Personnel abnormal behavior detection and alarm system based on real-time video

Abstract

The invention provides a personnel abnormal behavior detection and alarm system based on a real-time video, and relates to the technical field of behavior detection, and the system comprises a data collection module which is used for capturing a video image of a target monitoring area in real time, and carrying out the video processing, and obtaining first video data; the model training module is used for acquiring historical video data of a target monitoring area and identifying and extracting key features of target personnel so as to perform behavior modeling to obtain a first normal behavior model and a first abnormal behavior model; the behavior detection module is used for comparing current video frame data in the first video data with a first normal behavior model, and if an abnormal behavior exists, determining a first behavior detection result based on the current video frame data; and the alarm processing module is used for comprehensively determining alarm information of the target monitoring area according to the first behavior detection result and the monitoring area information, and carrying out abnormal behavior alarm. The behavior model can be adjusted in real time, and the monitoring flexibility and the behavior detection accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of behavior detection, and particularly to a system for detecting and alarming abnormal behaviors of personnel based on real-time video. Background Art

[0002] At present, with the rapid development of artificial intelligence technology, intelligent video surveillance technology has been widely applied.

[0003] However, traditional video surveillance systems mainly rely on manual monitoring, which has problems such as low work efficiency, easy fatigue, and easy omission. Existing intelligent systems that can automatically detect abnormal behaviors of personnel and give alarms have problems such as low monitoring efficiency and weak monitoring flexibility.

[0004] Therefore, the present invention provides a system for detecting and alarming abnormal behaviors of personnel based on real-time video. Summary of the Invention

[0005] The present invention provides a system for detecting and alarming abnormal behaviors of personnel based on real-time video to solve the problems of low monitoring efficiency and weak monitoring flexibility in the prior art.

[0006] The present invention provides a system for detecting and alarming abnormal behaviors of personnel based on real-time video, including: A data acquisition module: configured to capture video images of a target monitoring area in real time based on a preset camera and perform video processing to obtain first video data; A model training module: configured to obtain historical video data of the target monitoring area, identify and extract key features of the target personnel from the historical video data, and thus perform behavior modeling based on the extracted key features to obtain a first normal behavior model and a first abnormal behavior model; A behavior detection module: configured to compare the current video frame data in the first video data with the first normal behavior model, and if there is an abnormal behavior, compare the current video frame data with the first abnormal behavior model to determine a first behavior detection result; An alarm processing module: configured to comprehensively determine alarm information of the target monitoring area based on the first behavior detection result and monitoring area information and perform abnormal behavior alarm.

[0007] According to the data acquisition module provided by the present invention, it includes: A data acquisition unit: configured to capture video images of a target monitoring area in real time based on a preset camera to obtain initial video images; A data processing unit: configured to perform image preprocessing on the initial video images, and extract the initial video image with the highest image clarity in each detection period as the first video image corresponding to the detection period, so as to obtain first video data of the target monitoring in each detection period.

[0008] According to the model training module provided by the present invention, it includes: Historical data acquisition unit: used to extract historical video data from the data storage system of the target monitoring area to obtain the first historical video data; Key feature extraction unit: used to identify and extract the key features of the target person from the first historical video data based on image processing technology; Normal behavior model unit: used to model normal behavior based on the key features and using machine learning algorithms to obtain the first normal behavior model; Abnormal behavior model unit: used to expand the first normal behavior model so as to be able to identify abnormal behaviors that deviate from normal behaviors and obtain the first abnormal behavior model.

[0009] According to the abnormal behavior model unit provided by the present invention, it includes: Data collection sub-unit: used to extract known abnormal behavior data from the data storage system of the target monitoring area to obtain the first abnormal behavior data; Extraction and expansion sub-unit: used to randomly extract any set of abnormal key features of the first abnormal behavior data and add the abnormal key features to the first normal behavior model, thereby expanding the recognition ability of the first normal behavior model to obtain the first expanded model; Model training sub-unit: used to train the first expanded model with the remaining abnormal behavior data in the first abnormal behavior data and adjust the model parameters based on each training result to obtain the first abnormal behavior model; Threshold setting sub-unit: used to obtain the detection accuracy of abnormal behaviors in the target monitoring area, thereby determining the abnormal behavior threshold of the target monitoring area; Adjustment and optimization sub-unit: used to use the abnormal behavior threshold as the abnormal behavior threshold of the first abnormal behavior model.

