Artificial intelligence-based multi-path streaming data analysis method

Abstract

The application relates to the technical field of artificial intelligence, and particularly discloses a multi-path streaming media data analysis method based on artificial intelligence. The method is aimed at source confusion during smoke propagation across devices and connected spaces, identifies smoke changes in multi-path videos, includes connected areas lacking continuous records in a common scope to be investigated, uses historical videos and propagation conditions in the weakening and subsiding stages of smoke to verify whether external transmission and accumulated release can jointly explain each video, confirms the source only when the above explanation is not established and internal production can explain, further verifies a sub-range, and retains a range that cannot be distinguished and missing records. The application can reduce smoke source misjudgment caused by changes in observation ranges in video linkage and improve the reliability of abnormal source attribution.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to a multi-channel streaming media data analysis method based on artificial intelligence. Background Technology

[0002] In the construction of low-voltage intelligent systems in buildings and industrial parks, video surveillance undertakes tasks such as on-site status acquisition, anomaly identification, and remote verification. By accessing multiple streaming video feeds distributed in equipment rooms, corridors, and adjacent areas, it is possible to centrally view on-site images from different locations and use image recognition to extract suspected smoke. Upon detecting an anomaly, coordinated verification can be carried out by rotating and zooming the cameras and retrieving video footage, observing relevant equipment and areas, and providing image data for operation management and safety inspections.

[0003] In well-ventilated environments, the location where smoke appears may differ from the actual smoke-generating location. Joint verification often revolves around the initial suspected location, and the camera's observation range and video transmission content change accordingly. The continuous process of smoke entering, passing through, and dissipating between different areas may not be fully recorded. Subsequent clear and consistent local images may still only reflect the phenomenon of external smoke passing through relevant equipment, making it difficult to determine the source of the smoke. Initial attribution errors may also be reinforced during continuous verification, affecting anomaly localization and on-site investigation. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-channel streaming media data analysis method based on artificial intelligence to solve the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: The AI-based multi-stream streaming data analysis method includes the following steps: S1: Acquire multiple video feeds, ventilation data, flue gas calibration records, and flue gas recognition models. Input the multiple video feeds into the flue gas recognition model to extract flue gas changes and video clips. S2: Establish the transmission relationship based on ventilation data, and combine equipment and connected spaces where flue gas changes are not continuously recorded to form the scope to be investigated; S3: Based on the flue gas calibration records and propagation relationships, set the external input volume, accumulated volume, internal new volume, and ventilation opening delivery volume as quantities to be determined and associate them with video clips; S4: Tracing back the undemanded quantity along the propagation relationship, grouping the associated video clips into a clip group, retrieving the corresponding video and supplementing the shooting, and updating the clip group; S5: Combine the calculations of the same unknown quantity in each segment group, respectively prohibiting and allowing the addition of new flue gas within the range to be investigated, and obtain the solution results; S6: Based on the solution results and the segment group, the range to be investigated is divided into segments. The range where internal smoke production is determined to exist is recorded as the source range, and the source range or the range to be investigated is output.

[0006] As a further aspect of the present invention: the process of extracting flue gas changes and video clips is as follows: Set black and white adjacent background markers in the background of the ventilation opening, capture smoke-free images, record the shooting position and exposure parameters of the smoke-free images and multiple video streams, and read the acquisition time of the multiple video streams. Acquire videos of labeled flue gas areas and train a flue gas recognition model. Input multiple videos into the flue gas recognition model to extract the flue gas areas. Based on the correspondence between pixel values, exposure parameters and brightness in the camera calibration data, as well as the calibration error, the pixel values ​​of the background markers in the smokeless image and multiple video streams are converted into a range of brightness values. Select a smoke-free image based on the shooting location. Within the brightness range, divide the black-and-white brightness difference of the corresponding background marker outside the smoke area in the smoke-free image by the black-and-white brightness difference in the multiple video streams to obtain the ratio. Multiply the black-and-white brightness difference of the background marker in the multiple video streams by the ratio, and then divide it by the corresponding black-and-white brightness difference in the smoke-free image to obtain the transmittance range. Arrange the flue gas areas and transmittance ranges according to the time of collection to form flue gas changes, extract video clips of the corresponding time periods and associate them with ventilation openings and shooting positions; The periods when the background is obscured by non-smoke objects, the video is interrupted, or the transmittance range cannot be obtained are marked as observation missing, and the corresponding ventilation openings and start and end times are recorded.

