature generation and feature supplementation are performed based on defective area simulation features to form complete acquired image features. Feature supplementation is performed on acquired images at different acquisition intervals, specifically:

[0026] Based on the preset threshold range, determine whether the blurriness meets expectations;

[0027] If not, the character feature completeness is judged based on the character region information and character feature data of the collected image, and the region with feature defects is marked to form a character defect region;

[0028] Obtain relevant personnel information based on law enforcement tasks, and extract relevant personnel image data corresponding to the relevant personnel information from the system database;

[0029] Extracting image features from the image data of the relevant persons to form reference person features;

[0030] Build a GAN-based generative model, which includes a generator and a discriminator and sets a loss function;

[0031] Import the reference person characteristics as real data into the generative model, perform cyclic adversarial neural training based on the generator and the discriminator, and train until the discriminator and the generator reach Nash equilibrium, completing the training process;

[0032] Take a captured image as the analysis unit, import the corresponding character feature data as the basic feature data into the generation model, take the character defect area as the target area for feature generation, simulate and generate supplementary feature data through the generator, integrate the supplementary feature data with the basic feature data into character features, and use the discriminator to discriminate the results. If the discrimination result is false, the generator's simulated feature generation and character feature discrimination are repeated until it passes the discriminator, and the integrated character features are used as the complete captured image features.

[0033] Corresponding feature supplementary analysis is performed on the images collected at different collection intervals.

[0034] In this solution, based on the complete collected image features, secondary person recognition is performed on the collected images at different collection intervals, and information is extracted from the obtained person recognition results to form real-time person information, specifically:

[0035] The complete collected image features corresponding to the collected images at different collection intervals are imported into the image recognition model for secondary recognition, and the fuzziness calculation and analysis is performed based on the recognition results;

[0036] The secondary character recognition results are then used for information extraction to form real-time character information.

[0037] The second aspect of the present invention further provides a data collection and analysis system for a law enforcement recorder, the system comprising: a memory, a processor, the memory comprising a data collection and analysis program for a law enforcement recorder, the data collection and analysis program for a law enforcement recorder being executed by the processor to implement the following steps:

[0038] Obtain law enforcement video data in real time based on law enforcement recorders and pre-process the law enforcement video data;

[0039] Extracting an image set based on the law enforcement video data to obtain a real-time image set, performing scene recognition based on an image recognition model, and generating real-time scene information in combination with positioning information;

[0040] Based on the preset acquisition interval, the real-time image set is acquired, the person region is marked and the person is recognized on the acquired image, the corresponding person recognition result is obtained, the difference of the person recognition results at different acquisition intervals is determined, and the fuzziness of the person information is calculated;

[0041] Determine whether the blurriness meets expectations. If not, identify defective areas based on the captured images, obtain reference personnel features from the system database, build a GAN-based generative model, train the model based on reference personnel features, import the captured images into the generative model, and generate and supplement features based on defective area simulation features to form complete captured image features. Feature supplementation is performed on captured images at different acquisition intervals.

[0042] Based on the complete captured image features, secondary person recognition is performed on the captured images at different capture intervals, and information is extracted from the obtained person recognition results to form real-time person information;

[0043] The real-time character information and real-time scene information are integrated to obtain scene recognition data, and the scene recognition data is transmitted to the central dispatching platform in real time.

[0044] The third aspect of the present invention also provides a computer-readable storage medium, which includes a data acquisition and analysis program for a law enforcement recorder. When the data acquisition and analysis program for a law enforcement recorder is executed by a processor, the steps of the data acquisition and analysis method for a law enforcement recorder as described in any one of the above items are implemented.

[0045] The present invention discloses a data acquisition and analysis method and system for law enforcement recorders. The law enforcement recorder acquires video in real time, and extracts an image set after preprocessing. The image recognition model identifies the scene in combination with positioning information. Images are acquired at preset intervals, people are identified and marked, and the fuzziness of the person information is calculated. If the fuzziness does not meet the standard, the defective area is identified, and the GAN model and reference features are used to supplement the features and identify the person again. The person and scene information are integrated and transmitted to the central scheduling platform in real time. Through the present invention, the law enforcement recorder can achieve high adaptability to the environment, effectively increase the intelligence of the law enforcement recording process, the accuracy of data, and the efficiency of data acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flow chart of a data collection and analysis method for a law enforcement recorder of the present invention is shown;

[0047] Figure 2 A block diagram of a data acquisition and analysis system for a law enforcement recorder according to the present invention is shown. DETAILED DESCRIPTION

[0048] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0050] Figure 1 A flow chart of a data collection and analysis method for a law enforcement recorder of the present invention is shown.