[0010] According to the model training sub-unit provided by the present invention, it includes: Function determination block: used to obtain the task type of abnormal behavior detection, thereby determining the initial loss function; Model training block: used to train the first expanded model with the key features of each abnormal behavior data except the abnormal key features in the first abnormal behavior data; Training loss block: used to determine the function value of the initial loss function corresponding to each training sub-process in each training process of model training to obtain the first function value set; Curve determination block: used to input each function value in the first function value set into the same coordinate system in sequence and connect the curves to obtain the first function curve; Function trend block: used to judge the comprehensive curve trend of the first function curve. If the comprehensive curve trend is a trend of reducing loss, it is judged that the current training process is reasonable; If the comprehensive curve trend is not a curve of reducing loss, it is judged that the current training process is unreasonable and it is necessary to re-extract key features and train the model; Initial model block: used to complete model training on the first extended model based on a reasonable training process to obtain an initial abnormal behavior model; Function optimization block: used to determine the function generalization ability of the initial loss function based on the curve fluctuation of the first function curve; If the curve fluctuation is greater than the preset maximum curve fluctuation, it is judged that the function generalization ability of the initial loss function is poor and it is necessary to adjust the initial loss function to obtain the first loss function; If the curve fluctuation is not greater than the preset maximum curve fluctuation, it is judged that the function generalization ability of the initial loss function is strong and the initial loss function is used as the first loss function; Model optimization block: used to randomly extract abnormal behavior data from the first abnormal behavior data as a test set, and based on the test set, input it into the initial abnormal behavior model for model evaluation, and optimize the initial abnormal behavior model based on the model evaluation result to obtain the first abnormal behavior model.

[0011] According to the behavior detection module provided by the present invention, it includes: First comparison unit: used to extract the current video frame data from the first video data and make a first comparison between the current video frame data and the first normal behavior model; If the first comparison result determines that there is abnormal behavior in the current video frame data, the current video frame data is used as the first video frame data; Second comparison unit: used to make a second comparison between the first video frame data and the first abnormal behavior model, and determine the abnormal behavior type and abnormal behavior degree of the first video frame data based on the comparison result, so as to obtain the first behavior detection result of the target monitoring area.

[0012] According to the alarm processing module provided by the present invention, it includes: Alarm information determination unit: used to obtain the monitoring area information of the target monitoring area in real time and synthesize the first behavior detection result with the monitoring area information, so as to determine the alarm information of the target monitoring area; Behavior alarm unit: used to issue various forms of alarms based on the alarm information, so as to realize behavior alarm for the monitoring area.

[0013] According to the behavior alarm unit provided by the present invention, it includes: Information analysis sub-unit: used to classify and analyze information based on the alarm information, so as to determine the information type of the alarm information; Scheme determination subunit: used to determine the alarm level and alarm scheme of the corresponding type based on the information type of the alarm information; Behavior alarm subunit: used to perform corresponding forms of behavior alarms on the target monitoring area based on the alarm level and the corresponding alarm scheme.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: The personnel abnormal behavior detection and alarm system based on real-time video provided by the present invention processes real-time video images, combines historical video data for behavior modeling to obtain a first abnormal behavior model, thereby performing behavior detection on real-time video images, and can adjust the behavior model in real time, improving the monitoring flexibility and behavior detection accuracy. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a structural diagram of the personnel abnormal behavior detection and alarm system based on real-time video provided by the embodiments of the present invention. Detailed Embodiments

[0017] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0018] Embodiment 1: The embodiments of the present invention provide a personnel abnormal behavior detection and alarm system based on real-time video, as Figure 1 shown, including: Data acquisition module: used to capture video images of the target monitoring area in real time based on a preset camera and perform video processing to obtain first video data; Model training module: used to obtain historical video data of the target monitoring area, identify and extract key features of the target personnel from the historical video data, and thus perform behavior modeling based on the extracted key features to obtain a first normal behavior model and a first abnormal behavior model; Behavior detection module: used to compare the current video frame data in the first video data with the first normal behavior model. If there is an abnormal behavior, then compare the current video frame data with the first abnormal behavior model to determine the first behavior detection result; Alarm processing module: used to comprehensively determine the alarm information of the target monitoring area based on the first behavior detection result and the monitoring area information, and conduct abnormal behavior alarm.

[0019] In this embodiment, the preset camera refers to a camera device that is pre-installed and configured, and is used to capture video images of the target monitoring area in real time.

[0020] In this embodiment, the target monitoring area refers to the specific geographical area or spatial range monitored by the camera. Within the target monitoring area, the camera will capture and analyze video images to detect any abnormal or interesting events.