[0007] As a further aspect of the present invention: the scope to be investigated includes: Read the equipment layout, ventilation opening location and site boundary in the ventilation data, record the external space that is connected to the equipment through the ventilation opening as the connected space, associate the ventilation opening with the equipment interior or connected space on both sides, and establish the propagation relationship; Based on the changes in flue gas, the internal space of the equipment or connected space corresponding to the ventilation opening where flue gas appears is included in the scope of investigation, and the ventilation opening connecting the scope of investigation to the outside is found along the propagation relationship. Select video clips from both sides according to the ventilation opening and shooting position, and record the time period that is not covered by video clips and is missing from observation from the first identification of smoke to the current collection time as the recording gap; For ventilation openings with recording gaps on either side of the site boundary, incorporate the connected equipment interiors or adjacent spaces that are not included in the scope of investigation into the scope of investigation. Repeat the search, inspection, and merging until there are no recording gaps on either side of the found ventilation opening or it is located at the site boundary. Mark propagation relationships with gaps in the records at the merging points and site boundaries as unverified relationships, and associate the start and end times of the gaps in the records; mark propagation relationships in ventilation data where the connection points or airflow directions are not identified as unverified relationships.

[0008] As a further aspect of the present invention: setting the quantity to be determined and associating it with video segments includes: The time period from the first identification of flue gas to the current collection time is recorded as the judgment time period. According to the equipment interior, connected space and collection time period, the external input amount, the accumulated amount, the internal new amount and the ventilation opening delivery amount are set as non-negative undetermined quantities. The discharge and inflow volumes on both sides of the same ventilation opening during the same sampling period use the same required quantity, and the accumulated volume at the end of the previous sampling period and the beginning of the next sampling period use the same required quantity. Based on the space volume, airflow direction, air volume, temperature and humidity in the ventilation data, and combined with the flue gas conservation relationship, establish the flue gas transport relationship; Acquire measurement data of smoke inlet, smoke outlet, and accumulated release during commissioning, along with corresponding transmittance ranges. Combine this with the flue gas transport relationship to calibrate the ranges of transport parameters and flue gas optical parameters, thus forming a flue gas calibration record. Associate the quantities to be determined with video clips based on ventilation openings and collection time periods; when observations are missing, retain the corresponding quantities to be determined and flue gas transport relationships. During the judgment period, no new flue gas is allowed to be added within the area to be investigated. The amount of external input and the initial accumulation are retained. Based on the flue gas transport relationship, the range of transport parameters and flue gas optical parameters in the flue gas calibration record and the shooting position, the transmittance range corresponding to the value of the quantity to be determined is calculated. The calculated transmittance range and the transmittance range of the corresponding video segment are used together as constraints to solve the flue gas transport relationship. The values ​​of the unknown quantity that satisfy the constraints are recorded as the feasible value group.

[0009] As a further aspect of the present invention: composing and updating the fragment group includes: Read historical videos before the judgment period, extract the transmittance value range and set the quantity to be determined according to the collection period. Allow the addition of flue gas inside within the period covered by the historical video. Solve the problem by combining the flue gas transport relationship with the transmittance value range. Save all accumulated quantity values ​​that satisfy the constraints and the constraints between each accumulated quantity. Based on the changes in flue gas, select the exhaust video segments that record the weakening or disappearance of flue gas, trace back along the flue gas transport relationship to find the quantities to be calculated in the exhaust calculation, and form a segment group by combining the associated inlet video segments, the previous exhaust video segments, and the selected exhaust video segments. When tracing back to the accumulated quantity, continue reading the quantity to be determined from the previous acquisition period. When tracing back to the saved accumulated quantity constraints, inherit the corresponding accumulated quantity values ​​and the constraints between each accumulated quantity. When tracing back to the starting point of video recording, retain the starting accumulated quantity as the quantity to be determined and associate the segments that use the same quantity to be determined. Retrieve local video from cameras according to the ventilation openings and acquisition time periods within the segment group, connect video segments according to acquisition time, and record the cameras and acquisition time periods corresponding to video segments that have not yet been retrieved. Read the camera recording time period, check the untransmitted video and the missing observations, retain the quantities to be determined corresponding to the missing observations, supplement the shooting according to the ventilation openings corresponding to the missing observations, and extend the segment group according to the newly added acquisition time.