[0051] like Figure 1 As shown, the first aspect of the present invention provides a data collection and analysis method for a law enforcement recorder, comprising:

[0052] S102, obtaining law enforcement video data in real time based on the law enforcement recorder, and preprocessing the law enforcement video data;

[0053] S104, extracting an image set based on the law enforcement video data to obtain a real-time image set, performing scene recognition based on an image recognition model, and generating real-time scene information in combination with positioning information;

[0054] S106, based on a preset acquisition interval, performing image acquisition on the real-time image set, performing person region identification and person recognition on the acquired image, obtaining corresponding person recognition results, determining the difference of person recognition results at different acquisition intervals and calculating the fuzziness of person information;

[0055] S108, judging whether the blur meets expectations, if not, identifying defective areas based on the acquired images, obtaining reference personnel features from the system database, building a generative model based on GAN, training the model based on the reference personnel features, importing the acquired images into the generative model, and generating and supplementing features based on simulated features of the defective areas to form complete acquired image features, and performing feature supplementation on acquired images at different acquisition intervals;

[0056] S110, performing secondary person recognition on the collected images at different collection intervals based on the complete collected image features, and extracting information from the obtained person recognition results to form real-time person information;

[0057] S112, integrating the real-time person information and the real-time scene information to obtain scene recognition data, and transmitting the scene recognition data to the central scheduling platform in real time.

[0058] It should be noted that the real-time image set is an image data set sorted based on the time dimension.

[0059] According to an embodiment of the present invention, the real-time acquisition of law enforcement video data based on the law enforcement recorder and preprocessing of the law enforcement video data are specifically as follows:

[0060] Collect corresponding law enforcement video data in real time through law enforcement recorders;

[0061] The law enforcement video data is segmented and preprocessed for data enhancement.

[0062] It should be noted that data enhancement is usually performed at the single frame level.

[0063] According to an embodiment of the present invention, the image set is extracted based on the law enforcement video data to obtain a real-time image set, scene recognition is performed based on an image recognition model, and real-time scene information is generated in combination with positioning information, specifically:

[0064] Extract key frames from law enforcement video data to form a real-time image set;

[0065] Construct an image recognition model based on SVM, import the real-time image set into the image recognition model to perform scene area extraction and scene recognition, and obtain the scene recognition result;

[0066] Obtain positioning information through law enforcement recorders, and extract building information and environmental information within a preset range based on the networked map model using the positioning information as a reference point to obtain positioning scene information;

[0067] The scene recognition results and the positioning scene information are used to perform scene matching and scene information confirmation to form real-time scene information.

[0068] It should be noted that the real-time image set includes at least one image. The networked map model can use conventional map application software to analyze the scene data near the positioning point in real time under the networked condition, such as building information, store information, street information, etc., and then perform corresponding scene matching. It is worth mentioning that when performing video analysis scenes, since the law enforcement recorder is worn by law enforcement personnel or in a vehicle, the video clarity recorded is limited, and there may be image distortion and other situations. Therefore, the present invention can correct the erroneous scene information based on the recognition result by searching and extracting the positioning scene and matching it with the result of real-time video recognition, and the scene information based on video analysis can provide more accurate locations and scenes and upload them to the dispatch center to better analyze the real-time situation and support dispatch. Specific scene recognition is refined scenes such as shops, stalls, convenience stores, streets, communities, roads, etc. The SVM is a support vector machine. In image recognition tasks, SVM can be used for feature classification and recognition.

[0069] According to an embodiment of the present invention, based on a preset acquisition interval, image acquisition is performed on a real-time image set, person region identification and person recognition are performed on the acquired image, corresponding person recognition results are obtained, the difference of person recognition results at different acquisition intervals is determined, and the fuzziness of person information is calculated, specifically:

[0070] Based on a preset acquisition interval, the real-time image set is acquired to obtain acquired images at different acquisition intervals;

[0071] Importing the collected image into the image recognition model to identify the person area and extract the person features, and obtain the person area information and person feature data;

[0072] Associating the character area information with the scene information of the corresponding picture, importing the character feature data into the image recognition model for character recognition, and obtaining the character recognition result;

[0073] Determine the differences in person recognition results at different acquisition intervals and calculate the ambiguity of person information.

