A fuel operation video monitoring system

Abstract

The present invention provides a fuel operation video monitoring system, which relates to the field of video monitoring technology and includes: a data acquisition module: real-time monitoring of the fuel operation process to obtain a first operation video; a data processing module: extracting target key frames from the first operation video; a behavior analysis module: inputting the target key frames into a behavior recognition model and outputting behavior recognition results corresponding to the business type; a display module: visualizing the operation video and visualizing violations when violations are found in the behavior recognition results. By preprocessing and extracting key frames from the first operation video obtained by real-time monitoring of the fuel operation process to obtain target key frames; inputting the target key frames into the behavior recognition model to output behavior recognition results; visualizing the operation videos of all business types and visualizing violations when violations are found in the behavior recognition results, effective monitoring and management of the fuel operation process can be achieved, which is conducive to improving fuel production efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of video monitoring, and in particular to a fuel operation video monitoring system. Background Art

[0002] During fuel operations, such as coal and oil, strict adherence to operational specifications and safety standards is essential for measuring, sampling, preparing, and testing fuels like coal and oil. However, traditional manual monitoring methods suffer from numerous shortcomings, including low efficiency, significant human error, and numerous safety hazards, leading to low video monitoring accuracy. Therefore, improving the accuracy and efficiency of video monitoring has become a key area of ​​research.

[0003] Therefore, the present invention provides a fuel operation video monitoring system. Summary of the Invention

[0004] The present invention provides a fuel operation video supervision system, which is used to obtain target key frames by preprocessing and extracting key frames from a first operation video obtained by real-time monitoring of the fuel operation process; inputting the target key frames into a behavior recognition model to output behavior recognition results; visualizing operation videos of all business types, and when violations are found in the behavior recognition results, visualizing the violations, thereby realizing effective monitoring and management of the fuel operation process, which is conducive to improving fuel production efficiency.

[0005] The present invention provides a fuel operation video monitoring system, comprising:

[0006] Data acquisition module: used to monitor the fuel operation process in real time using a set camera to obtain a first operation video;

[0007] Data processing module: used to extract key frames from the current first job video to obtain target key frames;

[0008] Behavior analysis module: used to input the target key frame into the behavior recognition model and output the behavior recognition result of the corresponding business type;

[0009] Display module: used to visualize the operation videos of all business types, and when there are violations in the behavior recognition results, visualize the violation alarms.

[0010] Preferably, the data acquisition module includes:

[0011] Using the real-time monitoring function of the camera, the fuel operation process is continuously recorded to obtain the first operation video;

[0012] Uploading the first operation video to the video monitoring platform by using the scheduled upload function of the camera;

[0013] Using the device number of the currently set camera as the filtering condition, filter the business type from the preset camera-type list and mark the first operation video.

[0014] Preferably, the business types include four types: measurement business, sampling business, sample preparation business and testing business.

[0015] Preferably, the data processing module includes:

[0016] Processing unit: used for performing denoising and contrast enhancement processing on the acquired first operation video to obtain a key operation video;

[0017] Frame extraction unit: used to extract key frames from the key operation video using a pre-established frame extraction model to obtain target key frames, and mark the target key frames with business types.

[0018] Preferably, the behavior analysis module includes:

[0019] Model building unit: used to obtain model training samples using sample blocks; use the model generation block to establish several behavior recognition models based on the training samples obtained from the sample blocks; use the model matching block to filter out the first recognition model from the behavior recognition models;

[0020] Sample acquisition block: used to retrieve historical operation videos of various business types from the video surveillance database to perform positive and negative sample analysis and generate the first training sample;

[0021] Model generation block: used to adopt each set target algorithm in the set algorithm list, obtain several behavior recognition models based on the first training sample training model, and establish a behavior recognition model list;

[0022] Regularly optimizing all models in the behavior recognition model list;

[0023] Model matching block: used to match the target recognition model for the current business type based on the behavior recognition model list;

[0024] Behavior recognition unit: used to input the current target key frame into the target recognition model of the same business type, obtain the behavior recognition result, and mark the business type on the behavior recognition result.

