A surveillance video classification processing and push method based on data analysis

Abstract

The present invention belongs to the field of video processing technology, and specifically discloses a monitoring video classification processing and pushing method based on data analysis. The method collects environmental data and confirms the appropriate interval duration of frequency frame pulling, effectively solves the problem of insufficient adaptability of current video monitoring processing scenarios, ensures the rationality of monitoring video frame pulling, and thus can ensure the effectiveness of monitoring and the timeliness of abnormal response in a frequently and rapidly changing environment. At the same time, by setting algorithm models of multiple independent threads for video frame recognition, the video frames can be analyzed at the same time, which greatly shortens the overall processing time. In addition, by confirming the appropriate interval duration of frequency frame pulling, setting the algorithm models of multiple independent threads and setting the compression ratio, the capture efficiency and processing efficiency of key events are improved as much as possible on the basis of ensuring that video information can be obtained in time during critical periods, thereby improving navigation security.

Description

Technical Field

[0001] The present invention belongs to the technical field of video processing, and in particular, relates to a monitoring video classification processing and pushing method based on data analysis. Background Art

[0002] The maritime environment is complex and changeable, and ships face various risks when sailing at sea, such as bad weather, pirate attacks, mechanical failures, etc. By installing surveillance cameras, potential safety hazards can be discovered in a timely manner, and by classifying and processing surveillance videos, all aspects of the ship can be better managed.

[0003] Prior art, such as a Chinese invention patent application with application number 201910467214.2, discloses a method and system for publishing video information based on case scheduling, which includes: obtaining video surveillance data. Performing feature analysis on the video surveillance data and classifying the features according to preset feature extraction rules to generate multiple categories of video feature information. Obtaining case information of the case to be processed. Extracting video data related to the case to be processed from multiple categories of video feature information based on the case information. Sending the video data to each preset terminal according to the preset authority rules. In this way, a large amount of video surveillance data is automatically screened, the processing efficiency of the video data is improved, and multiple categories of video information are further screened according to the specific case information of the case to be processed to obtain video information related to the case to be processed, and by automatically pushing the screened video data to each preset terminal according to the preset authority rules, there is no need for manual extraction, which saves time and effort, and further improves the efficiency of case handling.

[0004] Another example of the prior art is a gas station oil unloading site safety supervision method and system disclosed in the Chinese invention patent application with application number 202010356918.5, which performs real-time safety monitoring of the gas station oil unloading operation through the gas station oil unloading site safety supervision system, including the gas station oil unloading site safety supervision system collecting real-time video and / or real-time image of each monitored object in the gas station oil unloading operation. All real-time videos and / or real-time images are classified and imported into the preset image recognition model corresponding to each monitored object for identification, and the detection results of whether there are safety hazards for each monitored object are obtained respectively. The detection results with safety hazards are warned, and the monitored objects with safety hazards and their processing results after intelligent identification and abnormal analysis are pushed to the user, thereby realizing unmanned supervision of the gas station oil unloading operation, and real-time alarms are issued for behaviors that do not meet the requirements and norms in the oil unloading operation, effectively reducing the occurrence of safety accidents in the oil unloading operation.

[0005] Regarding the above two technical solutions, it is obvious that the current classification processing of surveillance videos is biased towards the scene in which it is located and is suitable for relatively static and stable scenes. The scenes are relatively simple and insufficient attention is paid to the environmental factors in the scenes. There are also problems in the following aspects: 1. Insufficient scene adaptability. The light in the marine environment changes very frequently and the stability is poor. The impact of the environment is not currently considered, making it difficult for subsequent recognition and classification effects to meet expectations.

[0006] 2. The current processing efficiency is difficult to adapt to maritime scenarios. In the event of an emergency, maritime warnings need to respond quickly, and timeliness is required. Current video processing and push technologies pay less attention to processing speed, and thus cannot achieve timely response. Summary of the invention

[0007] In view of this, in order to solve the problems raised in the above background technology, a surveillance video classification processing and pushing method based on data analysis is now proposed.

[0008] The purpose of the present invention can be achieved through the following technical solutions: The present invention provides a monitoring video classification processing and push method based on data analysis, the method comprising: S1, monitoring environment data collection: real-time environment data collection through pre-installed environment sensor equipment.

[0009] S2. Monitoring video data acquisition: Determine the appropriate interval duration for pulling video frames corresponding to each monitoring key area, and pull video frames from the corresponding cameras installed in each pre-set monitoring key area according to the appropriate interval duration.