[0074] It should be noted that the association of the character area information with the scene information of the corresponding picture can perform fuzzy analysis on the characters in the same scene. In different collected images, due to the movement, vibration or poor ambient light of law enforcement personnel, there may be recognition deviations of the characters in the same scene. Therefore, in different collected images, the recognition of the same person may have different degrees of feature defects, etc., making it difficult to confirm the accurate information of the identified person. Therefore, the present invention collects images at different time intervals through the law enforcement recorder, identifies the character information, extracts the defective area, and generates simulated features based on the defective area to form feature data with higher integrity for identification.

[0075] It is worth mentioning that in actual law enforcement tasks, the identification and confirmation of relevant personnel often occur. For example, in the routine law enforcement inspection process of a store, it may involve the identification and confirmation of certain store clerks. For example, in the law enforcement process of crimes, it may involve the face recognition of offenders. Therefore, based on actual law enforcement tasks, the feature data of relevant persons (such as offenders) can be derived from the system data (generally facial features and body and limb feature data), and the relevant features can be trained based on the GAN model to obtain a model that can simulate and generate the character features of relevant personnel. In the subsequent image defect analysis, the relevant features can be supplemented by the generated model to more effectively identify people.

[0076] After supplementing the features, through fuzzy analysis and secondary recognition results, it is possible to more accurately determine whether the person is a related person (reference person), effectively confirm the person's information, provide effective law enforcement data support for the dispatch center, and realize information-based and efficient comprehensive law enforcement management.

[0077] It is worth mentioning that during the use of law enforcement recorders, due to the influence of actual work, such as the movement, vibration or poor ambient light of law enforcement officers, the obtained video data often appear distorted, blurred, and have low feature recognition, resulting in certain errors in the law enforcement information uploaded to the center. Especially for the person recognition process, the recognition result is more important;

[0078] The character recognition result includes the number of characters and the specific identity of the characters, and the collection interval is. The fuzziness calculation formula is as follows:

[0079] ;

[0080] represents the number of overlapping persons in the person recognition results between the i-th collected image and the i+1-th collected image, and its value represents an indicator of the consistency of the recognition results between the two collected images;

[0081] represents the number of people recognized in the i-th collected image, M represents the fuzziness, and N is the number of collected images, which is determined by the collection interval.

[0082] It can be seen that in multiple interval-collected images of the same scene, when the number of person recognitions varies greatly and the consistency of the recognition results decreases, the corresponding blurriness is larger. The blurriness reflects the credibility of real-time acquisition of video data for person recognition.

[0083] According to an embodiment of the present invention, the fuzziness is judged to be in accordance with expectations. If not, defective areas are identified based on the acquired images, reference personnel features are obtained from the system database, a generation model based on GAN is constructed, model training is performed based on the reference personnel features, the acquired images are imported into the generation model, and feature generation and feature supplementation are performed based on the simulated features of the defective areas to form complete acquired image features. Feature supplementation is performed on acquired images at different acquisition intervals, specifically:

[0084] Based on the preset threshold range, determine whether the blurriness meets expectations;

[0085] If not, the character feature completeness is judged based on the character region information and character feature data of the collected image, and the region with feature defects is marked to form a character defect region;

[0086] Obtain relevant personnel information based on law enforcement tasks, and extract relevant personnel image data corresponding to the relevant personnel information from the system database;

[0087] Extracting image features from the image data of the relevant persons to form reference person features;

[0088] Build a GAN-based generative model, which includes a generator and a discriminator and sets a loss function;

[0089] Import the reference person characteristics as real data into the generative model, perform cyclic adversarial neural training based on the generator and the discriminator, and train until the discriminator and the generator reach Nash equilibrium, completing the training process;

[0090] Take a captured image as the analysis unit, import the corresponding character feature data as the basic feature data into the generation model, take the character defect area as the target area for feature generation, simulate and generate supplementary feature data through the generator, integrate the supplementary feature data with the basic feature data into character features, and use the discriminator to discriminate the results. If the discrimination result is false, the generator's simulated feature generation and character feature discrimination are repeated until it passes the discriminator, and the integrated character features are used as the complete captured image features.

[0091] Corresponding feature supplementary analysis is performed on the images collected at different collection intervals.