[0021] In this embodiment, video processing refers to performing a series of processing operations on the captured video images, such as denoising, enhancement, compression, etc., to obtain video data that is more suitable for further analysis or storage. Video processing may also include the extraction and preprocessing of video frames.

[0022] In this embodiment, the first video data refers to the video data captured in real time after video processing, and is used for subsequent behavior analysis and detection.

[0023] In this embodiment, the historical video data refers to the video data captured and stored previously. These data are usually used to train behavior models to identify normal and abnormal behavior patterns.

[0024] In this embodiment, the key features refer to the important features extracted from the video data that can describe the behavior of the target person, such as movement trajectory, speed, posture, etc. These features are used to construct behavior models.

[0025] In this embodiment, behavior modeling refers to the process of using machine learning or deep learning algorithms to construct normal and abnormal behavior models based on the extracted key features, which can be used to detect abnormal behaviors in new video data.

[0026] In this embodiment, the first normal behavior model refers to a model of the normal behavior pattern constructed based on the historical video data, and is used to compare with the video data captured in real time to detect abnormal behaviors.

[0027] In this embodiment, the first abnormal behavior model refers to a model of the abnormal behavior pattern constructed based on the historical video data. When the video data captured in real time does not match the normal behavior model, this model will be used to further determine the type and degree of the abnormal behavior.

[0028] In this embodiment, the current video frame data refers to the current frame in the real-time captured video stream, which is used to compare with the normal and abnormal behavior models.

[0029] In this embodiment, the detection result of the first row refers to the result obtained by comparing the current video frame data with the normal and abnormal behavior models, which is used to determine whether there is abnormal behavior.

[0030] In this embodiment, the monitoring area information refers to the additional information related to the target monitoring area, such as location, importance, historical events, etc.

[0031] In this embodiment, the alarm information refers to the alarm notification generated by the system when abnormal behavior is detected, which includes information such as the description of the abnormal behavior, location, and time.

[0032] In this embodiment, the abnormal behavior alarm refers to the process of sending an alarm to relevant personnel by means of sound, light signal, message, etc. when the system detects abnormal behavior and generates alarm information.

[0033] The beneficial effects of the above technical solution are: By processing the real-time video images and combining with the historical video data for behavior modeling, the first abnormal behavior model is obtained, so as to perform behavior detection on the real-time video images, and the behavior model can be adjusted in real time, improving the monitoring flexibility and the accuracy of behavior detection.

[0034] Embodiment 2: Based on Embodiment 1, the data acquisition module includes: The data acquisition unit: It is used to capture the video images of the target monitoring area in real time based on the preset camera to obtain the initial video images. The data processing unit: It is used to perform image preprocessing on the initial video images and extract the initial video image with the highest image clarity in each detection period as the first video image of the corresponding detection period, so as to obtain the first video data of the target monitoring in each detection period.

[0035] In this embodiment, the preset camera refers to the camera device that has been pre-installed and configured, which is used to capture the video images of the target monitoring area in real time.

[0036] In this embodiment, the target monitoring area refers to the specific geographical area or spatial range monitored by the camera. Within the target monitoring area, the camera will capture and analyze the video images to detect any abnormal or interesting events.

[0037] In this embodiment, the initial video image refers to the original video image captured by the camera in real time without any processing or screening.

[0038] In this embodiment, image preprocessing refers to a series of operations performed on the initial video image to improve image quality, enhance image features, or prepare the image for further analysis. For example, image preprocessing includes denoising, enhancing contrast, adjusting brightness, etc.

[0039] In this embodiment, the detection period refers to a preset time interval used to divide a continuous video stream into a series of discrete time periods. The video images within each detection period will be processed and analyzed separately.

[0040] In this embodiment, image clarity refers to the degree of sharpness of image details, usually measured by indicators such as image sharpness and contrast. In surveillance videos, clarity is an important criterion for measuring image quality.

[0041] In this embodiment, the first video image refers to the video image with the highest image clarity selected from the initial video image within each detection period.

[0042] In this embodiment, the first video data for target surveillance refers to the video data set composed of the first video images within each detection period. These data sets provide clear and high-quality visual records of the surveillance area.

[0043] The beneficial effects of the above technical solution are: By processing real-time video images, combined with historical video data for behavior modeling, a first abnormal behavior model is obtained, and behavior detection can be performed. The behavior model can be adjusted in real time to improve the accuracy of behavior detection.