[0010] S3. Classification and packaging of surveillance videos: Classify and package the video frames of each key surveillance area to obtain video frames of various types, and transmit them to the corresponding pre-set algorithm models.

[0011] S4. Video recognition and feature construction: The transmitted video frames are recognized through the pre-set algorithm models to obtain the recognition results.

[0012] S5. Video push setting processing: Perform push setting analysis and output the push strategy corresponding to the output results of each algorithm model.

[0013] S6. Monitoring video push: Based on the pre-set user permissions, the push strategy of the corresponding output results of each algorithm model and the video features, corresponding push is performed.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) By collecting environmental data and confirming the appropriate interval duration of frame pulling, the present invention effectively solves the problem of insufficient adaptability of current video surveillance processing scenarios, ensures the rationality of monitoring video frame pulling, and thus can ensure the effectiveness of monitoring and the timeliness of abnormal response in a frequently and rapidly changing environment, while also ensuring the subsequent processing efficiency and the subsequent identification and classification effects to a certain extent.

[0015] (2) The present invention performs video frame recognition by setting up multiple independent thread algorithm models. First, the video frames can be analyzed simultaneously, which greatly shortens the overall processing time. Second, different key areas and different video frames can be detected simultaneously to quickly obtain analysis results. At the same time, the stability of video frame recognition is increased, avoiding mutual interference and mutual influence in the processing of different video frames.

[0016] (3) The present invention improves the capture efficiency and processing efficiency of key events as much as possible on the basis of ensuring that video information can be obtained in a timely manner during critical periods by confirming the appropriate interval length of frame pulling, setting an algorithm model of multiple independent threads and setting a compression ratio. It effectively solves the problem that the current processing efficiency is difficult to adapt to marine scenes, and can ensure that marine warnings can respond quickly in emergencies, thereby ensuring the timeliness of warning responses. At the same time, a reasonable compression ratio can reduce the pressure of data transmission and storage while ensuring video quality, thereby ensuring the processing speed of video processing and push. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 The present invention is a schematic flow chart of the steps for implementing the method. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1As shown, the present invention provides a monitoring video classification processing and pushing method based on data analysis, the method comprising: S1, monitoring environment data collection: real-time environment data collection through pre-installed environment sensor equipment.

[0021] S2. Monitoring video data acquisition: Determine the appropriate interval duration for pulling video frames corresponding to each monitoring key area, and pull video frames from the corresponding cameras installed in each pre-set monitoring key area according to the appropriate interval duration.

[0022] S3. Classification and packaging of surveillance videos: Classify and package the video frames of each key surveillance area to obtain video frames of various types, and transmit them to the corresponding pre-set algorithm models.

[0023] S4. Video recognition and feature construction: The transmitted video frames are recognized through the pre-set algorithm models to obtain the recognition results.

[0024] S5. Video push setting processing: Perform push setting analysis and output the push strategy corresponding to the output results of each algorithm model.

[0025] S6. Monitoring video push: Based on the pre-set user permissions, the push strategy of the corresponding output results of each algorithm model and the video features, corresponding push is performed.

[0026] Regarding step S1, exemplarily, the environmental sensing equipment is composed of meteorological sensors, sea condition sensors, light sensors, etc., wherein the meteorological sensors include wind speed sensors, wind direction sensors, air pressure sensors, humidity sensors, etc., and the sea condition sensors include wave sensors, water flow sensors, etc.

[0027] It should be added that sea condition information such as wave height and water flow speed is crucial to the navigation safety of ships. In severe sea conditions, ships may face greater risks and need to pull video frames more frequently to monitor the situation in key areas. The different lighting conditions during the day and at night also affect the effect of video monitoring. During the day with sufficient light, the video frame pulling time interval can be appropriately extended. At night or in low light conditions, the time interval needs to be shortened to ensure the effectiveness of monitoring. Changes in wind speed and direction may affect the stability and navigation safety of the ship, and changes in humidity may affect the operation of the equipment. The data from these sensors can be used to judge the current marine meteorological conditions, thereby adjusting the video frame pulling time interval. Therefore, the present invention mainly monitors from the dimensions of weather, sea conditions and lighting.

[0028] In a specific embodiment, the meteorological sensor data includes, but is not limited to, detection values ​​of meteorological detection indicators such as wind speed, wind direction angle, air pressure, and humidity at each monitoring time point.

[0029] Regarding step S2, it should be added that the monitoring and control areas include the mast and antenna area, cockpit area, deck area, bow area, stern area, engine room area, etc.