[0092] It should be noted that the presence of characteristic defects, that is, the distortion and incomplete features of the characters in the captured image, etc., is analyzed by characteristic defects, and the corresponding defective areas are marked and simulated features are generated subsequently. The character area information includes the marked area of ​​the entire character in the captured image, and the character defect area is generally a partial area of ​​the entire character area, that is, a partial area with characteristic defects. Each captured image corresponds to an independent character defect area. In the captured images with different capture intervals, each captured image includes the corresponding complete captured image features.

[0093] According to an embodiment of the present invention, based on the complete captured image features, secondary person recognition is performed on the captured images at different capture intervals, and information is extracted from the obtained person recognition results to form real-time person information, specifically:

[0094] The complete collected image features corresponding to the collected images at different collection intervals are imported into the image recognition model for secondary recognition, and the fuzziness calculation and analysis is performed based on the recognition results;

[0095] The secondary character recognition results are then used for information extraction to form real-time character information.

[0096] It should be noted that the ambiguity of the recognition results after feature supplementation will be greatly reduced, effectively improving the accuracy of the person's recognition information.

[0097] According to an embodiment of the present invention, it also includes:

[0098] Perform real-time image information analysis on the collected images, including light intensity, contrast, person size, and visible range information, and form basic image information;

[0099] Integrate the defect areas of the characters in all the collected images to form defect area information;

[0100] Perform image environment defect analysis based on basic image information and generate image defect categories in real time;

[0101] Through the image defect category and defect area information, the control countermeasures analysis of the shooting defects of the law enforcement recorder is carried out, and the control countermeasures information is obtained;

[0102] Generate real-time control instructions based on control countermeasure information, and send the real-time control instructions to the law enforcement recorder through the Internet of Things;

[0103] In each analysis cycle, the collected images and control strategy information are analyzed in real time and the control instructions are updated periodically.

[0104] It should be noted that the image environment defect analysis based on the basic information of the image and the real-time generation of image defect categories specifically involves category marking based on the image defects existing in the images collected in the use environment of the law enforcement recorder, including image defects such as insufficient light, excessive contrast, and too small a proportion of people, and generating corresponding control measures for the defects. The control measures include image magnification and reduction, contrast adjustment, light adjustment, illumination adjustment of the law enforcement recorder, and image magnification and reduction are generally performed on defective areas. In the present invention, the law enforcement recorder includes at least one processor, a memory, and a video control system.

[0105] It is worth mentioning that during the use of law enforcement recorders, due to the influence of environmental and human factors, the recorded real-time video data often has some image defects, which greatly reduces the scene recognition rate and affects the collection of law enforcement information. For example, the proportion of people in the picture is too small, the ambient light is dim, etc. Therefore, the present invention conducts real-time analysis of defective areas and image defect categories, sets corresponding countermeasures, such as local image magnification, image contrast, light intensity adjustment, etc., forms corresponding control instructions through countermeasure analysis, and changes the control instructions in real time based on different time periods (or different analysis cycles), so that the law enforcement recorder can achieve high adaptability to the environment. This method is highly practical and effectively increases the intelligence of the law enforcement recording process, the accuracy of data, and the efficiency of data collection.

[0106] Figure 2 A block diagram of a data acquisition and analysis system for a law enforcement recorder according to the present invention is shown.

[0107] The second aspect of the present invention further provides a data collection and analysis system 2 for a law enforcement recorder, the system comprising: a memory 21, a processor 22, the memory comprising a data collection and analysis program for a law enforcement recorder, the data collection and analysis program for a law enforcement recorder being executed by the processor to implement the following steps:

[0108] Obtain law enforcement video data in real time based on law enforcement recorders and pre-process the law enforcement video data;

[0109] Extracting an image set based on the law enforcement video data to obtain a real-time image set, performing scene recognition based on an image recognition model, and generating real-time scene information in combination with positioning information;

[0110] Based on the preset acquisition interval, the real-time image set is acquired, the person region is marked and the person is recognized on the acquired image, the corresponding person recognition result is obtained, the difference of the person recognition results at different acquisition intervals is determined, and the fuzziness of the person information is calculated;

[0111] Determine whether the blurriness meets expectations. If not, identify defective areas based on the captured images, obtain reference personnel features from the system database, build a GAN-based generative model, train the model based on reference personnel features, import the captured images into the generative model, and generate and supplement features based on defective area simulation features to form complete captured image features. Feature supplementation is performed on captured images at different acquisition intervals.