[0044] Embodiment 3: Based on Embodiment 2, the model training module includes: Historical data acquisition unit: Used to extract historical video data from the data storage system of the target surveillance area to obtain the first historical video data; Key feature extraction unit: Used to identify and extract the key features of the target person from the first historical video data based on image processing technology; Normal behavior model unit: Used to model normal behavior based on the key features and using machine learning algorithms to obtain the first normal behavior model; Abnormal behavior model unit: Used to expand the first normal behavior model so as to be able to identify abnormal behaviors that deviate from normal behaviors and obtain the first abnormal behavior model.

[0045] In this embodiment, the data storage system of the target surveillance area refers to the system that stores historical video data related to the target surveillance area, used for long-term storage and ready access to surveillance videos.

[0046] In this embodiment, the historical video data refers to the video data captured by the surveillance camera and stored in the data storage system in a certain period of time in the past. The historical video data can be a complete video file or a specific segment that has been processed or extracted.

[0047] In this embodiment, the first historical video data refers to a specific historical video data set extracted from a data storage system of a target monitoring area. The first historical video data set may be selected based on a time range, a monitoring area, or other screening conditions.

[0048] In this embodiment, image processing technology refers to various technologies for processing and analyzing images, including image enhancement, image transformation, image segmentation, feature extraction, etc. Image processing technology can be used to extract useful information from video data, such as the appearance characteristics and behavior patterns of target persons.

[0049] In this embodiment, key features refer to distinguishing and important features extracted from the target person during image processing. For example, key features may include facial features, gait features, clothing color, etc., which are used to uniquely identify the target person in subsequent analysis and recognition processes.

[0050] In this embodiment, in surveillance video analysis, a machine learning algorithm can be used to model the behavior patterns of target persons in order to predict and identify abnormal behaviors.

[0051] In this embodiment, the first normal behavior model refers to a model representing normal behavior established based on the key features of the target person and the machine learning algorithm. This model describes the behavior patterns that the target person may exhibit under normal circumstances, such as walking speed, dwell time, etc.

[0052] In this embodiment, model extension refers to adding new functions or improving the performance of the model based on the established model. In this context, model extension may involve extending the first normal behavior model into a model capable of identifying abnormal behavior, ie, the first abnormal behavior model.

[0053] In this embodiment, the first abnormal behavior model refers to an expanded model that can identify abnormal behaviors that deviate from the first normal behavior model. This model can be used to analyze the behavior of target persons in surveillance videos to detect possible security threats or illegal behaviors.

[0054] The beneficial effect of the above technical solution is: by constructing a first normal behavior model, and performing model expansion and model training to obtain a first abnormal behavior model, behavior detection is performed on real-time video images, and the behavior model can be adjusted in real time to improve monitoring flexibility and behavior detection accuracy.

[0055] Embodiment 4: Based on Example 3, the abnormal behavior model unit includes: Data collection sub-unit: used to extract the known abnormal behavior data in the data storage system of the target monitoring area to obtain the first abnormal behavior data; Extraction and extension sub-unit: used to randomly extract any set of abnormal key features of the first abnormal behavior data and add the abnormal key features to the first normal behavior model, thereby expanding the recognition ability of the first normal behavior model to obtain the first extended model; Model training sub-unit: used to train the first extended model with the remaining abnormal behavior data in the first abnormal behavior data and adjust the model parameters based on each training result to obtain the first abnormal behavior model; Threshold setting sub-unit: used to obtain the abnormal behavior detection accuracy of the target monitoring area, thereby determining the abnormal behavior threshold of the target monitoring area; Adjustment and optimization sub-unit: used to use the abnormal behavior threshold as the abnormal behavior threshold of the first abnormal behavior model.

[0056] In this embodiment, the known abnormal behavior data refers to the data that has been recorded and confirmed as abnormal behavior in the data storage system of the target monitoring area. For example, the data includes video clips, images, event logs, etc., for subsequent analysis and modeling.

[0057] In this embodiment, the first abnormal behavior data refers to a specific data set extracted from the known abnormal behavior data for subsequent processing and modeling. For example, the first abnormal behavior data is selected based on time range, abnormal type, or other filtering conditions.

[0058] In this embodiment, the abnormal key features refer to the features extracted from the abnormal behavior data that can distinguish normal behavior from abnormal behavior. These features may include the behavior patterns of the target personnel, appearance features, environmental factors, etc.

[0059] In this embodiment, the first normal behavior model refers to the model representing the normal behavior in the target monitoring area.

[0060] In this embodiment, the first extended model refers to the model that expands its recognition ability by adding abnormal key features to the first normal behavior model.

[0061] In this embodiment, model training refers to the process of training a machine learning model with data, aiming to enable the model to accurately identify and classify the input data. In the analysis of surveillance videos, model training usually involves using known behavior data to optimize the model parameters.

