Smart identification method and system for law enforcement recorders

Abstract

The application discloses a law enforcement recorder intelligent identification method and system, relates to the field of videos, and solves the problem of poor identification effect of the existing law enforcement recorder intelligent identification method, and comprises the following steps: S1: acquiring a recorder real-time analysis period, acquiring recorder law enforcement period division data and a characteristic law enforcement image quantity ratio by analyzing law enforcement videos in the recorder real-time analysis period, and obtaining recorder image identification data; S2: performing real-time audio analysis on the law enforcement recorder according to the recorder image identification data, and acquiring an audio period abnormality index; and S3: monitoring and early warning the recorder real-time analysis period according to the recorder image identification data and the audio period abnormality index. The application can fully play the role of the law enforcement recorder and improve the law enforcement efficiency and safety of law enforcement personnel.

Description

Technical Field

[0001] This invention belongs to the field of video and relates to image recognition technology, specifically a method and system for intelligent recognition of law enforcement recorders. Background Technology

[0002] The intelligent recognition method of law enforcement recorders has the following specific shortcomings in the law enforcement process:

[0003] 1. Existing intelligent recognition methods for law enforcement recorders cannot process the image data recorded by the recorders in real time, and cannot identify potential risks or dangerous situations in real time. This can easily put law enforcement officers in a passive state when responding, increasing the risk of performing their duties.

[0004] 2. Existing intelligent recognition methods for law enforcement recorders cannot simultaneously assess dangerous situations and issue alarms based on image and audio data recorded by the recorder in real time, thus failing to fully utilize the function of law enforcement recorders;

[0005] Therefore, we propose an intelligent recognition method and system for law enforcement recorders. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent recognition method and system for law enforcement recorders. This invention is based on acquiring the real-time analysis cycle of the recorder, acquiring multiple law enforcement monitoring images within that cycle, creating a law enforcement image recognition model to recognize these images, and dividing the real-time analysis cycle into a first type and a second type based on the recognition results. This yields the recorder's law enforcement cycle segmentation data, which is then analyzed to obtain the ratio of characteristic law enforcement images. This ratio and the segmentation data are defined as recorder image recognition data. Based on this image recognition data, voice acquisition is performed on the second type of law enforcement cycle to obtain multiple on-site audio recordings. A law enforcement audio recognition model is then created to identify these multiple on-site audio recordings to acquire multiple target law enforcement recorder audio recordings. By analyzing these target recordings, an audio cycle anomaly index corresponding to the second type of law enforcement cycle is obtained. Finally, the real-time analysis cycle of the recorder is monitored and alerted based on the recorder image recognition data and the audio cycle anomaly index.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent recognition method for law enforcement recorders, comprising the following specific steps:

[0008] Step S1: Obtain the real-time analysis cycle of the recorder, acquire multiple law enforcement monitoring images within the real-time analysis cycle of the recorder, create a law enforcement image recognition model to recognize the multiple law enforcement monitoring images, divide the real-time analysis cycle of the recorder into a first type of law enforcement cycle and a second type of law enforcement cycle based on the recognition results, obtain the law enforcement cycle division data of the recorder, analyze the second type of law enforcement cycle to obtain the ratio of the number of feature law enforcement images, and define the ratio of the number of feature law enforcement images and the law enforcement cycle division data of the recorder as the recorder image recognition data;

[0009] Step S2: Based on the image recognition data of the recorder, the second type of law enforcement cycle is voice acquired to obtain multiple law enforcement scene audios. A law enforcement audio recognition model is created to recognize multiple law enforcement scene audios to obtain multiple target law enforcement recorder audios. By analyzing the multiple target law enforcement recorder audios, the audio cycle anomaly index corresponding to the second type of law enforcement cycle is obtained.

[0010] Step S3: Monitor and issue early warnings for the real-time analysis cycle of the recorder based on the recorder's image recognition data and audio cycle anomaly index.

[0011] The intelligent recognition system for law enforcement recorders operates as follows:

[0012] Image recognition module: Used to acquire the real-time analysis cycle of the recorder, acquire multiple law enforcement monitoring images within the real-time analysis cycle of the recorder, create a law enforcement image recognition model to recognize the multiple law enforcement monitoring images, divide the real-time analysis cycle of the recorder into a first type of law enforcement cycle and a second type of law enforcement cycle based on the recognition results, obtain the law enforcement cycle division data of the recorder, analyze the second type of law enforcement cycle to obtain the ratio of the number of characteristic law enforcement images, and define the ratio of the number of characteristic law enforcement images and the law enforcement cycle division data of the recorder as the recorder image recognition data;

[0013] The speech recognition module is used to acquire speech data for the second type of law enforcement cycle based on the image recognition data of the recorder, obtain multiple law enforcement scene audios, create a law enforcement audio recognition model to recognize multiple law enforcement scene audios to obtain multiple target law enforcement recorder audios, and obtain the audio cycle anomaly index corresponding to the second type of law enforcement cycle by analyzing the audios of multiple target law enforcement recorders.