[0030] It should be added that frequent pulling of video frames may cause network congestion and affect the transmission of other important data. In this case, it is necessary to reasonably adjust the video frame pulling time interval according to the actual situation of the network bandwidth to ensure the stable operation of the system. At the same time, a large amount of video frame data requires a certain amount of storage capacity. If the storage capacity is limited, pulling video frames too frequently may cause the storage device to fill up quickly. Therefore, it is necessary to consider the limitation of storage capacity and appropriately adjust the video frame pulling time interval to balance data storage requirements and system performance. At the same time, it is difficult to detect the risks in time if the interval is large, so it is necessary to set the interval length for pulling video frames.

[0031] Specifically, determining the appropriate interval duration for pulling video frames corresponding to each key monitoring area includes: S21, extracting meteorological sensor data from the real-time collected environmental data, and then extracting the detection value of each meteorological detection index corresponding to each monitoring time point.

[0032] S22, extract the meteorological interference monitoring area set from the database, set the meteorological interference factor of each key monitoring area, and record it as , Indicates the monitoring key area number. .

[0033] S23, extracting sea condition sensor data from the real-time collected environmental data, and then extracting the detection value of each sea condition detection index corresponding to each monitoring time point, and at the same time extracting the sea condition interference monitoring area set from the database, and setting the sea condition interference factor of each key monitoring area in the same way as the setting method of the meteorological interference factor of each key monitoring area, recorded as .

[0034] S24, extracting the real-time detected light brightness corresponding to each key monitoring area from the real-time collected environmental data, and setting the light interference factor of each monitoring area, recorded as .

[0035] S25, record the pre-set video frame pulling interval length corresponding to each key monitoring area as , calculate the appropriate interval length for pulling video frames corresponding to each key monitoring area , , , and are the weight coefficients of meteorological interference factor, sea state interference factor and light and heat interference factor, .

[0036] It should be added that, on seagoing vessels, sea conditions have a huge impact on the safety of the ships. Severe sea conditions such as high waves and strong currents may cause the ship to shake violently, increasing the risk of collision, grounding or even capsizing. Key monitoring areas such as the ship's side area, bow area, and stern area are directly affected by sea conditions, and may have problems such as structural damage and equipment failure. For example, large waves hitting the ship's side may cause leakage, and strong currents may affect the maneuverability of the ship. In addition, sea conditions change relatively frequently and are difficult to predict, so a larger weight coefficient is given. Meteorological conditions such as wind speed, wind direction, and air pressure will also have an important impact on ships. Strong winds may cause ships to lose stability, and heavy rains may affect vision and equipment operation. Compared with sea conditions, meteorological changes are usually relatively slow, and ships can obtain certain information in advance through weather forecasts and other means to make corresponding preparations. Therefore, its impact may not be as direct and drastic as the sea condition, so the weight coefficient is smaller than the sea condition interference factor. Light directly determines visibility, while temperature affects safety. Frequent temperature changes or light changes directly reflect that there are certain abnormalities in the ship, but the impact is not as significant as the first two. In general, its weight coefficient is relatively small. For example, , and The possible values ​​are 0.45, 0.35 and 0.2 respectively.

[0037] The embodiments of the present invention effectively solve the problem of insufficient adaptability of current video surveillance processing scenarios by collecting environmental data and confirming the appropriate interval duration of frame pulling, thereby ensuring the rationality of monitoring video frame pulling, and further ensuring the effectiveness of monitoring and the timeliness of abnormal response in frequently and rapidly changing environments. At the same time, it also ensures the subsequent processing efficiency and the subsequent identification and classification effects to a certain extent.

[0038] Furthermore, in step S22, the meteorological interference factor of each key monitoring area is set, including: J1, comparing each key monitoring area with the meteorological interference monitoring area set.

[0039] J2. If a key monitoring area is not located in the set of meteorological interference monitoring areas, 0 is used as the meteorological interference factor of the key monitoring area.

[0040] J3. If a key monitoring area is located in the set of meteorological interference monitoring areas, the key monitoring area shall be recorded as an interference area, and the target detection value of each meteorological detection indicator shall be confirmed.

[0041] Among them, confirming the target detection value of each meteorological detection index includes: J31, calculating the average detection value of each meteorological detection index corresponding to each monitoring time point, and obtaining the average detection value of each meteorological detection index, which is recorded as , Indicates the meteorological detection index number. , and extract the maximum detection value of each meteorological detection index, recorded as .