[0062] In this embodiment, model parameters refer to adjustable values in a machine learning model, and these values determine the behavior and performance of the model. During the training process, the model parameters are adjusted to optimize the accuracy and efficiency of the model.

[0063] In this embodiment, the first abnormal behavior model refers to a model that has been trained and adjusted and can accurately identify and classify abnormal behaviors in the target monitoring area. It is used for real-time monitoring and anomaly detection.

[0064] In this embodiment, the abnormal behavior detection accuracy refers to the accuracy of the first abnormal behavior model in identifying abnormal behaviors. This metric is usually calculated by comparing the prediction results of the model with the actual behavior data.

[0065] In this embodiment, the abnormal behavior threshold refers to the threshold used to determine whether a behavior is considered abnormal. In the analysis of surveillance videos, the abnormal behavior threshold is set based on the output of the model (such as probability, score, etc.). When the output of the model exceeds this threshold, the behavior is considered abnormal. For example, there is an intelligent surveillance system that detects anomalies by analyzing behaviors in videos. The system contains a machine learning model that has been trained to identify normal behaviors and potential abnormal behaviors. The output of the model may be a probability value indicating the likelihood that a behavior is abnormal. Set an abnormal behavior threshold of 0.8. If the model analyzes a video clip and gives an abnormal probability of 0.9, then this behavior will be marked as abnormal because it exceeds the abnormal behavior threshold.

[0066] The beneficial effects of the above technical solution are: By constructing the first normal behavior model, and performing model extension and model training to obtain the first abnormal behavior model, behavior detection can be performed on real-time video images, and the behavior model can be adjusted in real time, improving the monitoring flexibility and behavior detection accuracy.

[0067] Embodiment 5: Based on Embodiment 4, the model training subunit includes: Function determination block: Used to obtain the task type of abnormal behavior detection, so as to determine the initial loss function; Model training block: Used to perform model training on the first extended model with the key features of each abnormal behavior data except the abnormal key features in the first abnormal behavior data; Training loss block: Used to determine the function value of the initial loss function corresponding to each training sub-process during each training process of model training, and obtain the first function value set; Curve determination block: Used to sequentially input each function value in the first function value set into the same coordinate system and connect the curves to obtain the first function curve; Function trend block: used to determine the comprehensive curve trend of the first function curve. If the comprehensive curve trend is a trend of reducing loss, it is determined that the current training process is reasonable; If the comprehensive curve trend is not a curve of reducing loss, it is determined that the current training process is unreasonable and the extraction of key features and model training need to be carried out again; Initial model block: used to complete model training on the first extended model based on a reasonable training process to obtain an initial abnormal behavior model; Function optimization block: used to determine the function generalization ability of the initial loss function based on the curve fluctuation of the first function curve; If the curve fluctuation is greater than the preset maximum curve fluctuation, it is determined that the function generalization ability of the initial loss function is poor and the initial loss function needs to be adjusted to obtain the first loss function; If the curve fluctuation is not greater than the preset maximum curve fluctuation, it is determined that the function generalization ability of the initial loss function is strong and the initial loss function is used as the first loss function; Model optimization block: used to randomly extract abnormal behavior data from the first abnormal behavior data as a test set, and perform model evaluation based on the test set input into the initial abnormal behavior model, and optimize the initial abnormal behavior model based on the model evaluation result to obtain the first abnormal behavior model.

[0068] In this embodiment, the task type of abnormal behavior detection refers to the specific nature or category of the abnormal behavior detection task. For example, it is a classification task (distinguishing normal behavior and abnormal behavior), a regression task (predicting the degree of behavior abnormality), or other types of tasks. The task type determines the subsequent model selection and loss function setting.

[0069] In this embodiment, the initial loss function in a machine learning model is used to measure the difference between the model's prediction result and the actual result. The initial loss function is selected before the start of model training and is used to guide the direction of model optimization.

[0070] In this embodiment, the key feature refers to the variable or attribute in the dataset that can significantly affect the model's prediction result. In abnormal behavior detection, the key features may include the speed, direction, duration, etc. of the behavior.

[0071] In this embodiment, the training subprocess refers to a step or stage in the model training process. In the iterative training process, each training subprocess updates the model's parameters and calculates the value of the loss function.

[0072] In this embodiment, the first function value set refers to the set composed of the function values of the initial loss function corresponding to each training subprocess in the model training process. These values are used for subsequent analysis of the model's training effect and the generalization ability of the loss function.