[0014] Intelligent early warning module: used to monitor and issue early warnings for the real-time analysis cycle of the recorder based on the recorder's image recognition data and audio cycle anomaly index.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0016] 1. This invention processes image data recorded by law enforcement recorders in real time by creating law enforcement image recognition models and law enforcement audio recognition models, which can effectively identify potential risks or dangerous situations and reduce the risks for law enforcement officers when performing their duties.

[0017] 2. This invention, by acquiring the abnormal index of audio cycle and comparing the number of characteristic law enforcement images recorded by the law enforcement recorder, simultaneously assesses dangerous situations and issues alarms using both image and audio data, thus fully leveraging the role of the law enforcement recorder. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0019] Figure 1 This is a diagram illustrating the implementation steps of the present invention;

[0020] Figure 2 This is an overall system block diagram of the present invention;

[0021] Figure 3 These are law enforcement images captured by the recorder in this invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1

[0024] Please see Figure 1 This invention provides a technical solution: an intelligent recognition method for law enforcement recorders, comprising the following specific steps:

[0025] Step S1: Obtain the real-time analysis cycle of the recorder, acquire multiple law enforcement monitoring images within the real-time analysis cycle of the recorder, create a law enforcement image recognition model to recognize the multiple law enforcement monitoring images, divide the real-time analysis cycle of the recorder into a first type of law enforcement cycle and a second type of law enforcement cycle based on the recognition results, obtain the law enforcement cycle division data of the recorder, analyze the second type of law enforcement cycle to obtain the ratio of the number of feature law enforcement images, and define the ratio of the number of feature law enforcement images and the law enforcement cycle division data of the recorder as the recorder image recognition data;

[0026] Step S1 further includes the following specific steps:

[0027] Step S11: When the law enforcement recorder is working, the time value corresponding to the current moment is marked as the first image feature time point, the time value corresponding to the feature monitoring period before the first image feature time point is marked as the second image feature time point, and the time period between the first image feature time point and the second image feature time point is marked as the recorder's real-time analysis cycle.

[0028] Step S12: Extract the law enforcement video images from the recorder frame by frame within the real-time analysis period of the recorder to obtain several law enforcement images from the recorder, and obtain the law enforcement image extraction data from the recorder.

[0029] Step S13: During the real-time analysis cycle of the recorder, randomly select several time points to obtain multiple video monitoring time points, and the time interval between each two consecutive video monitoring points is equal. Obtain the recorder law enforcement image corresponding to each video monitoring point by extracting data from the law enforcement image of the recorder, and obtain multiple law enforcement monitoring images.

[0030] Step S14: Acquire the number of law enforcement monitoring images within the real-time analysis period of the recorder to obtain the number of law enforcement images per period;

[0031] Step S15: Create a law enforcement image recognition model;

[0032] Step S15 further includes the following specific steps:

[0033] Step S151: Use big data crawling technology to obtain multiple law enforcement recorder images using law enforcement recorder images as keywords;

[0034] Step S152: Each law enforcement recorder image is manually labeled as a first type law enforcement recorder image, a second type law enforcement recorder image, and a third type law enforcement recorder image to obtain law enforcement recorder image labeling data;

[0035] Step S153: Divide the law enforcement recorder image labeling data into a law enforcement image training set and a law enforcement image test set according to the image training-test ratio;

[0036] Step S154: Create an image recognition model using an existing artificial intelligence platform, and train the image recognition model using a law enforcement image training set until each law enforcement recorder image in the law enforcement image training set is used to train the image recognition model once.

[0037] Step S155: Test the image recognition model using the law enforcement image test set and obtain the recognition accuracy. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model training is completed and the law enforcement image recognition model is obtained. When the recognition accuracy is less than the target recognition accuracy, continue to train the image recognition model using the law enforcement image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

[0038] Step S16: Use the law enforcement image recognition model to identify each law enforcement monitoring image in the real-time analysis period of the recorder, and divide the real-time analysis period of the recorder into the first type of law enforcement period and the second type of law enforcement period according to the recognition results to obtain the law enforcement period division data of the recorder.

[0039] Step S16 further includes the following specific steps:

[0040] Step S161: When the law enforcement image recognition model identifies any law enforcement monitoring image within the real-time analysis period of the recorder as a first-type law enforcement recorder image, the real-time analysis period of the recorder is divided into the first-type law enforcement period.

[0041] Step S162: When the law enforcement image recognition model identifies that any law enforcement monitoring image in the real-time analysis cycle of the recorder is not a first type of law enforcement recorder image, the real-time analysis cycle of the recorder is divided into the second type of law enforcement cycle.