[0042] J32. With the monitoring time point as the horizontal coordinate and the detection value as the vertical coordinate, construct the detection value change curve of each meteorological detection index, and calculate the slope of the curve, which is recorded as .

[0043] J33. Calculate the target detection value of each meteorological detection index. ,

[0044] .

[0045] J4. Compare the target detection value of each meteorological detection index with the set interference value of each meteorological detection index. Mark the meteorological detection index whose target detection value is greater than the corresponding interference value as the interference index. Count the number of interfering meteorological detection indicators and divide it by the number of meteorological detection indicators. Record the ratio as .

[0046] J5. The difference between the target detection value and the interference value of each interference index is recorded as , Indicates the interference indicator number, .

[0047] J6. Set the meteorological interference factor of the interference area ,

[0048] , and They represent the weight coefficients corresponding to the ratio of interference meteorological detection indicators and the difference of interference meteorological detection indicators, For setting The reference difference of interference detection indicators, , is a natural constant.

[0049] It should be added that the proportion of interference meteorological detection indicators reflects the proportion of specific meteorological conditions in the entire meteorological environment. The difference in meteorological detection indicators mainly reflects the degree of deviation between the current meteorological conditions and normal or expected meteorological conditions. For example, the difference between the current wind speed and the average wind speed, the difference between the temperature and the standard temperature, etc. Because the present invention mainly considers the meteorological interference situation, if the value of a single meteorological detection indicator exceeds a large amount, it has caused significant interference. Therefore, the weight coefficient corresponding to the difference in interference meteorological detection indicators is set to be greater than the weight coefficient corresponding to the proportion of interference meteorological detection indicators. For example, The value can be 0.6. The value can be 0.4.

[0050] In a specific embodiment, when the interference detection indicator is wind speed, the reference difference value of the interference detection indicator can be 15 meters per second; when the interference detection indicator is wind direction angle, the reference difference value of the interference detection indicator can be 30 degrees; when the interference detection indicator is air pressure, the reference difference value of the interference detection indicator can be 20 hPa; when the interference detection indicator is humidity, the reference difference value of the interference detection indicator can be 20%.

[0051] Exemplarily, the meteorological interference monitoring area centrally records the deck area, the side area, the mast and antenna area, the cockpit area, etc.

[0052] Among them, the specific impact of meteorological sensor data on the meteorological interference monitoring area is explained as follows: 1) Deck area: When the wind speed sensor detects a high wind speed, items on the deck may be blown off, and the risk of crew members working on the deck will also increase. At this time, the pulling interval of the deck area video should be shortened to timely detect potential safety hazards, such as loose equipment, cargo blown by the wind, etc. If the wind direction sensor shows that the wind direction is not conducive to the stability of the ship, such as a strong side wind, it may affect the placement of equipment on the deck and the actions of the crew. At this time, it is also necessary to pull the video frames of the deck area more frequently so that the equipment position can be adjusted in time or safety measures can be taken. The pulling interval can be adjusted according to the degree of change in wind direction. The greater the change, the shorter the pulling interval.

[0053] 2) Side area: Strong winds may increase the force of waves impacting the side of the ship, increasing the risk of damage to the side of the ship. When the meteorological sensor detects changes in wind speed and wave conditions, the interval for pulling the video of the side area should be shortened, and close attention should be paid to the condition of the side of the ship, such as cracks, leaks, etc. The pulling interval can be adjusted according to the combined impact of wave height and wind speed. The higher the risk, the shorter the interval.

[0054] 3) Mast and antenna area: Strong winds may cause the mast and antenna to shake more severely, and may even cause structural damage. The data from the wind speed sensor can be used as an important basis for adjusting the video pull interval in this area. When the wind speed exceeds a certain threshold, the pull interval should be shortened, and the status of the mast and antenna should be closely observed to ensure that their stability and communication functions are not affected.

[0055] 4) Cockpit area: Bad weather conditions will affect the ship's control and navigation safety. Data such as wind direction and wind speed are crucial for cockpit decision-making. Weather conditions such as rain and fog will affect the cockpit's vision. At this time, the video pull interval can be adjusted based on data such as humidity and visibility detected by the meteorological sensor to ensure safe operation of the cockpit.

[0056] Further, exemplarily in step S23, the sea condition interference monitoring area set records the ship side area, bow area, stern area and deck area, etc.