[0025] Preferably, the working process of the sample acquisition block includes:

[0026] Retrieve historical operation videos of various business types from the video surveillance database;

[0027] The historical homework videos are divided into standard behaviors and illegal behaviors, and the historical homework videos of standard behaviors are marked as the first positive samples, and the historical homework videos of illegal behaviors are marked as the first negative samples;

[0028] Counting the number and type of first positive samples and first negative samples corresponding to each business type;

[0029] If there are first positive samples and first negative samples corresponding to the business type whose sample types and sample quantities meet the corresponding set training sample conditions, then the first positive sample and first negative sample corresponding to the current business type are output as the first training samples;

[0030] If the sample type or sample quantity of the first positive sample corresponding to the business type does not meet the corresponding set training sample conditions, the current business type is marked as a positive sample supplement type;

[0031] Extracting the operation specification description of the positive sample supplement type from the business standard library for artificial simulation collection, obtaining a simulated positive sample to supplement the first positive sample of the current positive sample supplement type and outputting it as the first training sample;

[0032] When the sample type or sample quantity of the first negative sample corresponding to the business type does not meet the corresponding set training sample conditions, the current business type is marked as a negative sample supplement type;

[0033] The operation violation description of the negative sample supplement type is extracted from the business standard library for artificial simulation collection, and the simulated negative sample is obtained to supplement the first negative sample of the current negative sample supplement type and output as the first training sample.

[0034] Preferably, the working process of the model matching block includes:

[0035] From the model recognition database, extract the historical recognition records of each corresponding behavior recognition model using the current business type as the screening condition;

[0036] Marking the behavior recognition model with historical recognition records as the first recognition model;

[0037] Marking the behavior recognition model that does not have a historical recognition record as the second recognition model;

[0038] Extracting historical recognition accuracy and historical recognition time from all historical recognition records of the first recognition model for analysis to determine a first recognition effect coefficient of the first recognition model;

[0039] Determining a first recognition effect coefficient of the current second recognition model according to the simulation recognition accuracy and simulation time of the second recognition model;

[0040] When there is only a single recognition model whose first recognition effect coefficient is not less than the set recognition accuracy threshold, the current recognition model is output as the target recognition model for the current business type;

[0041] When there are multiple recognition models whose first recognition effect coefficients are not less than the set recognition accuracy threshold, the recognition model with the largest first recognition effect coefficient among the current recognition models is output as the target recognition model for the current business type;

[0042] When the first recognition effect coefficients of all recognition models are less than the set recognition accuracy threshold, the reference performance influence coefficient of the current recognition model is extracted from the set model influence mapping table and combined with the first recognition effect coefficient to calculate the recognition evaluation coefficient of the recognition model;

[0043] The recognition model with the highest recognition evaluation coefficient is output as the target recognition model for the current business type;

[0044] The calculation formula for the identification evaluation coefficient is as follows:

[0045] Where, Represented as the recognition evaluation coefficient of the current recognition model; It is expressed as the first recognition effect coefficient of the current recognition model; It is expressed as the weight of the impact of the model recognition effect on the recognition performance of the analysis model; It is expressed as the weight of the influence of the model performance on the recognition performance of the analysis model; It is expressed as the reference performance influence coefficient of the current recognition model; Represents the optimization time period of the current recognition model; Indicates the time between the current recognition model and the last model optimization; Expressed as the model degradation learning factor.