[0042] Step S17: When the real-time analysis cycle of the recorder is the first type of law enforcement cycle, the second type of law enforcement recorder image is acquired from multiple law enforcement monitoring images in the real-time analysis cycle of the recorder to obtain the number of second type images;

[0043] Step S18: Calculate the ratio of the number of second-type images to the number of law enforcement image cycles to obtain the feature law enforcement image quantity ratio;

[0044] Step S19: Define the ratio of the number of feature-based law enforcement images and the data of the law enforcement cycle division of the recorder as recorder image recognition data;

[0045] Step S2: Based on the image recognition data of the recorder, the second type of law enforcement cycle is voice acquired to obtain multiple law enforcement scene audios. A law enforcement audio recognition model is created to recognize multiple law enforcement scene audios to obtain multiple target law enforcement recorder audios. By analyzing the multiple target law enforcement recorder audios, the audio cycle anomaly index corresponding to the second type of law enforcement cycle is obtained.

[0046] Step S2 further includes the following specific steps:

[0047] Step S21: Obtain image recognition data from the recorder, and obtain the ratio of the number of feature law enforcement images and the law enforcement cycle division data from the recorder based on the image recognition data from the recorder;

[0048] Step S22: Obtain the recorder audio data corresponding to the second type of law enforcement cycle, and use audio software to divide the recorder audio data into several different types of law enforcement scene audio to obtain multiple law enforcement scene audios;

[0049] Step S23: Create a law enforcement audio recognition model;

[0050] Step S23 further includes the following specific steps:

[0051] Step S231: Use big data crawling technology to obtain multiple law enforcement recorder audio clips using the audio clips of law enforcement recorders as keywords;

[0052] Step S232: Each law enforcement recorder audio clip is manually labeled as either target law enforcement recorder audio or non-target law enforcement recorder audio, thus obtaining law enforcement recorder audio labeling data;

[0053] Step S233: Divide the law enforcement recorder audio tag data into a law enforcement audio training set and a law enforcement audio test set according to the audio training-test ratio;

[0054] Step S234: Create an audio recognition model using an existing artificial intelligence platform, and train the audio recognition model using the law enforcement audio training set until each piece of law enforcement recorder audio in the law enforcement audio training set is used to train the audio recognition model once.

[0055] Step S235: Test the audio recognition model using the law enforcement audio test set and obtain the audio recognition accuracy. When the audio recognition accuracy is greater than or equal to the target audio recognition accuracy, the audio recognition model training is completed and the law enforcement audio recognition model is obtained. When the audio recognition accuracy is less than the target audio recognition accuracy, continue to train the audio recognition model using the law enforcement audio training set until the audio recognition accuracy is greater than or equal to the target audio recognition accuracy.

[0056] Step S24: Use the law enforcement audio recognition model to identify multiple law enforcement scene audios, and acquire the audios that are identified as target law enforcement recorder audios to obtain multiple target law enforcement recorder audios;

[0057] Step S25: Name the obtained target law enforcement recorder audios as First Target Law Enforcement Audio to Target a, respectively;

[0058] Step S26: Obtain the periodic audio loudness corresponding to the first target law enforcement audio to the a-th target law enforcement audio within the second type of law enforcement cycle, and obtain the average loudness of the first audio cycle to the average loudness of the a-th audio cycle;

[0059] Step S27: Obtain the duration of each period corresponding to the first target law enforcement audio to the a-th target law enforcement audio within the second type of law enforcement cycle, and obtain the duration of the first audio cycle to the duration of the a-th audio cycle;

[0060] Step S28: Calculate the audio cycle anomaly index corresponding to the second type of law enforcement cycle by combining the average loudness of the first audio cycle to the average loudness of the a-th audio cycle and the duration of the first audio cycle to the duration of the a-th audio cycle.

[0061] The audio cycle anomaly index corresponding to the second type of law enforcement cycle is calculated using the following formula:

[0062] ;

[0063] Wherein, Yyz is the audio cycle anomaly index corresponding to the second type of law enforcement cycle, Scx1 to Scxa are the duration of the first audio cycle to the duration of the a-th audio cycle, and Xdc1 to Xdca are the average loudness of the first audio cycle to the average loudness of the a-th audio cycle.

[0064] Step S3: Monitor and issue early warnings for the real-time analysis cycle of the recorder based on the recorder's image recognition data and audio cycle anomaly index;

[0065] Step S3 further includes the following specific steps:

[0066] Step S31: Obtain image recognition data from the recorder, and obtain law enforcement cycle division data from the recorder based on the image recognition data;

[0067] Step S32: When the real-time analysis cycle of the recorder is the first type of law enforcement cycle, the law enforcement recorder automatically issues a video warning;

[0068] Step S33: When the real-time analysis cycle of the recorder is the second type of law enforcement cycle, further analysis of the real-time analysis cycle of the recorder is conducted.