[0073] In this embodiment, the first function curve is a graph obtained by sequentially inputting each function value in the first function value set into the same coordinate system and connecting them with a curve. This curve reflects the changing trend of the loss function value during the model training process.

[0074] In this embodiment, the comprehensive curve trend refers to the overall direction or trend of the first function curve, such as rising, falling, or fluctuating. The comprehensive curve trend is used to judge whether the current training process is reasonable.

[0075] In this embodiment, the function generalization ability refers to the performance ability of the loss function on unseen data. A loss function with strong generalization ability can guide the model to achieve better performance on the test set.

[0076] In this embodiment, the preset maximum curve fluctuation is a threshold set when judging the generalization ability of the loss function. If the fluctuation of the first function curve exceeds this threshold, it is considered that the generalization ability of the loss function is poor.

[0077] In this embodiment, model evaluation is a process of evaluating the performance of a trained model using a test set. The model evaluation results are used to judge the accuracy and generalization ability of the model, and the model is optimized accordingly.

[0078] In this embodiment, the first abnormal behavior model is a machine learning model that can be used for abnormal behavior detection after training and optimization. It is obtained based on the first abnormal behavior data and a reasonable training process.

[0079] The beneficial effects of the above technical solution are: By constructing the first normal behavior model, and performing model extension and model training to obtain the first abnormal behavior model, the behavior detection of real-time video images can be carried out, and the behavior model can be adjusted in real time, improving the monitoring flexibility and the accuracy of behavior detection.

[0080] Embodiment 6: Based on Embodiment 3, the behavior detection module includes: The first comparison unit: used to extract the current video frame data from the first video data and perform a first comparison between the current video frame data and the first normal behavior model; If the first comparison result determines that there is an abnormal behavior in the current video frame data, the current video frame data is used as the first video frame data; The second comparison unit: used to perform a second comparison between the first video frame data and the first abnormal behavior model, and determine the type and degree of the abnormal behavior of the first video frame data based on the comparison result, so as to obtain the first behavior detection result of the target monitoring area.

[0081] In this embodiment, the current video frame data is a single video frame extracted from the first video data. A video is composed of a series of consecutive video frames, and each frame is a static image. The current video frame data refers to the frame being analyzed.

[0082] In this embodiment, the first comparison is a process of comparing the current video frame data with the first normal behavior model. This comparison may be achieved by the model classifying or scoring the video frame, with the aim of determining whether there is an abnormal behavior in the video frame.

[0083] In this embodiment, the first comparison result is the output of the first comparison, that is, the judgment result of the first normal behavior model on the current video frame data. If the first normal behavior model believes that there is an abnormal behavior in the video frame, then the first comparison result is that there is an abnormality.

[0084] In this embodiment, the first video frame data means that if the first comparison result determines that the current video frame data has an abnormal behavior, then this video frame data is "named" as the first video frame data for subsequent analysis.

[0085] In this embodiment, the second comparison is a process of comparing the first video frame data (i.e., the video frame previously determined to have an abnormal behavior) with the first abnormal behavior model. This comparison aims to determine the type and degree of the abnormal behavior.

[0086] In this embodiment, the abnormal behavior type is the category or classification of the abnormal behavior. For example, the abnormal behavior may be intrusion, fighting, falling, etc. The first abnormal behavior model should be able to identify these different types of abnormal behaviors.

[0087] In this embodiment, the abnormal behavior degree is the severity or scope of influence of the abnormal behavior. This can be measured by scores, probabilities, or other metrics output by the model.

[0088] In this embodiment, the first behavior detection result is the result obtained after the above analysis and comparison of the first video frame data. This result should include the type and degree of the abnormal behavior, as well as other possible relevant information (such as timestamp, location, etc.). The first behavior detection result can be used to trigger an alarm, record an event, or conduct further analysis.

[0089] The beneficial effects of the above technical solution are: By performing behavior detection on the video frame data at the current moment in the real-time video image based on the first abnormal behavior model, behavior detection can be carried out timely and accurately, thereby triggering a behavior alarm in a timely manner, improving the monitoring flexibility and the accuracy of behavior detection.

[0090] Embodiment 7: Based on Embodiment 6, the alarm processing module includes: Alarm information determination unit: It is used to obtain the monitoring area information of the target monitoring area in real time, and comprehensively combine the first-line detection result with the monitoring area information to determine the alarm information of the target monitoring area; Behavior alarm unit: It is used to issue multi-form alarms based on the alarm information, so as to realize the behavior alarm of the monitoring area.