[0057] Among them, the specific impact of sea condition sensor data on the sea condition interference monitoring area is explained as follows: 1. Ship side area: When the sea condition sensor detects an increase in wave height, the force of the wave impact on the ship side will increase, which may cause damage to the ship side or leakage. At this time, the interval for pulling the video in the ship side area should be shortened so that potential problems can be discovered in time and corresponding measures can be taken. At the same time, changes in water flow speed will also affect the safety of the ship side area. If the water flow speed increases, the relative position of the ship and surrounding objects may change, increasing the risk of collision. By pulling video frames in the ship side area more frequently, changes in the surrounding environment can be observed in time to adjust the position and heading of the ship.

[0058] 2. Bow area: Data such as wave direction and height detected by the sea condition sensor have an important impact on the video pulling interval in the bow area. When the wave hits the bow head-on, it may cause greater pressure on the ship structure, and it is necessary to pull video frames more frequently to observe the damage to the bow and the operating status of the equipment. The pulling interval can be adjusted according to the intensity and direction of the waves. The higher the risk, the shorter the interval. And if the water flow speed is fast and inconsistent with the ship's heading, the bow may deviate from the scheduled route. The video pulling interval in the bow area needs to be adjusted in time to ensure that the crew can detect and correct the heading deviation in time.

[0059] 3. Stern area: When the wave height is large or the water flow speed is fast, the stability of the stern may be affected, which in turn affects the normal operation of the equipment. At this time, the pulling interval of the stern area video should be shortened and the operating status of the equipment should be closely monitored.

[0060] 4. Deck area: When the sea conditions are poor, the interval for pulling the video of the deck area should be shortened so as to timely detect cargo displacement, loose equipment, etc., and take corresponding fixing and protection measures. At the same time, the activities of crew members on the deck will also be affected by the sea conditions. In severe sea conditions, the risk of crew members working on the deck increases, and the deck area needs to be monitored more frequently to ensure the safety of the crew. For example, when the wave height exceeds a certain level, the frequency of video monitoring of the crew activity area on the deck can be increased to detect dangerous situations in time and remind the crew to pay attention to safety.

[0061] It should be noted that the present invention is based on settings under normal circumstances and can be adjusted and optimized according to specific circumstances in actual applications.

[0062] Furthermore, regarding setting the light interference factor of each monitoring area in step S24, it includes: taking the detection time as the horizontal coordinate and the light brightness as the vertical coordinate, constructing the light brightness change curve of each key monitoring area, and extracting the number of fluctuation points and the interval detection time between each fluctuation point.

[0063] The number of fluctuation points in each key monitoring area is recorded as , and the interval detection time between the fluctuation points is averaged to obtain the average interval detection time of each key monitoring area, which is recorded as .

[0064] Set the light and heat interference factor of each monitoring area ,

[0065] , and They respectively represent the number of fluctuation points and interval detection duration for setting reference.

[0066] In a specific embodiment, The specific value of is mainly set according to the number of detection time points. When the number of detection time points is 20, The specific value can be 3. The value of is mainly set according to the detection time. The interval between the first detection time point and the last detection time point is counted from the light brightness change curve of each key monitoring area as the cumulative interval time, recorded as , .

[0067] Regarding the classification of the pulled video frames of each key monitoring area in step S3, it includes: using image recognition technology to identify various instrument display screens in each video frame corresponding to each key monitoring area, classifying each video frame containing the display screen image as an instrument display class, and adding corresponding key monitoring area labels.

[0068] Image recognition technology is used to identify the people in each video frame corresponding to each key monitoring area, and each video frame containing a person's image is classified as a person posture monitoring class, and the corresponding key monitoring area label is added.

[0069] Image recognition technology is used to identify the waves in each video frame corresponding to each key monitoring area, and each video frame containing wave images is classified as a sea condition monitoring category, and the corresponding key monitoring area label is added.

[0070] Image recognition technology is used to identify the hull structure in each video frame corresponding to each key monitoring area, and each video frame containing the hull structure is classified as a hull structure monitoring class, and the corresponding key monitoring area label is added.

[0071] It can be understood that image recognition technology is a relatively mature technology currently available, and its specific recognition process will not be described in detail here. Adding corresponding monitoring key area labels refers to adding corresponding monitoring key area labels to each classified video frame, that is, each video frame has a monitoring key area label added to facilitate subsequent identification and processing.

[0072] The algorithm models pre-set in step S4 are, in order, a personnel detection model, an equipment detection model, a sea condition analysis model, and a potential safety hazard detection model, and each algorithm model is an independent thread.

[0073] Furthermore, each transmitted video frame is identified through each pre-set algorithm model, including: each video frame of the personnel posture monitoring class is identified through a pre-set personnel detection model to obtain each personnel feature and use it as the output feature of the personnel detection model.