[0069] Step S33 further includes the following specific steps:

[0070] Step S331: Obtain the ratio of the number of feature-based law enforcement images based on the image recognition data from the recorder;

[0071] Step S332: Obtain the audio periodicity index;

[0072] Step S333: Calculate the video anomaly coefficient corresponding to the second type of law enforcement cycle by combining the ratio of the number of feature law enforcement images and the audio cycle anomaly index;

[0073] The video anomaly coefficient is calculated using the following formula:

[0074] ;

[0075] Where Spy is the video anomaly coefficient, Yyz is the audio periodic anomaly index, and Tsb is the ratio of the number of feature law enforcement images;

[0076] Step S334: Obtain the video anomaly coefficient threshold, compare the video anomaly coefficient with the video anomaly coefficient threshold, and issue an early warning for the real-time analysis cycle of the recorder in the second type of law enforcement cycle based on the comparison result.

[0077] Step S334 further includes the following specific steps:

[0078] Step S3341: Obtain the threshold for the ratio of the number of feature-enforcement images and the threshold for the audio periodicity anomaly index;

[0079] Step S3342: Calculate the video anomaly coefficient threshold corresponding to the second type of law enforcement cycle by combining the feature law enforcement image quantity ratio threshold and the audio cycle anomaly index threshold;

[0080] The threshold for the video anomaly coefficient is calculated using the following formula:

[0081] ;

[0082] Where Spy is the video anomaly coefficient, Yyz is the audio periodic anomaly index threshold, and Tsb is the feature law enforcement image number ratio threshold.

[0083] Step S3343: When the video anomaly coefficient is greater than or equal to the video anomaly coefficient threshold, the law enforcement recorder automatically issues a video warning;

[0084] Step S3344: When the video anomaly coefficient is less than the video anomaly coefficient threshold, the law enforcement recorder will not issue a video warning.

[0085] In this application, if a corresponding calculation formula appears, the above calculation formula is a dimensionless calculation. The weighting coefficient, proportional coefficient and other coefficients in the formula are set to quantify each parameter to obtain a result value. The size of the weighting coefficient and proportional coefficient is only required to not affect the proportional relationship between the parameter and the result value.

[0086] Example 2

[0087] Please see Figure 2Based on another concept of the same invention, an intelligent recognition system for law enforcement recorders is proposed, including an image recognition module, a voice recognition module, an intelligent early warning module, and a server. The image recognition module, the voice recognition module, and the intelligent early warning module are respectively connected to the server, and the server controls the image recognition module, the voice recognition module, and the intelligent early warning module respectively.

[0088] The image recognition module acquires the real-time analysis cycle of the recorder, acquires multiple law enforcement monitoring images within the real-time analysis cycle, creates a law enforcement image recognition model to recognize the multiple law enforcement monitoring images, and divides the real-time analysis cycle of the recorder into a first type of law enforcement cycle and a second type of law enforcement cycle based on the recognition results, thereby obtaining the law enforcement cycle division data of the recorder. The second type of law enforcement cycle is then analyzed to obtain the ratio of the number of characteristic law enforcement images. The ratio of the number of characteristic law enforcement images and the law enforcement cycle division data of the recorder are defined as the recorder image recognition data.

[0089] When the law enforcement recorder is working, the time value corresponding to the current moment is marked as the first image feature time point, the time value corresponding to the feature monitoring period before the first image feature time point is marked as the second image feature time point, and the time period between the first image feature time point and the second image feature time point is marked as the recorder's real-time analysis cycle.

[0090] It should be noted here that:

[0091] In this application, the time length corresponding to the feature monitoring period mentioned here is specifically 30 seconds;

[0092] In this application, as the time value corresponding to the current moment changes, the first feature monitoring time point and the second feature monitoring time point also change accordingly, thereby realizing the dynamic update of the recorder's real-time analysis cycle.

[0093] The law enforcement video images from the recorder are extracted frame by frame during the real-time analysis period of the recorder to obtain several law enforcement images from the recorder, thus obtaining the law enforcement image extraction data from the recorder.

[0094] During the real-time analysis cycle of the recorder, several time points are randomly selected to obtain multiple video monitoring time points. The time interval between each two consecutive video monitoring points is equal. The law enforcement image corresponding to each video monitoring point is obtained by extracting data from the law enforcement image of the recorder, resulting in multiple law enforcement monitoring images.

[0095] The number of law enforcement monitoring images within the real-time analysis period of the recorder is acquired to obtain the number of law enforcement images per period.