[0046] Preferably, the display module includes:

[0047] Comprehensive display unit: used by the video supervision platform to classify, store and visualize the first operation videos received regularly according to business types, and provide video display functions;

[0048] Violation visualization unit: used to analyze behavior recognition results. When the obtained behavior recognition results contain violations, the violation display area is set in the video display interface through the highlighting function to visualize the violation and the corresponding violation-related information;

[0049] User interaction unit: used to assign corresponding access rights to target users based on their departments;

[0050] The target user uses a smart device to log in to the video supervision platform and, based on the assigned access rights, has limited access to the operation video and violation records.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] The target key frame is obtained by preprocessing and key frame extraction of the first operation video obtained by real-time monitoring of the fuel operation process; the target key frame is input into the behavior recognition model to output the behavior recognition result; the operation video of all business types is visualized, and when there are violations in the behavior recognition results, the violations are visualized. This can realize effective monitoring and management of the fuel operation process, which is conducive to improving fuel production efficiency.

[0053] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0054] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0056] Figure 1 This is a structural diagram of a fuel operation video monitoring system in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0058] The embodiment of the present invention provides a fuel operation video monitoring system, such as Figure 1 Shown, including:

[0059] Data acquisition module: used to monitor the fuel operation process in real time using a set camera to obtain a first operation video;

[0060] Data processing module: used to extract key frames from the current first job video to obtain target key frames;

[0061] Behavior analysis module: used to input the target key frame into the behavior recognition model and output the behavior recognition result of the corresponding business type;

[0062] Display module: used to visualize the operation videos of all business types, and when there are violations in the behavior recognition results, visualize the violation alarms.

[0063] In this embodiment, the set camera refers to a pre-installed camera for clearly capturing the business operation process; the first operation video refers to a video obtained by using the set camera to monitor the corresponding fuel operation process of different business types in real time; the target key frame is the key frame extracted from the pre-processed first operation video; the behavior recognition result includes behavior category, behavior judgment and behavior confidence, wherein the behavior category refers to the identified behavior type, such as personnel intervention, personnel leaving the post, vehicle weighing, etc.; behavior judgment refers to the judgment result of whether the current behavior is normative or illegal. If the currently identified behavior is judged to be normative, the identified behavior is described as normative behavior. If the currently identified behavior is judged to be illegal, the identified behavior is described as illegal behavior; behavior confidence refers to the credibility of the recognition result.

[0064] The beneficial effects of the above technical solution are: the target key frame is obtained by preprocessing and key frame extraction of the first operation video obtained by real-time monitoring of the fuel operation process; the target key frame is input into the behavior recognition model to output the behavior recognition result; the operation videos of all business types are visualized, and when there are violations in the behavior recognition results, the violations are visualized, which can realize effective monitoring and management of the fuel operation process, which is conducive to improving fuel production efficiency.

[0065] An embodiment of the present invention provides a fuel operation video monitoring system, wherein the data acquisition module includes:

[0066] Using the real-time monitoring function of the camera, the fuel operation process is continuously recorded to obtain the first operation video;

[0067] Uploading the first operation video to the video monitoring platform by using the scheduled upload function of the camera;

[0068] Using the device number of the currently set camera as the filtering condition, filter the business type from the preset camera-type list and mark the first operation video.

[0069] In this embodiment, the set camera refers to a pre-installed camera for clearly capturing the business operation process; the real-time monitoring function refers to the set camera monitoring the fuel operation process in real time and recording the fuel operation process; the periodic upload function refers to the set camera uploading the first operation video to the video supervision platform according to a preset time period; the first operation video refers to the video obtained by using the set camera to monitor the corresponding fuel operation process of different business types in real time; the equipment number refers to the unique identifier assigned to each set camera; the preset camera-type list consists of the equipment number of the set camera and the business type of the fuel operation; the business type is a type screened out from the preset camera-type list based on the matching condition of the equipment number of the current set camera, including four types of measurement business, sampling business, sample preparation business and testing business.

[0070] The beneficial effect of the above technical solution is: by utilizing the real-time monitoring function of the set camera, the fuel operation process is continuously recorded to obtain the first operation video, and the first business type is marked on the first operation video, which can provide data support for subsequent operation video supervision.

[0071] An embodiment of the present invention provides a fuel operation video monitoring system, wherein the data processing module includes:

[0072] Processing unit: used for performing denoising and contrast enhancement processing on the acquired first operation video to obtain a key operation video;

[0073] Frame extraction unit: used to extract key frames from the key operation video using a pre-established frame extraction model to obtain target key frames, and mark the target key frames with business types.