[0010] As a further aspect of the present invention, the solution results include: The range of transmittance values ​​for video segments is defined as the constraint on the quantity to be determined, and the range of values ​​for the quantity to be determined before each segment group is solved individually is recorded. By combining the observation constraints and flue gas transport relationships of each segment group, the feasible value set can be solved; During the judgment period, the addition of new flue gas within the scope to be investigated is prohibited and allowed respectively, and all segment groups are solved together; when the addition of new flue gas within the scope is prohibited, there is a feasible solution, the feasible value group is updated and the scope to be investigated is determined to be undetermined; when the constraint is not feasible when the addition of new flue gas within the scope is prohibited, but when the addition of new flue gas within the scope is allowed, there is a feasible solution, and the scope to be investigated is determined to have internal smoke generation. The observation constraints corresponding to each segment group are temporarily removed, while the flue gas transport relationship and the unsolved quantities used by each segment group are retained. The solution is then re-solved from the recorded range of unsolved quantities. If the constraints are still infeasible when adding new flue gas is prohibited, they are removed. If a feasible solution is obtained again when adding new flue gas is prohibited, or if the determination is not completed, the observation constraints are restored. Read the air volume and temperature and humidity during the debugging period, record the corresponding value range as the calibration range, and obtain the current air volume and temperature and humidity; when it is not feasible to constrain the addition of internal smoke, the current air volume and temperature and humidity exceed the calibration range, or the judgment cannot be completed, the range to be investigated is judged as undetermined. The determination that the area to be investigated contains internal smoke production or is undetermined is taken as the solution result, and the last group of segments with observation constraints is associated with the solution result.

[0011] As a further aspect of this invention: the scope to be investigated is divided according to the solution results and the segment group, including: Based on the solution results, the area to be investigated that is determined to have internal smoke production is recorded as the source area, and associated with the judgment time period; The source range is divided along the internal ventilation openings continuously recorded by the segment group within the judgment period to obtain sub-ranges, and the flue gas transport relationship of the original range to be investigated and the quantities to be determined that are used by each segment group are retained. Within the judgment period, sub-ranges are selected sequentially, allowing new smoke to be added within the remaining sub-ranges, and prohibiting and allowing new smoke to be added within the selected sub-ranges respectively, and then solving all segment groups together; The sub-range where adding new flue gas into the prohibited sub-range is not feasible under constraints, but where adding new flue gas into the allowed sub-range has a feasible solution, is denoted as the source range. Sub-ranges that meet the conditions are retained, and the original source range is retained if no sub-ranges meet the conditions. Save the segment group corresponding to the solution results. When the camera turns, zooms, or the video exits the analysis, the saved video segments are retained. For subsequent periods without shooting, the corresponding undetermined quantities are retained. The propagation relationship of the corresponding ventilation opening is marked as an unverified relationship. Output the source range and judgment period. If it is determined that the range to be investigated is not output on time, it will also output the camera and acquisition period corresponding to the unverified relationship and the video clip that has not yet been retrieved.

[0012] The beneficial effects of this invention compared to the prior art are as follows: This invention improves the reliability of attributing abnormal smoke in low-voltage intelligent video surveillance, reducing the possibility of smoke passing through equipment or being released from accumulated smoke being misjudged as smoke originating inside the equipment. Even after camera rotation, zooming, or changes in video transmission content, existing attribution criteria can still be integrated with subsequent observations, reducing the loss of judgment criteria due to changes in the observation range and the repeated reinforcement of initial attribution biases. The narrowing of the source range is consistent with changes in smoke within the relevant space, reducing the premature identification of equipment based solely on clear local images or consistent multiple video feeds. For sources that cannot be immediately distinguished, the investigation range consistent with existing records is retained, clarifying unverified propagation relationships and the location of videos to be retrieved, improving the targeting of cross-camera verification and supplementary observations, and providing a range basis for remote monitoring, anomaly analysis, and on-site troubleshooting in low-voltage intelligent scenarios. Attached Figure Description

[0013] The invention will now be further described with reference to the accompanying drawings.