[0091] In this embodiment, the monitoring area information refers to additional information related to the target monitoring area, such as location, importance, historical events, etc.

[0092] In this embodiment, the alarm information refers to an alarm notification generated by the system when an abnormal behavior is detected, including information such as a description of the abnormal behavior, location, and time.

[0093] In this embodiment, the multi-form alarm means that there may be multiple ways or means for the system to issue an alarm, such as a sound alarm, a visual alarm (such as a flashing light), a text message alarm, an email alarm, etc. The multi-form alarm can ensure that the system can effectively attract attention and convey the alarm information in different situations.

[0094] In this embodiment, the abnormal behavior alarm refers to the process of issuing an alarm to relevant personnel by means of sound, light signal, message, etc. when the system detects an abnormal behavior and generates alarm information.

[0095] The beneficial effects of the above technical solution are: By classifying and analyzing the early warning information and then performing abnormal behavior alarms, the behavior alarms for the target monitoring area can be made more accurate and effective.

[0096] Embodiment 8: Based on Embodiment 7, the behavior alarm unit includes: Information analysis sub-unit: It is used to classify and analyze the information based on the alarm information to determine the information type of the alarm information; Scheme determination sub-unit: It is used to determine the alarm level and alarm scheme of the corresponding type based on the information type of the alarm information; Behavior alarm sub-unit: It is used to perform corresponding-form behavior alarms on the target monitoring area based on the alarm level and the corresponding alarm scheme.

[0097] In this embodiment, information classification is the process of classifying alarm information according to its nature, content, or urgency and other characteristics. Information classification helps the system or personnel to more quickly identify and process different types of alarm information.

[0098] In this embodiment, information analysis is the process of conducting in-depth research and interpretation on alarm information, aiming to extract more valuable information, such as the type of abnormal behavior, potential threats, scope of influence, etc. Information analysis helps to formulate a more accurate alarm scheme.

[0099] In this embodiment, the category or type to which the information type alarm information belongs after classification. The information type may be divided based on characteristics such as the nature of the abnormal behavior, the degree of influence, and the urgency.

[0100] In this embodiment, the alarm level is the level of urgency or importance determined according to the information type of the alarm information. The alarm level may be divided into different levels such as high, medium, and low, which are used to guide the formulation and implementation of subsequent alarm plans.

[0101] In this embodiment, the alarm plan is also a response measure or action plan formulated for specific types of alarm information. The alarm plan may include elements such as the form of the alarm (such as sound, light, text message, etc.), the content of the alarm, and the alarm frequency, aiming to ensure that the alarm information can be conveyed to relevant personnel or systems in a timely and accurate manner.

[0102] In this embodiment, the behavior alarm refers to the alarm or warning issued by the monitoring system or personnel to the target monitoring area according to the alarm plan, which is used to remind relevant personnel to pay attention to abnormal behaviors or take corresponding measures. The behavior alarm may be carried out in various forms such as sound, light, and text message.

[0103] The beneficial effects of the above technical solutions are as follows: By classifying and analyzing the early warning information, the behavior alarm for the target monitoring area becomes more accurate and effective.

[0104] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personnel abnormal behavior detection and alarm system based on real-time video, characterized in that, Including: Data acquisition module: used to capture video images of the target monitoring area in real time based on a preset camera, and perform video processing to obtain first video data; Model training module: used to obtain historical video data of the target monitoring area, identify and extract key features of the target person from the historical video data, and thus perform behavior modeling based on the extracted key features to obtain a first normal behavior model and a first abnormal behavior model; Behavior detection module: used to compare the current video frame data in the first video data with the first normal behavior model. If there is an abnormal behavior, then compare the current video frame data with the first abnormal behavior model to determine the first behavior detection result; Alarm processing module: used to comprehensively determine the alarm information of the target monitoring area based on the first behavior detection result and the monitoring area information to perform abnormal behavior alarm.

2. The personnel abnormal behavior detection and alarm system based on real-time video according to claim 1, characterized in that The data acquisition module includes: Data acquisition unit: used to capture video images of the target monitoring area in real time based on a preset camera to obtain initial video images; Data processing unit: used to perform image preprocessing on the initial video images, and extract the initial video image with the highest image clarity in each detection period as the first video image of the corresponding detection period, so as to obtain the first video data of the target monitoring in each detection period.