[0096] Create a law enforcement image recognition model;

[0097] Specifically as follows:

[0098] Multiple images from law enforcement recorders were obtained using big data web scraping technology with images from law enforcement recorders as keywords;

[0099] Each law enforcement recorder image was manually labeled as a type 1 law enforcement recorder image, a type 2 law enforcement recorder image, and a type 3 law enforcement recorder image, thus obtaining law enforcement recorder image labeling data;

[0100] It should be noted here that:

[0101] In this application, the first type of law enforcement recorder images referred to herein are law enforcement images with high social harm, specifically including but not limited to violent law enforcement, punitive law enforcement, and violent incidents. The second type of law enforcement recorder images referred to herein are law enforcement images with moderate social harm, specifically including but not limited to routine arrests, small-scale conflicts, and non-physical resistance to law enforcement. The third type of law enforcement recorder images referred to herein are law enforcement images without social harm, specifically including but not limited to law enforcement patrols, routine traffic enforcement, and lawful inspections. Please refer to [link / reference needed]. Figure 3 ;

[0102] The law enforcement recorder image labeling data is divided into a law enforcement image training set and a law enforcement image test set according to the image training-test ratio;

[0103] It should be noted here that:

[0104] In this application, the image training-to-test ratio is specifically set to 7:3, that is, the ratio of the number of law enforcement recorder images in the law enforcement image training set to the number of law enforcement image test sets is 7:3.

[0105] An image recognition model is created using an existing artificial intelligence platform. The model is then trained using a law enforcement image training set until each law enforcement recorder image in the training set is used to train the image recognition model once.

[0106] The image recognition model is tested using a law enforcement image test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model is trained and the law enforcement image recognition model is obtained. When the recognition accuracy is less than the target recognition accuracy, the image recognition model is trained again using the law enforcement image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

[0107] It should be noted here that:

[0108] In this application, the target recognition accuracy rate is specifically set at 95%.

[0109] The law enforcement image recognition model is used to identify each law enforcement monitoring image in the real-time analysis period of the recorder. Based on the recognition results, the real-time analysis period of the recorder is divided into the first type of law enforcement period and the second type of law enforcement period, and the law enforcement period division data of the recorder is obtained.

[0110] Specifically as follows:

[0111] When the law enforcement image recognition model identifies any one of the multiple law enforcement monitoring images in the real-time analysis cycle of the recorder as a first-type law enforcement recorder image, the real-time analysis cycle of the recorder is classified as a first-type law enforcement cycle.

[0112] When the law enforcement image recognition model identifies any law enforcement monitoring image in the real-time analysis cycle of the recorder that is not a first-type law enforcement recorder image, the real-time analysis cycle of the recorder is classified as a second-type law enforcement cycle.

[0113] When the real-time analysis cycle of the recorder is the first type of law enforcement cycle, the second type of law enforcement recorder images are acquired from multiple law enforcement monitoring images in the real-time analysis cycle of the recorder to obtain the number of second type images;

[0114] Calculate the ratio of the number of second-type images to the number of law enforcement image cycles to obtain the characteristic law enforcement image quantity ratio;

[0115] The ratio of the number of feature-based law enforcement images and the data on the division of law enforcement cycles by recorders are defined as recorder image recognition data.

[0116] The image recognition module acquires image recognition data from the recorder and transmits it to the voice recognition module and the intelligent early warning module;

[0117] The speech recognition module acquires speech data from the recorder image recognition data for the second type of law enforcement cycle, obtaining multiple law enforcement scene audios. It then creates a law enforcement audio recognition model to identify the multiple law enforcement scene audios to obtain multiple target law enforcement recorder audios. By analyzing the multiple target law enforcement recorder audios, it obtains the audio cycle anomaly index corresponding to the second type of law enforcement cycle.

[0118] Acquire image recognition data from the recorder, and based on the image recognition data, obtain the ratio of the number of feature-based law enforcement images and the data on the division of law enforcement cycles from the recorder.

[0119] It should be noted here that:

[0120] The data on the classification of law enforcement cycles involving recorders mentioned here includes the first type of law enforcement cycle and the second type of law enforcement cycle;

[0121] Acquire the recorder audio data corresponding to the second type of law enforcement cycle, and use audio software to divide the recorder audio data into several different types of law enforcement scene audio, thus obtaining multiple law enforcement scene audios;

[0122] Create a law enforcement audio recognition model;

[0123] Specifically as follows:

[0124] Multiple law enforcement recorder audio clips were obtained using big data crawling technology with law enforcement recorder audio as the keyword.

[0125] Each law enforcement recorder audio clip was manually labeled as either target law enforcement recorder audio or non-target law enforcement recorder audio, thus obtaining law enforcement recorder audio labeling data.

[0126] It should be noted here that:

[0127] In this application, the target law enforcement recorder audio refers to law enforcement audio with high social harm, including but not limited to background sounds of insults, fighting, and mass riots. The non-target law enforcement recorder audio refers to law enforcement audio without social harm, including but not limited to law enforcement patrol audio, routine traffic law enforcement audio, and legal inspection audio.