[0074] In this embodiment, the first operation video refers to a video obtained by using a set camera to monitor the corresponding fuel operation process of different business types in real time; the set camera refers to a pre-installed camera for clearly capturing the business operation process; the key operation video is obtained after denoising and contrast enhancement processing of the acquired first operation video; the target key frame is a key frame obtained by extracting the key frame of the key operation video using a frame extraction model; the frame extraction model is obtained by training the convolutional neural network using training data established after preprocessing and feature extraction of a preset amount of operation video data containing all business types, wherein the extracted features include high-level semantic features, low-level visual features, etc. High-level semantic features refer to semantic information related to specific objects, scenes or behaviors expressed in video frames, such as character actions and object categories; low-level semantic features refer to basic visual information in video frames, including color, texture, shape, etc.

[0075] The beneficial effect of the above technical solution is: by denoising and enhancing the contrast of the acquired first operation video and then extracting the key frame, the target key frame is obtained, which can provide data support for subsequent operation video supervision.

[0076] An embodiment of the present invention provides a fuel operation video monitoring system, wherein the behavior analysis module includes:

[0077] Model building unit: used to obtain model training samples using sample blocks; use the model generation block to establish several behavior recognition models based on the training samples obtained from the sample blocks; use the model matching block to filter out the first recognition model from the behavior recognition models;

[0078] Sample acquisition block: used to perform positive and negative sample analysis on historical operation videos of various business types retrieved from the video surveillance database to generate the first training sample;

[0079] Model generation block: used to adopt each set target algorithm in the set algorithm list, obtain several behavior recognition models based on the first training sample training model, and establish a behavior recognition model list;

[0080] Regularly optimizing all models in the behavior recognition model list;

[0081] Model matching block: used to match the target recognition model for the current business type based on the behavior recognition model list;

[0082] Behavior recognition unit: used to input the current target key frame into the target recognition model of the same business type, obtain the behavior recognition result, and mark the business type on the behavior recognition result.

[0083] In this embodiment, the video surveillance database is composed of historical operation videos of various business types, where the business types include four types: measurement business, sampling business, sample preparation business, and testing business; the set algorithm list refers to a list composed of multiple set target algorithms, including convolutional neural networks, recurrent neural networks, and dual-stream networks, etc.

[0084] In this embodiment, the behavior recognition model list is composed of several behavior recognition models obtained by using a set target algorithm and a first training sample training model, wherein the first training sample is obtained by performing positive and negative sample analysis and supplementation on historical homework videos; optimization processing refers to regularly optimizing all models in the behavior recognition model list according to the corresponding set model optimization strategy, for example, optimizing all models in the behavior recognition model list according to the corresponding set model optimization strategy every 3 months; wherein the set model optimization strategy includes model parameter adjustment, training data enhancement, etc.

[0085] In this embodiment, the target recognition model refers to a behavior recognition model that is selected from the behavior recognition model list and is used to identify and process the target key frames of the current business type; the behavior recognition result includes behavior category, behavior judgment, and behavior confidence, wherein the behavior category refers to the identified behavior type, such as personnel intervention, personnel leaving the post, vehicle weighing, etc.; behavior judgment refers to the judgment result of whether the current behavior is normative or in violation of the rules. If the currently identified behavior is judged to be normative, the identified behavior is described as normative behavior. If the currently identified behavior is judged to be in violation of the rules, the identified behavior is described as in violation of the rules; behavior confidence refers to the credibility of the recognition result.

[0086] The beneficial effect of the above technical solution is: by analyzing the dynamic priority of each pre-established behavior recognition model, screening the behavior recognition model with the largest dynamic priority coefficient to perform behavior recognition on the acquired target key frame, which can effectively ensure the accuracy of behavior recognition.