[0014] Figure 1 This is a flowchart illustrating the multi-channel streaming media data analysis method based on artificial intelligence according to the present invention. Figure 2 This is a schematic diagram of the process by which the solution is obtained in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figures 1-2 As shown, this invention is a multi-channel streaming media data analysis method based on artificial intelligence, comprising the following steps: S1: Acquire multiple video feeds, ventilation data, flue gas calibration records, and flue gas recognition models. Input the multiple video feeds into the flue gas recognition model to extract flue gas changes and video clips. In a preferred embodiment of the present invention, the process of extracting changes in flue gas and video clips is as follows: Multiple video feeds are input from the low-voltage intelligent video surveillance equipment, image frames and acquisition times are retained, and the clocks of all cameras are adjusted to the same time base. Ventilation data is retrieved from on-site equipment documentation, and flue gas calibration records are retrieved from the commissioning logs. The flue gas calibration records contain the range of transport parameters describing the changes in flue gas inflow, outflow, and accumulation, as well as the range of optical parameters describing the relationship between flue gas and light attenuation, for subsequent solution purposes.

[0017] Matte black and white markers are placed on a fixed background behind and around the ventilation openings, with the black and white portions on the same plane. Each marker is illuminated by the same light source, and the light path from the light source to the marker avoids the area through which smoke passes. The camera observes that the light path of the marker behind the opening passes through the ventilation opening. In a smoke-free state, smoke-free images are captured according to the actual shooting position and focal length. The camera position, orientation, focal length, exposure time, aperture, and gain are recorded, and the correspondence between each marker and the ventilation opening is saved. During video processing, smoke-free images are matched according to the shooting position and focal length. The sampling position uses pixels within the corresponding black and white portion, excluding pixels at the black-and-white boundary and on the outer contour.

[0018] The training videos include images from different shooting positions, lighting conditions, and smoke concentrations, as well as images with no smoke, steam, reflections, and object obstruction. Smoke areas are labeled frame-by-frame, with smoke pixels marked as 1 and other pixels as 0. The training and validation sets are divided into 8:2 ratios based on the complete video recordings, with the same continuous video not being split into two groups simultaneously. Images are scaled while maintaining aspect ratio and padded to 512x512 pixels, with synchronous transformation applied. The pixel values ​​of the three channels of the color image are divided by 255.

[0019] The smoke recognition model employs a convolutional structure connecting an encoder and a decoder. The encoder has four stages, each performing two 3x3 convolutions followed by linear rectification. Each 3x3 convolution has a stride of 1, with edge padding of 1 pixel. The output channel counts for each stage are 32, 64, 128, and 256, respectively. At the end of each stage, a 2x2 max pooling operation halves the image's dimensions. This is followed by two more 3x3 convolutions and linear rectifications, resulting in a 512-channel output. The decoder also has four stages. Each stage first uses bilinear interpolation to double the image's dimensions. The pre-pooling results of the same size as the encoder are then concatenated by channel, followed by two more 3x3 convolutions and linear rectifications. The output channel counts are 256, 128, 64, and 32, respectively. Finally, a 1x1 convolution and logistic function are used to output a single-channel smoke probability of the same size as the input image.

[0020] Convolutional weights are initialized using Kamin, with a bias of 0. Pixel-level binary cross-entropy is used to calculate the difference between the predicted probability and the labeled probability. The convolutional weights are updated through backpropagation and stochastic gradient descent with momentum. In this embodiment, 8 images are input per batch, with a learning rate of 0.01 and momentum of 0.9, and training is performed for 100 epochs. In each epoch, the validation set loss is calculated, and the model with the lowest loss is saved. When analyzing multi-channel video, the model is input frame by frame. The output is padded and restored to its original size according to the scaling ratio. Pixels with a probability of not less than 0.5 constitute the smoke region. Brightness calculation uses the original image without model scaling and normalization.