3. The real-time video-based personnel abnormal behavior detection and alarm system according to claim 2, characterized in that, The model training module includes: Historical data acquisition unit: used to extract historical video data from the data storage system of the target monitoring area to obtain first historical video data; Key feature extraction unit: used to identify and extract key features of the target person from the first historical video data based on image processing technology; Normal behavior model unit: used to perform modeling of normal behavior based on the key features and using machine learning algorithms to obtain a first normal behavior model; Abnormal behavior model unit: used to expand the first normal behavior model so as to be able to identify abnormal behaviors deviating from normal behaviors to obtain a first abnormal behavior model.

4. The real-time video-based personnel abnormal behavior detection and alarm system according to claim 3, characterized in that, The abnormal behavior model unit includes: Data collection subunit: used to extract known abnormal behavior data from the data storage system of the target monitoring area to obtain first abnormal behavior data; Extraction and expansion subunit: used to randomly extract any set of abnormal key features of the first abnormal behavior data and add the abnormal key features to the first normal behavior model, so as to expand the recognition ability of the first normal behavior model to obtain a first expanded model; Model training subunit: used to perform model training on the first expanded model with the remaining abnormal behavior data in the first abnormal behavior data, and adjust the model parameters based on each training result to obtain a first abnormal behavior model; Threshold setting subunit: used to obtain the abnormal behavior detection accuracy of the target monitoring area to determine the abnormal behavior threshold of the target monitoring area; Adjustment and optimization subunit: used to use the abnormal behavior threshold as the abnormal behavior threshold of the first abnormal behavior model.

5. The real-time video-based abnormal behavior detection and alarm system for personnel according to claim 4, characterized in that, The model training subunit includes: Function determination block: used to obtain the task type of abnormal behavior detection to determine the initial loss function; Model training block: used to perform model training on the first extended model with the key features of each abnormal behavior data in the first abnormal behavior data except for the abnormal key features; Training loss block: used to determine the function values of the initial loss function corresponding to each training subprocess during each training process of model training, and obtain the first function value set; Curve determination block: used to sequentially input each function value in the first function value set into the same coordinate system and connect the curves to obtain the first function curve; Function trend block: used to judge the comprehensive curve trend of the first function curve. If the comprehensive curve trend is a trend of reducing loss, it is judged that the current training process is reasonable; If the comprehensive curve trend is not a curve of reducing loss, it is judged that the current training process is unreasonable and the extraction of key features and model training need to be carried out again; Initial model block: used to complete the model training of the first extended model based on a reasonable training process to obtain the initial abnormal behavior model; Function optimization block: used to determine the function generalization ability of the initial loss function based on the curve fluctuation of the first function curve; If the curve fluctuation is greater than the preset maximum curve fluctuation, it is judged that the function generalization ability of the initial loss function is poor and the initial loss function needs to be adjusted to obtain the first loss function; If the curve fluctuation is not greater than the preset maximum curve fluctuation, it is judged that the function generalization ability of the initial loss function is strong and the initial loss function is used as the first loss function; Model optimization block: used to randomly extract abnormal behavior data from the first abnormal behavior data as the test set, input it into the initial abnormal behavior model for model evaluation based on the test set, and optimize the initial abnormal behavior model based on the model evaluation results to obtain the first abnormal behavior model.

6. The personnel abnormal behavior detection and alarm system based on real-time video according to claim 3, characterized in that, Behavior detection module, including: First comparison unit: used to extract the current video frame data from the first video data and perform the first comparison on the current video frame data with the first normal behavior model; If the first comparison result determines that the current video frame data has abnormal behavior, the current video frame data is used as the first video frame data; Second comparison unit: used to perform the second comparison on the first video frame data with the first abnormal behavior model and determine the abnormal behavior type and abnormal behavior degree of the first video frame data based on the comparison result, so as to obtain the first behavior detection result of the target monitoring area.

7. The personnel abnormal behavior detection and alarm system based on real-time video according to claim 6, characterized in that, Alarm processing module, including: Alarm information determination unit: used to obtain the monitoring area information of the target monitoring area in real time and synthesize the first behavior detection result with the monitoring area information to determine the alarm information of the target monitoring area; Behavior alarm unit: used to issue various forms of alarms based on the alarm information to realize behavior alarm for the monitoring area.

8. The personnel abnormal behavior detection and alarm system based on real-time video according to claim 7, characterized in that Behavior alarm unit, including: Information analysis subunit: used to perform information classification and information analysis based on the alarm information to determine the information type of the alarm information; Scheme determination subunit: used to determine the alarm level and alarm scheme of the corresponding type based on the information type of the alarm information; Behavior alarm subunit: used to perform corresponding form of behavior alarm on the target monitoring area based on the alarm level and the corresponding alarm scheme.

Images