[0087] An embodiment of the present invention provides a fuel operation video monitoring system, wherein the working process of the sample acquisition block includes:

[0088] Retrieve historical operation videos of various business types from the video surveillance database;

[0089] The historical homework videos are divided into standard behaviors and illegal behaviors, and the historical homework videos of standard behaviors are marked as the first positive samples, and the historical homework videos of illegal behaviors are marked as the first negative samples;

[0090] Counting the number and type of first positive samples and first negative samples corresponding to each business type;

[0091] If there are first positive samples and first negative samples corresponding to the business type whose sample types and sample quantities meet the corresponding set training sample conditions, then the first positive sample and first negative sample corresponding to the current business type are output as the first training samples;

[0092] If the sample type or sample quantity of the first positive sample corresponding to the business type does not meet the corresponding set training sample conditions, the current business type is marked as a positive sample supplement type;

[0093] Extracting the operation specification description of the positive sample supplement type from the business standard library for artificial simulation collection, obtaining a simulated positive sample to supplement the first positive sample of the current positive sample supplement type and outputting it as the first training sample;

[0094] When the sample type or sample quantity of the first negative sample corresponding to the business type does not meet the corresponding set training sample conditions, the current business type is marked as a negative sample supplement type;

[0095] The operation violation description of the negative sample supplement type is extracted from the business standard library for artificial simulation collection, and the simulated negative sample is obtained to supplement the first negative sample of the current negative sample supplement type and output as the first training sample.

[0096] In this embodiment, the video surveillance database is composed of historical operation videos of various business types, where the business types include four types: measurement business, sampling business, sample preparation business, and testing business; the first positive sample refers to the historical operation video of standardized behavior; the first negative sample refers to the historical operation video of illegal behavior; the setting of the training sample size is predetermined.

[0097] In this embodiment, sample types include three types: behavior types, scene types, and behavior patterns, among which behavior types refer to different behaviors of people or equipment, such as carrying goods and weighing vehicles; scene types refer to work scenes, such as warehouses and production lines; behavior patterns refer to specific behavior requirements, such as carrying with both hands and using equipment for sampling; setting training sample conditions means that the sample types are complete and the number of samples reaches a preset number; positive sample supplement type refers to a business type whose sample type or sample number of the first positive sample does not meet the corresponding set training sample conditions; operation specification description refers to a written description of the standardized operation actions of each business type determined in advance by the enterprise; simulated positive samples refer to videos obtained by artificial simulation based on operation specification descriptions; negative sample supplement type refers to a business type whose sample type or sample number of the first negative sample does not meet the corresponding set training sample conditions.

[0098] In this embodiment, the operation violation description refers to a written description of the illegal operation actions of each business type determined in advance by the enterprise. For example, the operation violation description of the measurement business includes the inconsistency of the number of people on the vehicle twice when weighing a heavy vehicle and a light vehicle; the operation violation description of the sampling business includes opening the sampling machine observation port / sample collection barrel, the car compartment sampling machine, and people entering sensitive areas when the sampling machine is running, etc. during the operation of the sampling machine; the simulated negative sample refers to the video obtained by artificial simulation based on the operation violation description.

[0099] The beneficial effect of the above technical solution is: by performing quantitative analysis on the obtained positive and negative samples, and when the samples do not meet the set conditions, using artificial simulation to enrich the training samples, it can provide effective data support for model establishment, thereby ensuring the recognition performance of the behavior recognition model.

[0100] An embodiment of the present invention provides a fuel operation video monitoring system, wherein the working process of the model matching block includes:

[0101] From the model recognition database, extract the historical recognition records of each corresponding behavior recognition model using the current business type as the screening condition;

[0102] Marking the behavior recognition model with historical recognition records as the first recognition model;

[0103] Marking the behavior recognition model that does not have a historical recognition record as the second recognition model;

[0104] Extracting historical recognition accuracy and historical recognition time from all historical recognition records of the first recognition model for analysis to determine a first recognition effect coefficient of the first recognition model;

[0105] Determining a first recognition effect coefficient of the current second recognition model according to the simulation recognition accuracy and simulation time of the second recognition model;