[0021] During camera calibration, a luminance meter is used to measure the brightness of surfaces at multiple levels, and images are taken under different exposure parameters. The pixel values ​​of the brightness channel in the decoded image are recorded along with the measured brightness. The corresponding relationships are saved according to the exposure parameters, and linear interpolation is used between adjacent calibrated pixel values. Repeated shooting is performed, and the absolute difference between the converted brightness and the measured brightness is recorded according to the exposure parameters and brightness segments. The maximum difference for each segment is taken and added to the luminance meter's measurement error as the calibration error. During actual calculation, the corresponding relationships are read, and the converted brightness is subtracted from and added to the calibration error, which are used as the lower and upper limits of brightness, respectively. The same image brightness processing settings are used for both shooting and calibration.

[0022] The lower and upper limits of brightness for each fixed sampled pixel of the background marker are averaged to obtain the brightness range of the black and white parts. The lower limit of white brightness minus the upper limit of black brightness gives the lower limit of the black-white brightness difference; the upper limit of white brightness minus the lower limit of black brightness gives the upper limit of the black-white brightness difference. The intersection of the brightness difference range and the non-negative range is taken; if there is no intersection, it is recorded as a missing observation. The smoke distribution corresponding to adjacent black and white parts is similar, and the common additional scattered brightness of the two parts cancels out during the difference calculation.

[0023] For each frame, a background marker located entirely outside the smoke area and not obscured is selected. The smoke-free black-and-white brightness difference is divided by the current black-and-white brightness difference to obtain a correction ratio for the illumination change. The lower limit of the ratio is obtained by dividing the lower limit of the smoke-free brightness difference by the upper limit of the current brightness difference, and the upper limit is obtained by dividing the upper limit of the smoke-free brightness difference by the lower limit of the current brightness difference. This ratio is applied sequentially to each background marker in the current image, including those located within the smoke area.

[0024] Multiply the current black-and-white brightness difference of each measured mark by the ratio, and then divide by the corresponding smokeless black-and-white brightness difference for that mark. When all brightness differences are non-negative, multiply the lower limit of the current brightness difference by the lower limit of the ratio, and then divide by the corresponding upper limit of the smokeless brightness difference to obtain the lower limit of transmittance; multiply the upper limit of the current brightness difference by the upper limit of the ratio, and then divide by the corresponding lower limit of the smokeless brightness difference to obtain the upper limit of transmittance. When the measured mark is used to calculate the ratio, the same brightness that appears repeatedly is kept with the same value, corresponding to a transmittance of 1. Take the intersection of the calculation interval with 0 to 1 to obtain the transmittance value interval.

[0025] The system saves the smoke area and the transmittance range corresponding to each marker at the time of acquisition, forming a record of smoke changes. From the initial acquisition time of smoke identification, images are continuously captured up to the current acquisition time, retaining images during periods of smoke weakening, intermittent activity, and dissipation. When the shooting position changes, the video is segmented at the time of change, and the video clips are associated with the ventilation openings and the shooting position.

[0026] When comparing the sampling area with the smoke-free image, check the outer contour and black-and-white boundary of the marker. If the sampling area is covered by the outline of a newly added object that is not identified as smoke, record the occlusion of the non-smoke object. Mark the observation as missing if there is video frame interruption, pixel saturation, lack of calibration data for corresponding exposure parameters or pixel values, lack of matching smoke-free images or markers outside the smoke area, divisor range containing 0, or transmittance range not intersecting with 0 to 1. Do not enter transmittance values. Record the corresponding ventilation opening and the start and end times of the missing period. If the missing period is continuous, extend the missing period, but do not record the missing period as smoke-free.

[0027] S2: Establish the transmission relationship based on ventilation data, and combine equipment and connected spaces where flue gas changes are not continuously recorded to form the scope to be investigated; In another preferred embodiment of the present invention, the scope to be investigated includes: Read the equipment layout, ventilation opening locations, and site boundaries. The site boundaries are the outer boundaries of the pre-defined analysis area in the ventilation data, which do not change with camera field of view, and are recorded separately from the boundaries of the area to be investigated, which change with merging. Based on the equipment casing, walls, and ventilation opening locations, determine the spaces on both sides connected to each ventilation opening; these external spaces connected to the equipment interior through ventilation openings are recorded as connected spaces. Save the correspondence between each ventilation opening and the interior of the equipment on both sides or the connected spaces, forming a propagation relationship.

[0028] Locate the ventilation openings where flue gas is detected by observing changes in the flue gas flow, and include the corresponding equipment interiors or connected spaces within the scope of investigation. The scope of investigation consists of the equipment interiors and connected spaces currently included in the analysis. Inspect each ventilation opening along the propagation path, including those openings with one side inside the scope of investigation and the other side outside the scope of investigation in this inspection.