[0074] Each video frame of the instrument display class is identified through a pre-set device detection model to obtain the characteristics of each device and use it as the output feature of the device detection model.

[0075] Through the pre-set sea condition analysis model, each video frame of the sea condition monitoring class is identified to obtain the characteristics of each sea condition and use them as the output characteristics of the sea condition analysis model.

[0076] Each video frame of the hull structure monitoring class is identified through a pre-set safety hazard detection model to obtain various safety features, which are used as output features of the safety hazard detection model.

[0077] The output features of the personnel detection model, equipment detection model, sea condition analysis model and safety hazard detection model are taken as recognition results.

[0078] In a specific embodiment, the personnel detection model can specifically adopt a target detection algorithm based on deep learning, such as the YOLO series algorithm or FasterR-CNN, etc. Personnel features include wearing features and the number of personnel, facial posture features of personnel and motion posture features, wherein the wearing features refer to whether safety equipment is worn, such as a helmet and a life jacket, and the motion posture features are specifically such as the body inclination angle, and the facial posture features of personnel are such as the overall facial inclination angle and the degree of eye opening, etc., wherein the motion posture features can be used to determine whether a person has fallen, and the facial posture features of personnel can be used to determine the fatigue of personnel, etc.

[0079] The sea condition analysis model can be specifically integrated with multiple image analysis algorithms such as edge detection algorithm, texture recognition algorithm and target tracking algorithm, etc., which is used to analyze the sea condition information around the ship. The sea condition characteristics include wave height, water flow speed, wind direction, etc. The equipment detection model and the safety hazard detection model have the same composition as the sea condition analysis model, among which the characteristics of each equipment are the images of each angle corresponding to each equipment, etc., and the safety characteristics include cracks, rust and tilt, etc.

[0080] It should also be added that the hull structure includes the side hull, mast, railings and antennas, etc.

[0081] The embodiment of the present invention performs video frame recognition by setting up algorithm models of multiple independent threads. First, the video frames can be analyzed simultaneously, which greatly shortens the overall processing time. Secondly, different key areas and different video frames can be detected at the same time to quickly obtain analysis results. At the same time, the stability of video frame recognition is increased, avoiding mutual interference and mutual influence in the processing of different video frames.

[0082] The push setting analysis described in step S5 includes: extracting the output features of the personnel detection model, the equipment detection model, the sea condition analysis model and the safety hazard detection model from the recognition results, setting the output video frame compression requirement factor corresponding to each algorithm model, recorded as , Indicates the algorithm model number, .

[0083] The preset video push compression ratio and the upper limit of the suitable compression ratio range are respectively recorded as and , calculate the appropriate compression ratio of the output video frame corresponding to each algorithm model, denoted as , , and serves as the push strategy for the corresponding output results of each algorithm model.

[0084] Furthermore, the setting of the compression requirement factor includes: extracting the features of each person corresponding to each video frame and the added monitoring key area label from the output features of the person detection model.

[0085] The characteristics of each person in the same monitoring key area label are compared with the corresponding preset reference personnel characteristics one by one, and the number of matching personnel characteristics is counted and divided by the number of personnel characteristics. The ratio is recorded as the characteristic matching ratio, and the characteristic matching ratio of each monitoring key area label is obtained. The average characteristic matching ratio is obtained by mean calculation and recorded as .

[0086] The key monitoring areas whose feature matching ratio is less than the set reference matching ratio are recorded as warning areas, and the number of warning areas is counted.

[0087] The added monitoring key area labels corresponding to each video frame are compared, the same monitoring key area labels are filtered, and the number of remaining monitoring key area labels after filtering is counted as the number of involved monitoring key areas.

[0088] The ratio of the number of warning areas to the number of monitoring areas is recorded as the warning monitoring ratio, ,Will

[0089] As the output video frame compression requirement factor corresponding to the person detection model, and They are the weight coefficients corresponding to the set early warning monitoring ratio and feature matching ratio respectively.

[0090] The output video frame compression demand factors corresponding to the equipment detection model, sea condition analysis model and safety hazard detection model are set in the same way as the setting method of the output video frame compression demand factors corresponding to the personnel detection model, and then the output video frame compression demand factors corresponding to each algorithm model are integrated to obtain.

[0091] In a specific embodiment, The value can be 0.6. The value can be 0.4.

[0092] It should be added that compressing video frames can reduce the amount of data, reduce transmission costs and time. Compression is particularly important when pushing data through channels with limited bandwidth. When the analysis results processed by the model are pushed to relevant personnel, efficient data transmission is also required to ensure the timeliness and reliability of the push.