[0112] Based on the complete captured image features, secondary person recognition is performed on the captured images at different capture intervals, and information is extracted from the obtained person recognition results to form real-time person information;

[0113] The real-time character information and real-time scene information are integrated to obtain scene recognition data, and the scene recognition data is transmitted to the central dispatching platform in real time.

[0114] It should be noted that the real-time image set is an image data set sorted based on the time dimension.

[0115] According to an embodiment of the present invention, the real-time acquisition of law enforcement video data based on the law enforcement recorder and preprocessing of the law enforcement video data are specifically as follows:

[0116] Collect corresponding law enforcement video data in real time through law enforcement recorders;

[0117] The law enforcement video data is segmented and preprocessed for data enhancement.

[0118] It should be noted that data enhancement is usually performed at the single frame level.

[0119] According to an embodiment of the present invention, the image set is extracted based on the law enforcement video data to obtain a real-time image set, scene recognition is performed based on an image recognition model, and real-time scene information is generated in combination with positioning information, specifically:

[0120] Extract key frames from law enforcement video data to form a real-time image set;

[0121] Construct an image recognition model based on SVM, import the real-time image set into the image recognition model to perform scene area extraction and scene recognition, and obtain the scene recognition result;

[0122] Obtain positioning information through law enforcement recorders, and extract building information and environmental information within a preset range based on the networked map model using the positioning information as a reference point to obtain positioning scene information;

[0123] The scene recognition results and the positioning scene information are used to perform scene matching and scene information confirmation to form real-time scene information.

[0124] It should be noted that the real-time image set includes at least one image. The networked map model can use conventional map application software to analyze the scene data near the positioning point in real time under the networked condition, such as building information, store information, street information, etc., and then perform corresponding scene matching. It is worth mentioning that when performing video analysis scenes, since the law enforcement recorder is worn by law enforcement personnel or in a vehicle, the video clarity recorded is limited, and there may be image distortion and other situations. Therefore, the present invention can correct the erroneous scene information based on the recognition result by searching and extracting the positioning scene and matching it with the result of real-time video recognition, and the scene information based on video analysis can provide more accurate locations and scenes and upload them to the dispatch center to better analyze the real-time situation and support dispatch. Specific scene recognition is refined scenes such as shops, stalls, convenience stores, streets, communities, roads, etc. The SVM is a support vector machine. In image recognition tasks, SVM can be used for feature classification and recognition.

[0125] According to an embodiment of the present invention, based on a preset acquisition interval, image acquisition is performed on a real-time image set, person region identification and person recognition are performed on the acquired image, corresponding person recognition results are obtained, the difference of person recognition results at different acquisition intervals is determined, and the fuzziness of person information is calculated, specifically:

[0126] Based on a preset acquisition interval, the real-time image set is acquired to obtain acquired images at different acquisition intervals;

[0127] Importing the collected image into the image recognition model to identify the person area and extract the person features, and obtain the person area information and person feature data;

[0128] Associating the character area information with the scene information of the corresponding picture, importing the character feature data into the image recognition model for character recognition, and obtaining the character recognition result;

[0129] Determine the differences in person recognition results at different acquisition intervals and calculate the ambiguity of person information.

[0130] It should be noted that the association of the character area information with the scene information of the corresponding picture can perform fuzzy analysis on the characters in the same scene. In different collected images, due to the movement, vibration or poor ambient light of law enforcement personnel, there may be recognition deviations of the characters in the same scene. Therefore, in different collected images, the recognition of the same person may have different degrees of feature defects, etc., making it difficult to confirm the accurate information of the identified person. Therefore, the present invention collects images at different time intervals through the law enforcement recorder, identifies the character information, extracts the defective area, and generates simulated features based on the defective area to form feature data with higher integrity for identification.

[0131] It is worth mentioning that in actual law enforcement tasks, the identification and confirmation of relevant personnel often occur. For example, in the routine law enforcement inspection process of a store, it may involve the identification and confirmation of certain store clerks. For example, in the law enforcement process of crimes, it may involve the face recognition of offenders. Therefore, based on actual law enforcement tasks, the feature data of relevant persons (such as offenders) can be derived from the system data (generally facial features and body and limb feature data), and the relevant features can be trained based on the GAN model to obtain a model that can simulate and generate the character features of relevant personnel. In the subsequent image defect analysis, the relevant features can be supplemented by the generated model to more effectively identify people.