[0106] When there is only a single recognition model whose first recognition effect coefficient is not less than the set recognition accuracy threshold, the current recognition model is output as the target recognition model for the current business type;

[0107] When there are multiple recognition models whose first recognition effect coefficients are not less than the set recognition accuracy threshold, the recognition model with the largest first recognition effect coefficient among the current recognition models is output as the target recognition model for the current business type;

[0108] When the first recognition effect coefficients of all recognition models are less than the set recognition accuracy threshold, the reference performance influence coefficient of the current recognition model is extracted from the set model influence mapping table and combined with the first recognition effect coefficient to calculate the recognition evaluation coefficient of the recognition model;

[0109] The recognition model with the highest recognition evaluation coefficient is output as the target recognition model for the current business type;

[0110] The calculation formula for the identification evaluation coefficient is as follows:

[0111] Where, Represented as the recognition evaluation coefficient of the current recognition model; It is expressed as the first recognition effect coefficient of the current recognition model; It is expressed as the weight of the impact of the model recognition effect on the recognition performance of the analysis model; It is expressed as the weight of the influence of the model performance on the recognition performance of the analysis model; It is expressed as the reference performance influence coefficient of the current recognition model; Represents the optimization time period of the current recognition model; Indicates the time between the current recognition model and the last model optimization; Expressed as the model degradation learning factor.

[0112] In this embodiment, the model recognition database is used to store historical recognition records of each behavior recognition model corresponding to each business type; the historical recognition records include historical recognition accuracy, historical recognition duration, historical recognition start time, historical recognition end time, and historical behavior recognition results, etc.; the first recognition model refers to a behavior recognition model with historical recognition records; the second recognition model refers to a behavior recognition model with no historical recognition records.

[0113] In this embodiment, the calculation formula of the first recognition effect coefficient of the first recognition model is as follows:

[0114] Where, It is represented as the first recognition effect coefficient of the current first recognition model; It is expressed as the average historical recognition accuracy of the current first recognition model; It is represented by the average historical length of the current first recognition model; Expressed as the average of the average history lengths of all first-identification models; It is expressed as the weight of the impact of recognition accuracy on the recognition effect of the analysis model; It is expressed as the weight of the impact of recognition time on the recognition effect of the analysis model; e is expressed as the natural base number;

[0115] The calculation formula of the first recognition effect coefficient of the second recognition model is as follows:

[0116] Where, 2 represents the first recognition effect coefficient of the current second recognition model; It is represented as the simulated recognition accuracy of the current second recognition model; It is represented by the simulation recognition time of the current second recognition model; It is expressed as the weight of the impact of recognition accuracy on the recognition effect of the analysis model; It is expressed as the weight of the impact of recognition time on the recognition effect of the analysis model; e is expressed as the natural base number;

[0117] Among them, the influence weights of recognition accuracy and recognition time on the recognition effect of the analysis model are obtained by solving the matrix constructed after pairwise comparison and relative importance scoring using the hierarchical analysis method.

[0118] In this embodiment, the recognition effect threshold is set in advance and is generally 0.85; the model impact mapping table is set to consist of the business type, the behavior recognition model corresponding to the business type, the initial performance impact coefficient of the behavior recognition model corresponding to the business type, and the performance impact coefficient after each model optimization, wherein the initial performance impact coefficient refers to the performance impact coefficient of the behavior recognition model that has not been optimized; the reference performance impact coefficient refers to the performance impact coefficient of the current recognition model extracted from the set model impact mapping table, which is the closest to the current moment after optimization; the recognition evaluation coefficient is used to evaluate the model recognition effect of the current recognition model.

[0119] In this embodiment, the performance impact coefficient is obtained by weighted averaging the performance indicator parameters obtained by evaluating the behavior recognition model using the set performance impact indicators, wherein the set performance impact indicators include data quality, algorithm complexity, and required computing resources; the weights assigned to the performance indicator parameters are obtained by solving the matrix constructed after pairwise comparison and relative importance scoring using the hierarchical analysis method.