[0029] Based on the locations of ventilation openings, background markers, and shooting positions, video clips recording changes in smoke on both sides were compiled. The start and end times of the data collection were limited to the period from the first identification of smoke to the current data collection time. Overlapping or consecutive coverage periods on the same side were merged, and then the missing observation periods on that side were subtracted. The remaining coverage periods were subtracted from the entire inspection period to obtain the recording gaps on that side, and the start and end times of consecutive missing periods were saved separately.

[0030] For ventilation openings located inside the site boundary, if there is a recording gap on either side, the equipment interior or connected space on the other side that has not yet been included in the scope of the investigation shall be included in the scope of the investigation.

[0031] After each merge, the ventilation openings between the area to be inspected and the outside are re-identified. Connections already within the area to be inspected will not trigger repeated merges, and ventilation openings leading to the outside of the newly merged space will continue to be inspected at the same time intervals. The merging process ends when each remaining ventilation opening meets the requirement of having no recorded gaps on both sides or being located at the site boundary. Spaces outside the site boundary will not be merged further.

[0032] Propagation relationships with gaps in recording at the merging point and site boundary are marked as unverified relationships, and the start and end times of the recording gaps are retained. For propagation relationships where the connection location or airflow direction is not identified, the known connection content is retained and the unidentified items are recorded, and these are marked as unverified relationships; only the start and end times of actual recording gaps are recorded. The merged scope to be investigated and the unverified relationships are listed below.

[0033] S3: Based on the flue gas calibration records and propagation relationships, set the external input volume, accumulated volume, internal new volume, and ventilation opening delivery volume as quantities to be determined and associate them with video clips; In a preferred embodiment of the present invention, setting the quantity to be determined and associating it with a video segment includes: The period from the first identification of flue gas to the current collection time is used as the determination period, and the collection period is divided according to the video collection time and the time of air volume change. 'i' represents the number of the space inside the equipment or connected space, 'j' represents the number of the ventilation opening, and 'k' represents the collection period number. 'M' represents the accumulated volume; the subscripts 'i' and 'k' represent the mass of flue gas particles in space 'i' at the beginning of period 'k', and the subscript 'k' plus 1 represents the end of the period; 'F' represents the mass of flue gas particles transported by the opening during the corresponding period, and 'G' represents the mass of newly generated flue gas particles in the space during the corresponding period. All the above mass units are kilograms and are set as non-negative quantities to be determined.

[0034] The flue gas transport relationship is specifically constituted by the mass conservation equations for each space and the transport equations for each opening, which respectively define how the flue gas volume increases or decreases and how it moves with the airflow. A represents the set of openings entering the corresponding space, and B represents the set of openings leaving the corresponding space. The openings are listed according to their actual connectivity, and the mass conservation equations are established: ; The same F is referenced on both sides of the same opening, and the same M is referenced at the boundary of adjacent time periods. E represents the external inflow amount to the corresponding space during the corresponding time period, in kilograms and non-negative. D represents the set of openings entering the space from outside the area to be investigated. E is related to the boundary transport amount by the following formula. The conservation equation already includes these inflow amounts, so E will not be added again: ; A space with a uniform mixing approximation was established using multi-point concentration measurement after debugging. The average concentration of the space was obtained by dividing the accumulated volume by the volume. Q represents the opening volumetric flow rate under the on-site temperature and humidity, in cubic meters per second; H represents the length of the collection period, in seconds; V represents the space volume, in cubic meters; R represents the ratio of the opening concentration to the average concentration of the upstream space, which is a transport parameter. The subscript k is associated with the collection period, and R is shared for the same opening under the same operating conditions. Implicit Euler discretization was used to approximate the average concentration of the period with the concentration at the end of the period, and the opening transport equation was established, where i is the upstream space of opening j: ; When upstream space outside the boundary lacks computational data, its input remains unknown. The same time discretization method is used for both commissioning and actual calculations. During commissioning, the processes of smoke intake, cessation of smoke intake, and continued exhaust of accumulated smoke are recorded. Concentration meters calibrated using filter quality sampling are used to independently measure the concentration at the opening and in the space. The delivery and accumulation volumes are obtained by combining airflow, duration, and volume, while simultaneously reading video transmittance. Substituting the actual measurements into the transport equation and the following optical relationships, and considering the difference between the calculated and measured values ​​according to measurement error limits, the parameter range and common limitations are determined. These, along with applicable ventilation conditions, are saved as a flue gas calibration record.