[0132] After supplementing the features, through fuzzy analysis and secondary recognition results, it is possible to more accurately determine whether the person is a related person (reference person), effectively confirm the person's information, provide effective law enforcement data support for the dispatch center, and realize information-based and efficient comprehensive law enforcement management.

[0133] It is worth mentioning that during the use of law enforcement recorders, due to the influence of actual work, such as the movement, vibration or poor ambient light of law enforcement officers, the obtained video data often appear distorted, blurred, and have low feature recognition, resulting in certain errors in the law enforcement information uploaded to the center. Especially for the person recognition process, the recognition result is more important;

[0134] The character recognition result includes the number of characters and the specific identity of the characters, and the collection interval is. The fuzziness calculation formula is as follows:

[0135] ;

[0136] represents the number of overlapping persons in the person recognition results between the i-th collected image and the i+1-th collected image, and its value represents an indicator of the consistency of the recognition results between the two collected images;

[0137] represents the number of people recognized in the i-th collected image, M represents the fuzziness, and N is the number of collected images, which is determined by the collection interval.

[0138] It can be seen that in multiple interval-collected images of the same scene, when the number of person recognitions varies greatly and the consistency of the recognition results decreases, the corresponding blurriness is larger. The blurriness reflects the credibility of real-time acquisition of video data for person recognition.

[0139] According to an embodiment of the present invention, the fuzziness is judged to be in accordance with expectations. If not, defective areas are identified based on the acquired images, reference personnel features are obtained from the system database, a generation model based on GAN is constructed, model training is performed based on the reference personnel features, the acquired images are imported into the generation model, and feature generation and feature supplementation are performed based on the simulated features of the defective areas to form complete acquired image features. Feature supplementation is performed on acquired images at different acquisition intervals, specifically:

[0140] Based on the preset threshold range, determine whether the blurriness meets expectations;

[0141] If not, the character feature completeness is judged based on the character region information and character feature data of the collected image, and the region with feature defects is marked to form a character defect region;

[0142] Obtain relevant personnel information based on law enforcement tasks, and extract relevant personnel image data corresponding to the relevant personnel information from the system database;

[0143] Extracting image features from the image data of the relevant persons to form reference person features;

[0144] Build a GAN-based generative model, which includes a generator and a discriminator and sets a loss function;

[0145] Import the reference person characteristics as real data into the generative model, perform cyclic adversarial neural training based on the generator and the discriminator, and train until the discriminator and the generator reach Nash equilibrium, completing the training process;

[0146] Take a captured image as the analysis unit, import the corresponding character feature data as the basic feature data into the generation model, take the character defect area as the target area for feature generation, simulate and generate supplementary feature data through the generator, integrate the supplementary feature data with the basic feature data into character features, and use the discriminator to discriminate the results. If the discrimination result is false, the generator's simulated feature generation and character feature discrimination are repeated until it passes the discriminator, and the integrated character features are used as the complete captured image features.

[0147] Corresponding feature supplementary analysis is performed on the images collected at different collection intervals.

[0148] It should be noted that the presence of characteristic defects, that is, the distortion and incomplete features of the characters in the captured image, etc., is analyzed by characteristic defects, and the corresponding defective areas are marked and simulated features are generated subsequently. The character area information includes the marked area of ​​the entire character in the captured image, and the character defect area is generally a partial area of ​​the entire character area, that is, a partial area with characteristic defects. Each captured image corresponds to an independent character defect area. In the captured images with different capture intervals, each captured image includes the corresponding complete captured image features.

[0149] According to an embodiment of the present invention, based on the complete captured image features, secondary person recognition is performed on the captured images at different capture intervals, and information is extracted from the obtained person recognition results to form real-time person information, specifically:

[0150] The complete collected image features corresponding to the collected images at different collection intervals are imported into the image recognition model for secondary recognition, and the fuzziness calculation and analysis is performed based on the recognition results;

[0151] The secondary character recognition results are then used for information extraction to form real-time character information.

[0152] It should be noted that the ambiguity of the recognition results after feature supplementation will be greatly reduced, effectively improving the accuracy of the person's recognition information.