[0120] The beneficial effect of the above technical solution is that by performing dynamic performance evaluation and historical recognition accuracy analysis on each behavior recognition model, the most suitable behavior recognition model can be screened out for behavior recognition, thereby effectively improving the accuracy of behavior recognition.

[0121] An embodiment of the present invention provides a fuel operation video monitoring system, wherein the display module includes:

[0122] Comprehensive display unit: used by the video supervision platform to classify, store and visualize the first operation videos received regularly according to business types, and provide video display functions;

[0123] Violation visualization unit: used to analyze behavior recognition results. When the obtained behavior recognition results contain violations, the violation display area is set in the video display interface through the highlighting function to visualize the violation and the corresponding violation-related information;

[0124] User interaction unit: used to assign corresponding access rights to target users based on their departments;

[0125] The target user uses a smart device to log in to the video supervision platform and, based on the assigned access rights, has limited access to the operation video and violation records.

[0126] In this embodiment, the video supervision platform is used to achieve efficient management, classified storage, visual display, and intelligent identification and processing of regularly received operation videos; business types include measurement business, sampling business, sample preparation business, and testing business.

[0127] In this embodiment, the video display function includes full-screen playback, list playback, fast forward and other functions; the highlight display function refers to marking the specific location and time of the violation by highlighting in the set violation display area of ​​the video display interface; the violation-related information includes the violation type, violation time and violation location, etc.; the set violation display area refers to the area used to display the violation and the corresponding violation-related information; the target user refers to the user who successfully logs in to the video supervision platform; the purpose of assigning corresponding access rights to the target user according to the department to which the target user belongs is to ensure that the user can only access data within his or her authority, thereby improving the security and confidentiality of the data; smart devices include mobile phones, tablets, etc., which can achieve remote access.

[0128] The beneficial effect of the above technical solution is: by classifying and storing the first operation video according to business type, highlighting and visualizing violations, supervisors can view the video content in real time or on demand, which helps to promptly discover potential problems or violations.

[0129] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A fuel operation video monitoring system, characterized in that: include: Data acquisition module: used to monitor the fuel operation process in real time using a set camera to obtain a first operation video; Data processing module: used to extract key frames from the current first job video to obtain target key frames; Behavior analysis module: used to input the target key frame into the behavior recognition model and output the behavior recognition result of the corresponding business type; Display module: used to visualize operation videos of all business types and visualize violation alerts when violations are found in the behavior recognition results; The behavior analysis module includes: Model building unit: used to obtain model training samples using sample blocks; use the model generation block to establish several behavior recognition models based on the training samples obtained from the sample blocks; use the model matching block to filter out the first recognition model from the behavior recognition models; Sample acquisition block: used to retrieve historical operation videos of various business types from the video surveillance database to perform positive and negative sample analysis and generate the first training sample; Model generation block: used to adopt each set target algorithm in the set algorithm list, obtain several behavior recognition models based on the first training sample training model, and establish a behavior recognition model list; Regularly optimizing all models in the behavior recognition model list; Model matching block: used to match the target recognition model for the current business type based on the behavior recognition model list; Behavior recognition unit: used to input the current target key frame into the target recognition model of the same business type, obtain the behavior recognition result, and mark the business type on the behavior recognition result; The working process of the model matching block includes: From the model recognition database, extract the historical recognition records of each corresponding behavior recognition model using the current business type as the screening condition; Marking the behavior recognition model with historical recognition records as the first recognition model; Marking the behavior recognition model that does not have a historical recognition record as the second recognition model; Extracting historical recognition accuracy and historical recognition time from all historical recognition records of the first recognition model for analysis to determine a first recognition effect coefficient of the first recognition model; Determining a first recognition effect coefficient of the current second recognition model according to the simulation recognition accuracy and simulation time of the second recognition model; When there is only a single recognition model whose first recognition effect coefficient is not less than the set recognition accuracy threshold, the current recognition model is output as the target recognition model for the current business type; When there are multiple recognition models whose first recognition effect coefficients are not less than the set recognition accuracy threshold, the recognition model with the largest first recognition effect coefficient among the current recognition models is output as the target recognition model for the current business type; When the first recognition effect coefficients of all recognition models are less than the set recognition accuracy threshold, the reference performance influence coefficient of the current recognition model is extracted from the set model influence mapping table and combined with the first recognition effect coefficient to calculate the recognition evaluation coefficient of the recognition model; The recognition model with the highest recognition evaluation coefficient is output as the target recognition model for the current business type; The calculation formula for the identification evaluation coefficient is as follows: Where, Represented as the recognition evaluation coefficient of the current recognition model; It is expressed as the first recognition effect coefficient of the current recognition model; It is expressed as the weight of the impact of the model recognition effect on the recognition performance of the analysis model; It is expressed as the weight of the influence of the model performance on the recognition performance of the analysis model; It is expressed as the reference performance influence coefficient of the current recognition model; Represents the optimization time period of the current recognition model; Indicates the time between the current recognition model and the last model optimization; Expressed as the model degradation learning factor.