[0093] The embodiment of the present invention improves the capture efficiency and processing efficiency of key events as much as possible on the basis of ensuring that video information can be obtained in time during critical periods by confirming the appropriate interval length of frame pulling, setting the algorithm model of multiple independent threads and setting the compression ratio. It effectively solves the problem that the current processing efficiency is difficult to adapt to marine scenes, and can ensure that marine warnings can respond quickly in emergencies, thereby ensuring the timeliness of warning responses. At the same time, a reasonable compression ratio can reduce the pressure of data transmission and storage while ensuring video quality, thereby ensuring the processing speed of video processing and push.

[0094] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they shall all fall within the protection scope of the present invention.

Claims

1. A surveillance video classification processing and pushing method based on data analysis, characterized in that: include: S1. Monitoring environment data collection: Real-time environmental data collection through pre-installed environmental sensor equipment; S2, monitoring video data acquisition: determine the appropriate interval length for pulling video frames corresponding to each key monitoring area, and pull video frames from the corresponding cameras installed in each pre-set key monitoring area according to the appropriate interval length; S3, surveillance video classification and packaging: classify and package the video frames of each key surveillance area, obtain video frames of each type, and transmit them to the corresponding pre-set algorithm models; S4, video recognition and feature construction: recognize each transmitted video frame through each pre-set algorithm model to obtain the recognition result; S5, video push setting processing: perform push setting analysis and output push strategies corresponding to the output results of each algorithm model; S6, monitoring video push: based on the pre-set user permissions, push strategies of the corresponding output results of each algorithm model and video features, corresponding push is performed; The step of determining the appropriate interval duration for pulling video frames corresponding to each key monitoring area includes: Extract meteorological sensor data from real-time collected environmental data, and then extract the detection value of each meteorological detection index corresponding to each monitoring time point; Extract the meteorological interference monitoring area set from the database and set the meteorological interference factor of each key monitoring area, denoted as , Indicates the monitoring key area number. ; The sea condition sensor data is extracted from the real-time collected environmental data, and then the detection values ​​of each sea condition detection index corresponding to each monitoring time point are extracted. At the same time, the sea condition interference monitoring area set is extracted from the database. The sea condition interference factor of each key monitoring area is set in the same way as the meteorological interference factor of each key monitoring area, which is recorded as ; The real-time detected light brightness corresponding to each key monitoring area is extracted from the real-time collected environmental data, and the light interference factor of each monitoring area is set, which is recorded as ; The pre-set video frame extraction interval duration corresponding to each key monitoring area is recorded as , calculate the appropriate interval length for pulling video frames corresponding to each key monitoring area , , , and are the weight coefficients of meteorological interference factor, sea state interference factor and light and heat interference factor, ; The meteorological interference factors of each key monitoring area are set, including: Compare each key monitoring area with the set of meteorological interference monitoring areas; If a key monitoring area is not located in the set of meteorological interference monitoring areas, 0 is used as the meteorological interference factor of the key monitoring area; If a key monitoring area is located in the set of meteorological interference monitoring areas, the key monitoring area is recorded as an interference area, and the target detection value of each meteorological detection index is confirmed; The target detection value of each meteorological detection index is compared with the set interference value of each meteorological detection index. The meteorological detection index with a target detection value greater than the corresponding interference value is marked as an interference index. The number of interfering meteorological detection indicators is counted and divided by the number of meteorological detection indicators. The ratio is recorded as ; The difference between the target detection value and the interference value of each interference index is recorded as , Indicates the interference indicator number, ; Set the weather interference factor for the interference area , , and They represent the weight coefficients corresponding to the ratio of interference meteorological detection indicators and the difference of interference meteorological detection indicators, For setting The reference difference of the interference detection index, , is a natural constant; Determining the target detection value of each meteorological detection index includes: The detection values ​​of each meteorological detection index corresponding to each monitoring time point are averaged to obtain the average detection value of each meteorological detection index, which is recorded as , Indicates the meteorological detection index number. , and extract the maximum detection value of each meteorological detection index, recorded as ; With the monitoring time point as the horizontal coordinate and the detection value as the vertical coordinate, a detection value change curve of each meteorological detection index is constructed, and the slope of the curve is calculated, which is recorded as ; Calculate the target detection value of each meteorological detection index , ; The light interference factor of each monitoring area is set, including: With the detection time as the horizontal axis and the light brightness as the vertical axis, a light brightness change curve of each key monitoring area is constructed, from which the number of fluctuation points and the interval detection time between each fluctuation point are extracted; The number of fluctuation points in each key monitoring area is recorded as , and the interval detection time between the fluctuation points is averaged to obtain the average interval detection time of each key monitoring area, which is recorded as ; Set the light and heat interference factor of each monitoring area , , and They respectively represent the number of fluctuation points and interval detection duration for setting reference.