[0128] The audio tagging data from law enforcement recorders was divided into a law enforcement audio training set and a law enforcement audio test set according to the audio training-test ratio.

[0129] It should be noted here that:

[0130] In this application, the audio training-to-test ratio is specifically set to 7:3, that is, the ratio of the number of law enforcement recorder audios in the law enforcement audio training set to the number of law enforcement audio test sets is 7:3.

[0131] An audio recognition model is created using an existing artificial intelligence platform. The audio recognition model is then trained using a law enforcement audio training set until each segment of law enforcement recorder audio in the law enforcement audio training set is used to train the audio recognition model once.

[0132] The audio recognition model is tested using the law enforcement audio test set, and the audio recognition accuracy is obtained. When the audio recognition accuracy is greater than or equal to the target audio recognition accuracy, the audio recognition model is trained and the law enforcement audio recognition model is obtained. When the audio recognition accuracy is less than the target audio recognition accuracy, the audio recognition model is trained again using the law enforcement audio training set until the audio recognition accuracy is greater than or equal to the target audio recognition accuracy.

[0133] It should be noted here that:

[0134] The target audio recognition accuracy rate mentioned here is specifically set to 95% in this application;

[0135] The law enforcement audio recognition model is used to identify multiple law enforcement scene audios, and the audios that are identified as target law enforcement recorder audios are acquired to obtain multiple target law enforcement recorder audios;

[0136] The obtained audio recordings from multiple target law enforcement recorders are named as follows: First Target Law Enforcement Audio to Target a Law Enforcement Audio;

[0137] It should be noted here that:

[0138] In this application, 'a' refers to the quantity value corresponding to the audio from the target law enforcement recorder, and 'a' is an integer greater than 0.

[0139] The loudness of the periodic audio corresponding to the first target law enforcement audio to the a-th target law enforcement audio within the second type of law enforcement cycle is obtained respectively, and the average loudness of the first audio cycle to the average loudness of the a-th audio cycle is obtained.

[0140] The duration of each period corresponding to the first target law enforcement audio to the ath target law enforcement audio within the second type of law enforcement cycle is obtained, and the duration of the first audio cycle to the duration of the ath audio cycle is obtained.

[0141] The average loudness of the first audio cycle to the average loudness of the a-th audio cycle and the duration of the first audio cycle to the duration of the a-th audio cycle are used to calculate the audio cycle anomaly index corresponding to the second type of law enforcement cycle.

[0142] The audio cycle anomaly index corresponding to the second type of law enforcement cycle is calculated using the following formula:

[0143] ;

[0144] Wherein, Yyz is the audio cycle anomaly index corresponding to the second type of law enforcement cycle, Scx1 to Scxa are the duration of the first audio cycle to the duration of the a-th audio cycle, and Xdc1 to Xdca are the average loudness of the first audio cycle to the average loudness of the a-th audio cycle.

[0145] The speech recognition module acquires the audio cycle anomaly index and sends it to the intelligent early warning module;

[0146] The intelligent early warning module monitors and issues warnings based on the recorder's image recognition data and audio cycle anomaly index to analyze the recorder's real-time cycle.

[0147] Acquire image recognition data from the recorder, and obtain law enforcement cycle segmentation data from the recorder based on the image recognition data;

[0148] When the real-time analysis cycle of the recorder is the first type of law enforcement cycle, the law enforcement recorder will automatically issue a video warning.

[0149] If the real-time analysis cycle of the recorder is the second type of law enforcement cycle, then the real-time analysis cycle of the recorder will be further analyzed.

[0150] Specifically as follows:

[0151] The ratio of the number of feature-rich law enforcement images is obtained based on the image recognition data from the recorder.

[0152] Obtain the audio cycle anomaly index;

[0153] The video anomaly coefficient corresponding to the second type of law enforcement cycle is obtained by calculating the ratio of the number of feature law enforcement images and the audio cycle anomaly index.

[0154] The video anomaly coefficient is calculated using the following formula:

[0155] ;

[0156] Where Spy is the video anomaly coefficient, Yyz is the audio periodic anomaly index, and Tsb is the ratio of the number of feature law enforcement images;

[0157] Obtain the video anomaly coefficient threshold, compare the video anomaly coefficient with the video anomaly threshold, and issue an early warning for the real-time analysis cycle of the recorder in the second type of law enforcement cycle based on the comparison result.

[0158] Specifically as follows:

[0159] Acquire the threshold for the number of feature-rich law enforcement images and the threshold for the audio periodicity anomaly index;

[0160] The video anomaly coefficient threshold corresponding to the second type of law enforcement cycle is obtained by calculating the threshold of the ratio of the number of feature law enforcement images and the threshold of the audio cycle anomaly index.