[0153] According to an embodiment of the present invention, it also includes:

[0154] Perform real-time image information analysis on the collected images, including light intensity, contrast, person size, and visible range information, and form basic image information;

[0155] Integrate the defect areas of the characters in all the collected images to form defect area information;

[0156] Perform image environment defect analysis based on basic image information and generate image defect categories in real time;

[0157] Through the image defect category and defect area information, the control countermeasures analysis of the shooting defects of the law enforcement recorder is carried out, and the control countermeasures information is obtained;

[0158] Generate real-time control instructions based on control countermeasure information, and send the real-time control instructions to the law enforcement recorder through the Internet of Things;

[0159] In each analysis cycle, the collected images and control strategy information are analyzed in real time and the control instructions are updated periodically.

[0160] It should be noted that the image environment defect analysis based on the basic information of the image and the real-time generation of image defect categories specifically involves category marking based on the image defects existing in the images collected in the use environment of the law enforcement recorder, including image defects such as insufficient light, excessive contrast, and too small a proportion of people, and generating corresponding control measures for the defects. The control measures include image magnification and reduction, contrast adjustment, light adjustment, illumination adjustment of the law enforcement recorder, and image magnification and reduction are generally performed on defective areas. In the present invention, the law enforcement recorder includes at least one processor, a memory, and a video control system.

[0161] It is worth mentioning that during the use of law enforcement recorders, due to the influence of environmental and human factors, the recorded real-time video data often has some image defects, which greatly reduces the scene recognition rate and affects the collection of law enforcement information. For example, the proportion of people in the picture is too small, the ambient light is dim, etc. Therefore, the present invention conducts real-time analysis of defective areas and image defect categories, sets corresponding countermeasures, such as local image magnification, image contrast, light intensity adjustment, etc., forms corresponding control instructions through countermeasure analysis, and changes the control instructions in real time based on different time periods (or different analysis cycles), so that the law enforcement recorder can achieve high adaptability to the environment. This method is highly practical and effectively increases the intelligence of the law enforcement recording process, the accuracy of data, and the efficiency of data collection.

[0162] The third aspect of the present invention also provides a computer-readable storage medium, which includes a data acquisition and analysis program for a law enforcement recorder. When the data acquisition and analysis program for a law enforcement recorder is executed by a processor, the steps of the data acquisition and analysis method for a law enforcement recorder as described in any one of the above items are implemented.

[0163] The present invention discloses a data acquisition and analysis method and system for law enforcement recorders. The law enforcement recorder acquires video in real time, and extracts an image set after preprocessing. The image recognition model identifies the scene in combination with positioning information. Images are acquired at preset intervals, people are identified and marked, and the fuzziness of the person information is calculated. If the fuzziness does not meet the standard, the defective area is identified, and the GAN model and reference features are used to supplement the features and identify the person again. The person and scene information are integrated and transmitted to the central scheduling platform in real time. Through the present invention, the law enforcement recorder can achieve high adaptability to the environment, effectively increase the intelligence of the law enforcement recording process, the accuracy of data, and the efficiency of data acquisition.