2. A fuel operation video monitoring system according to claim 1, characterized in that: The data acquisition module includes: Using the real-time monitoring function of the camera, the fuel operation process is continuously recorded to obtain the first operation video; Uploading the first operation video to the video monitoring platform by using the scheduled upload function of the camera; Using the device number of the currently set camera as the filtering condition, filter the business type from the preset camera-type list and mark the first operation video.

3. A fuel operation video monitoring system according to claim 2, characterized in that: Business types include measurement business, sampling business, sample preparation business and testing business.

4. A fuel operation video monitoring system according to claim 1, characterized in that: The data processing module includes: Processing unit: used for performing denoising and contrast enhancement processing on the acquired first operation video to obtain a key operation video; Frame extraction unit: used to extract key frames from the key operation video using a pre-established frame extraction model to obtain target key frames, and mark the target key frames with business types.

5. A fuel operation video monitoring system according to claim 1, characterized in that: The working process of the sample acquisition block includes: Retrieve historical operation videos of various business types from the video surveillance database; The historical homework videos are divided into standard behaviors and illegal behaviors, and the historical homework videos of standard behaviors are marked as the first positive samples, and the historical homework videos of illegal behaviors are marked as the first negative samples; Counting the number and type of first positive samples and first negative samples corresponding to each business type; If there are first positive samples and first negative samples corresponding to the business type whose sample types and sample quantities meet the corresponding set training sample conditions, then the first positive sample and first negative sample corresponding to the current business type are output as the first training samples; If the sample type or sample quantity of the first positive sample corresponding to the business type does not meet the corresponding set training sample conditions, the current business type is marked as a positive sample supplement type; Extracting the operation specification description of the positive sample supplement type from the business standard library for artificial simulation collection, obtaining a simulated positive sample to supplement the first positive sample of the current positive sample supplement type and outputting it as the first training sample; When the sample type or sample quantity of the first negative sample corresponding to the business type does not meet the corresponding set training sample conditions, the current business type is marked as a negative sample supplement type; The operation violation description of the negative sample supplement type is extracted from the business standard library for artificial simulation collection, and the simulated negative sample is obtained to supplement the first negative sample of the current negative sample supplement type and output as the first training sample.

6. A fuel operation video monitoring system according to claim 1, characterized in that: The display module includes: Comprehensive display unit: used by the video supervision platform to classify, store and visualize the first operation videos received regularly according to business types, and provide video display functions; Violation visualization unit: used to analyze behavior recognition results. When the obtained behavior recognition results contain violations, the violation display area is set in the video display interface through the highlighting function to visualize the violation and the corresponding violation-related information; User interaction unit: used to assign corresponding access rights to target users based on their departments; The target user uses a smart device to log in to the video supervision platform and, based on the assigned access rights, has limited access to the operation video and violation records.

Images