[0161] The threshold for the video anomaly coefficient is calculated using the following formula:

[0162] ;

[0163] Where Spy is the video anomaly coefficient, Yyz is the audio periodic anomaly index threshold, and Tsb is the feature law enforcement image number ratio threshold.

[0164] When the video anomaly coefficient is greater than or equal to the video anomaly coefficient threshold, the law enforcement recorder will automatically issue a video warning.

[0165] When the video anomaly coefficient is less than the video anomaly coefficient threshold, the law enforcement recorder will not issue a video warning.

[0166] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for intelligent recognition using law enforcement recorders, characterized in that: include: Step S1: Obtain the real-time analysis cycle of the recorder. Acquire multiple law enforcement monitoring images within the real-time analysis cycle of the recorder. Create a law enforcement image recognition model to recognize the multiple law enforcement monitoring images. Based on the recognition results, divide the real-time analysis cycle of the recorder into a first type of law enforcement cycle and a second type of law enforcement cycle to obtain the law enforcement cycle division data of the recorder. Analyze the second type of law enforcement cycle to obtain the ratio of the number of characteristic law enforcement images. Define the ratio of the number of characteristic law enforcement images and the law enforcement cycle division data of the recorder as the recorder image recognition data. The first type of law enforcement recorder images are law enforcement images with high social harm, including violent law enforcement, punitive law enforcement, and violent incidents. The second type of law enforcement recorder images are law enforcement images with moderate social harm, including routine arrest law enforcement, small-scale conflicts, and non-physical resistance law enforcement. Step S2: Based on the image recognition data of the recorder, the second type of law enforcement cycle is voice acquired to obtain multiple law enforcement scene audios. A law enforcement audio recognition model is created to recognize the multiple law enforcement scene audios and to acquire multiple target law enforcement recorder audios. By analyzing the multiple target law enforcement recorder audios, the audio cycle anomaly index corresponding to the second type of law enforcement cycle is obtained. Step S3: Monitor and issue early warnings for the real-time analysis cycle of the recorder based on the recorder's image recognition data and audio cycle anomaly index; Step S1 further includes the following steps: Step S14: Acquire the number of law enforcement monitoring images within the real-time analysis period of the recorder to obtain the law enforcement image period quantity value; Step S15: Create a law enforcement image recognition model; Step S16: Use the law enforcement image recognition model to identify each law enforcement monitoring image in the real-time analysis period of the recorder, and divide the real-time analysis period of the recorder into the first type of law enforcement period and the second type of law enforcement period according to the recognition results to obtain the law enforcement period division data of the recorder. Step S17: When the real-time analysis cycle of the recorder is the second type of law enforcement cycle, the second type of law enforcement recorder images in multiple law enforcement monitoring images of the real-time analysis cycle of the recorder are acquired to obtain the number of second type images; Step S18: Calculate the ratio of the number of second-type images to the number of law enforcement image cycles to obtain the feature law enforcement image quantity ratio; Step S19: Define the ratio of the number of feature-based law enforcement images and the data of the law enforcement cycle division of the recorder as recorder image recognition data; Step S2 further includes the following specific steps: Step S21: Obtain image recognition data from the recorder, and obtain the ratio of the number of feature law enforcement images and the law enforcement cycle division data from the recorder based on the image recognition data from the recorder; Step S22: Obtain the recorder audio data corresponding to the second type of law enforcement cycle, and use audio software to divide the recorder audio data into several different types of law enforcement scene audio to obtain multiple law enforcement scene audios; Step S23: Create a law enforcement audio recognition model; Step S24: Use the law enforcement audio recognition model to identify multiple law enforcement scene audios, and acquire the audios that are identified as target law enforcement recorder audios to obtain multiple target law enforcement recorder audios; Step S25: Name the obtained target law enforcement recorder audios as First Target Law Enforcement Audio to Target a, respectively; Step S26: Obtain the periodic audio loudness corresponding to the first target law enforcement audio to the a-th target law enforcement audio within the second type of law enforcement cycle, and obtain the average loudness of the first audio cycle to the average loudness of the a-th audio cycle; Step S27: Obtain the duration of each period corresponding to the first target law enforcement audio to the a-th target law enforcement audio within the second type of law enforcement cycle, and obtain the duration of the first audio cycle to the duration of the a-th audio cycle; Step S28: Calculate the audio cycle anomaly index corresponding to the second type of law enforcement cycle by combining the average loudness of the first audio cycle to the average loudness of the a-th audio cycle and the duration of the first audio cycle to the duration of the a-th audio cycle; The audio cycle anomaly index corresponding to the second type of law enforcement cycle is calculated using the following formula: ； Wherein, Yyz is the audio cycle anomaly index corresponding to the second type of law enforcement cycle, Scx1 to Scxa are the duration of the first audio cycle to the duration of the a-th audio cycle, and Xdc1 to Xdca are the average loudness of the first audio cycle to the average loudness of the a-th audio cycle. Step S1 further includes the following specific steps: Step S11: When the law enforcement recorder is working, the time value corresponding to the current moment is marked as the first image feature time point, the time value corresponding to the feature monitoring period before the first image feature time point is marked as the second image feature time point, and the time period between the first image feature time point and the second image feature time point is marked as the recorder's real-time analysis cycle.