2. A surveillance video classification processing and pushing method based on data analysis as claimed in claim 1, characterized in that: The classifying of the pulled video frames of each key monitoring area includes: Use image recognition technology to identify various instrument display screens in each video frame corresponding to each key monitoring area, classify each video frame containing the display screen as an instrument display class, and add corresponding key monitoring area labels; Use image recognition technology to identify the people in each video frame corresponding to each key monitoring area, classify each video frame containing the person's image into the person posture monitoring category, and add the corresponding key monitoring area label; Use image recognition technology to identify the waves in each video frame corresponding to each key monitoring area, classify each video frame containing the wave image as a sea condition monitoring category, and add the corresponding key monitoring area label; Image recognition technology is used to identify the hull structure in each video frame corresponding to each key monitoring area, and each video frame containing the hull structure is classified as a hull structure monitoring class, and the corresponding key monitoring area label is added.

3. A surveillance video classification processing and pushing method based on data analysis as claimed in claim 1, characterized in that: The pre-set algorithm models are respectively a personnel detection model, an equipment detection model, a sea condition analysis model and a safety hazard detection model, and each algorithm model is an independent thread.

4. A surveillance video classification processing and pushing method based on data analysis as claimed in claim 3, characterized in that: The identifying each transmitted video frame by using each preset algorithm model includes: Through the pre-set personnel detection model, each video frame of the personnel posture monitoring class is identified to obtain the characteristics of each person and use them as the output features of the personnel detection model; Using a pre-set device detection model, each video frame of the instrument display class is identified to obtain each device feature, which is used as the output feature of the device detection model; By identifying each video frame of the sea condition monitoring class of the pre-set sea condition analysis model, each sea condition feature is obtained and used as the output feature of the sea condition analysis model; Through the pre-set potential safety hazard detection model, each video frame of the hull structure monitoring class is identified to obtain various safety features, which are used as output features of the potential safety hazard detection model; The output features of the personnel detection model, equipment detection model, sea condition analysis model and safety hazard detection model are taken as recognition results.

5. A surveillance video classification processing and pushing method based on data analysis as claimed in claim 4, characterized in that: The push setting analysis includes: The output features of the personnel detection model, the equipment detection model, the sea condition analysis model and the potential safety hazard detection model are extracted from the recognition results, and the output video frame compression requirement factor corresponding to each algorithm model is set, which is recorded as , Indicates the algorithm model number, ; The preset video push compression ratio and the upper limit of the suitable compression ratio range are respectively recorded as and , calculate the appropriate compression ratio of the output video frame corresponding to each algorithm model, denoted as , , and serves as the push strategy for the corresponding output results of each algorithm model.

6. A surveillance video classification processing and pushing method based on data analysis as claimed in claim 5, characterized in that: The step of setting the compression requirement factor includes: Extract the features of each person in each video frame and the added key monitoring area labels from the output features of the person detection model; The characteristics of each person in the same monitoring key area label are compared with the corresponding preset reference personnel characteristics one by one, and the number of matching personnel characteristics is counted and divided by the number of personnel characteristics. The ratio is recorded as the characteristic matching ratio, and the characteristic matching ratio of each monitoring key area label is obtained. The average characteristic matching ratio is obtained by mean calculation and recorded as ; The key monitoring areas whose feature matching ratio is less than the set reference matching ratio are recorded as warning areas, and the number of warning areas is counted; Compare the added monitoring key area labels corresponding to each video frame, filter the same monitoring key area labels, and count the number of monitoring key area labels remaining after filtering as the number of monitoring key areas involved; The ratio of the number of warning areas to the number of monitoring areas is recorded as the warning monitoring ratio, ,Will As the output video frame compression requirement factor corresponding to the person detection model, and They are the weight coefficients corresponding to the set early warning monitoring ratio and feature matching ratio; The output video frame compression demand factors corresponding to the equipment detection model, sea condition analysis model and safety hazard detection model are set in the same way as the setting method of the output video frame compression demand factors corresponding to the personnel detection model, and then the output video frame compression demand factors corresponding to each algorithm model are integrated to obtain.

Images