[0164] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0165] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0166] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0167] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0168] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0169] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A data collection and analysis method for law enforcement recorders, characterized in that: include: Obtain law enforcement video data in real time based on law enforcement recorders and pre-process the law enforcement video data; Extracting an image set based on the law enforcement video data to obtain a real-time image set, performing scene recognition based on an image recognition model, and generating real-time scene information in combination with positioning information; Based on the preset acquisition interval, the real-time image set is acquired, the person region is marked and the person is recognized on the acquired image, the corresponding person recognition result is obtained, the difference of the person recognition results at different acquisition intervals is determined, and the fuzziness of the person information is calculated; Determine whether the blurriness meets expectations. If not, identify defective areas based on the captured images, obtain reference personnel features from the system database, build a GAN-based generative model, train the model based on reference personnel features, import the captured images into the generative model, and generate and supplement features based on defective area simulation features to form complete captured image features. Feature supplementation is performed on captured images at different acquisition intervals. Based on the complete captured image features, secondary person recognition is performed on the captured images at different capture intervals, and information is extracted from the obtained person recognition results to form real-time person information; Integrate real-time character information and real-time scene information to obtain scene recognition data, and transmit the scene recognition data to the central dispatching platform in real time; The method includes performing image acquisition on a real-time image set based on a preset acquisition interval, performing person region identification and person recognition on the acquired image, obtaining corresponding person recognition results, judging the differences of person recognition results at different acquisition intervals, and calculating the fuzziness of person information, specifically: Based on a preset acquisition interval, the real-time image set is acquired to obtain acquired images at different acquisition intervals; Importing the collected image into the image recognition model to identify the person area and extract the person features, and obtain the person area information and person feature data; Associating the character area information with the scene information of the corresponding picture, importing the character feature data into the image recognition model for character recognition, and obtaining the character recognition result; Determine the differences in person recognition results at different acquisition intervals and calculate the ambiguity of person information; Among them, the judgment of whether the fuzziness meets expectations, if not, the defect area is identified based on the collected image, the reference personnel characteristics are obtained from the system database, a generation model based on GAN is constructed, the model is trained based on the reference personnel characteristics, the collected image is imported into the generation model, and the defect area simulation characteristics are generated and supplemented to form a complete collection image feature, and the collected images at different collection intervals are supplemented with features, specifically: Based on the preset threshold range, determine whether the blurriness meets expectations; If not, the completeness of the character features is judged based on the character region information and the character feature data of the collected image, and the region with feature defects is marked to form a character defect region; Obtain relevant personnel information based on law enforcement tasks, and extract relevant personnel image data corresponding to the relevant personnel information from the system database; Extracting image features from the image data of the relevant persons to form reference person features; Build a GAN-based generative model, which includes a generator and a discriminator and sets a loss function; Import the reference person characteristics as real data into the generative model, perform cyclic adversarial neural training based on the generator and the discriminator, and train until the discriminator and the generator reach Nash equilibrium, completing the training process; Take a captured image as the analysis unit, import the corresponding character feature data as the basic feature data into the generation model, take the character defect area as the target area for feature generation, simulate and generate supplementary feature data through the generator, integrate the supplementary feature data with the basic feature data into character features, and use the discriminator to discriminate the results. If the discrimination result is false, the generator's simulated feature generation and character feature discrimination are repeated until it passes the discriminator, and the integrated character features are used as the complete captured image features. Corresponding feature supplementary analysis is performed on the images collected at different collection intervals; The fuzziness calculation formula is as follows: ； represents the number of overlapping persons in the person recognition results between the i-th collected image and the i+1-th collected image, and its value represents an indicator of the consistency of the recognition results between the two collected images; represents the number of people recognized in the i-th collected image, represents the fuzziness, The number of collected images is determined by the collection interval.

2. The data collection and analysis method for a law enforcement recorder according to claim 1 is characterized in that: The real-time acquisition of law enforcement video data based on the law enforcement recorder and preprocessing of the law enforcement video data are specifically as follows: Collect corresponding law enforcement video data in real time through law enforcement recorders; The law enforcement video data is segmented and preprocessed for data enhancement.

3. The data collection and analysis method for a law enforcement recorder according to claim 2 is characterized in that: The image set is extracted based on the law enforcement video data to obtain a real-time image set, scene recognition is performed based on the image recognition model, and real-time scene information is generated in combination with positioning information, specifically: Extract key frames from law enforcement video data to form a real-time image set; Construct an image recognition model based on SVM, import the real-time image set into the image recognition model to extract scene areas and recognize scenes, and obtain scene recognition results; Obtain positioning information through law enforcement recorders, and extract building information and environmental information within a preset range based on the networked map model using the positioning information as a reference point to obtain positioning scene information; The scene recognition results and the positioning scene information are used to perform scene matching and scene information confirmation to form real-time scene information.

4. The data collection and analysis method for a law enforcement recorder according to claim 1 is characterized in that: Based on the complete collected image features, secondary person recognition is performed on the collected images at different collection intervals, and information is extracted from the obtained person recognition results to form real-time person information, specifically: The complete collected image features corresponding to the collected images at different collection intervals are imported into the image recognition model for secondary recognition, and the fuzziness calculation and analysis is performed based on the recognition results; The secondary character recognition results are then used for information extraction to form real-time character information.

5. A data collection and analysis system for law enforcement recorders, characterized in that: The system includes: a memory and a processor, wherein the memory includes a data acquisition and analysis program for a law enforcement recorder, and when the data acquisition and analysis program for a law enforcement recorder is executed by the processor, the steps of the data acquisition and analysis method for a law enforcement recorder as described in any one of claims 1 to 4 are implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a data acquisition and analysis program for a law enforcement recorder. When the data acquisition and analysis program for a law enforcement recorder is executed by a processor, the steps of the data acquisition and analysis method for a law enforcement recorder as described in any one of claims 1 to 4 are implemented.

Images