2. The intelligent recognition method for law enforcement recorders according to claim 1, characterized in that, Step S1 further includes the following steps: Step S12: Capture the law enforcement video images of the recorder frame by frame within the real-time analysis period of the recorder to obtain several law enforcement images of the recorder and obtain law enforcement image capture data of the recorder. Step S13: During the real-time analysis cycle of the recorder, several time points are randomly selected to obtain multiple video monitoring time points. The time interval between any two consecutive video monitoring points is equal. Based on the data captured from the law enforcement images of the recorder, the law enforcement images corresponding to each video monitoring point are obtained, resulting in multiple law enforcement monitoring images.

3. The intelligent recognition method for law enforcement recorders according to claim 1, characterized in that, Step S15 further includes the following specific steps: Step S151: Use big data crawling technology to obtain multiple law enforcement recorder images using law enforcement recorder images as keywords; Step S152: Each law enforcement recorder image is manually labeled as a first type law enforcement recorder image, a second type law enforcement recorder image, and a third type law enforcement recorder image to obtain law enforcement recorder image labeling data; Step S153: Divide the law enforcement recorder image labeling data into a law enforcement image training set and a law enforcement image test set according to the image training-test ratio.

4. The intelligent recognition method for law enforcement recorders according to claim 1, characterized in that, Step S154: Create an image recognition model using an existing artificial intelligence platform, and train the image recognition model using a law enforcement image training set until each law enforcement recorder image in the law enforcement image training set is used to train the image recognition model once. Step S155: Test the image recognition model using the law enforcement image test set and obtain the recognition accuracy. When the recognition accuracy is greater than or equal to the target recognition accuracy, the image recognition model training is complete, and the law enforcement image recognition model is obtained. When the recognition accuracy is less than the target recognition accuracy, continue to train the image recognition model using the law enforcement image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.

5. The intelligent recognition method for law enforcement recorders according to claim 1, characterized in that, Step S16 further includes the following specific steps: Step S161: When the law enforcement image recognition model identifies any law enforcement monitoring image within the real-time analysis period of the recorder as a first-type law enforcement recorder image, the real-time analysis period of the recorder is divided into the first-type law enforcement period. Step S162: When the law enforcement image recognition model identifies that any law enforcement monitoring image in the real-time analysis cycle of the recorder is not a first type of law enforcement recorder image, the real-time analysis cycle of the recorder is divided into the second type of law enforcement cycle.

6. The intelligent recognition method for law enforcement recorders according to claim 1, characterized in that, Step S3 further includes the following specific steps: Step S31: Obtain image recognition data from the recorder, and obtain law enforcement cycle division data from the recorder based on the image recognition data; Step S32: When the real-time analysis cycle of the recorder is the first type of law enforcement cycle, the law enforcement recorder automatically issues a video warning; Step S33: When the real-time analysis cycle of the recorder is the second type of law enforcement cycle, further analysis of the real-time analysis cycle of the recorder is conducted.

7. A law enforcement recorder intelligent recognition system, applicable to the law enforcement recorder intelligent recognition method according to any one of claims 1-6, characterized in that, The specific working process of each module of the intelligent recognition system is as follows: Image recognition module: Used to acquire the real-time analysis cycle of the recorder, acquire multiple law enforcement monitoring images within the real-time analysis cycle of the recorder, create a law enforcement image recognition model to recognize the multiple law enforcement monitoring images, divide the real-time analysis cycle of the recorder into a first type of law enforcement cycle and a second type of law enforcement cycle based on the recognition results, obtain the law enforcement cycle division data of the recorder, analyze the second type of law enforcement cycle to obtain the ratio of the number of characteristic law enforcement images, and define the ratio of the number of characteristic law enforcement images and the law enforcement cycle division data of the recorder as the recorder image recognition data; The speech recognition module is used to acquire speech data for the second type of law enforcement cycle based on the image recognition data of the recorder, obtain multiple law enforcement scene audios, create a law enforcement audio recognition model to recognize multiple law enforcement scene audios to obtain multiple target law enforcement recorder audios, and obtain the audio cycle anomaly index corresponding to the second type of law enforcement cycle by analyzing the audios of multiple target law enforcement recorders. Intelligent early warning module: used to monitor and issue early warnings for the real-time analysis cycle of the recorder based on the recorder's image recognition data and audio cycle anomaly index.

Images