[0035] p represents the optical path number determined by the camera and the background marker, P represents the set of spaces traversed by the optical path, L represents the length of the optical path within the corresponding space in meters, determined by the shooting position, marker position, and spatial boundary. K represents the mass extinction coefficient of the corresponding optical path and space, in square meters per kilogram, representing the extinction intensity of a unit mass concentration of flue gas over a unit optical path, belonging to the optical parameters of flue gas. The subscript k is associated with the acquisition time period; the same optical path and space share K for the same time period under the same operating conditions. T represents the transmittance of the corresponding optical path at the beginning of the time period, and ln represents the natural logarithm. Concentration and transmittance are related by the following formula: ; The superscript minus and plus signs in T represent the lower and upper limits of the transmittance range extracted from the video, respectively. Taking the natural logarithm of the positive transmittance and its opposite, with the endpoints reversed, yields the observation constraints: ; When the lower limit of transmittance is 0, there is no upper limit on the right side. The unknown quantity is associated with the video segment according to the opening and acquisition time period. When observations are missing, only the corresponding observation constraints are missing; the unknown quantity and transport equations are retained. During the determination time period, G is set to 0 for each space within the investigation range, and the initial M and E remain non-negative unknowns.

[0036] When solving simultaneously, all conservation equations, transport equations, observation constraints, and parameter limitations for all time periods are placed in the same constraint group, with only one variable being repeatedly involved. First, linear constraints are established using upper and lower limits of the parameter intervals. For example, the upper and lower limits of R are substituted into the transport equation to limit the upper and lower bounds of F. In optical relations, the total extinction calculated by the lower limit of the coefficients must not exceed the upper limit of observation, and the total extinction calculated by the upper limit of the coefficients must not be lower than the lower limit of observation. The objective function is set to 0, and the simplex method of linear programming is used to find the mass values ​​that satisfy these constraints.

[0037] After obtaining the mass value, temporarily fix the mass, and then substitute R and K as unknowns back into the original equation and calibration constraints. At this point, all constraints are linear with respect to the parameters. Continue to use the simplex method to find a set of common parameters. When the mass and parameters together satisfy all the original constraints, record the mass value as a feasible value set. Only when the aforementioned linear constraints have been determined to be unsolvable are the corresponding parameter intervals excluded; if common parameters cannot be assigned to the mass values, alternately bisect the parameter intervals that still have width, and continue checking according to the segmentation level. If the check is not completed, retain the undetermined state, and do not consider a single solution failure as all constraints being unsolvable.

[0038] S4: Tracing back the undemanded quantity along the propagation relationship, grouping the associated video clips into a clip group, retrieving the corresponding video and supplementing the shooting, and updating the clip group; In a preferred embodiment of the present invention, composing and updating a fragment group includes: Historical videos corresponding to the area to be investigated and adjacent openings before the judgment period are read. Smoke-free images are selected according to the shooting location, and the transmittance value range is extracted using the aforementioned method. M, F, G, and E are set according to the historical acquisition period. The incremental quantity G within the history is kept as a non-negative unknown, and the same M is used for the historical end point and the starting point of the judgment period. Mass conservation, opening transport, video observation, and calibration parameter constraints are added spatially and time-by-time according to the aforementioned equations. The historical video gaps only lack the corresponding observation constraints.

[0039] The historical simultaneous solution follows the aforementioned parameter range partitioning and linear programming operations, adding all historical videos to the same constraint group simultaneously, without selecting parameters separately for each video. For example, if an opening connects space 1 and space 2, the same transport quantity is simultaneously recorded as the discharge item in space 1 and the entry item in space 2; after any video restricts the transport quantity, the other video continues to use the transport quantity restricted by the same constraint. The accumulated quantity at the end of the previous time period is also used as the initial accumulated quantity of the next time period, and all spaces and time periods are solved together.

[0040] The accumulated quantities that satisfy all constraints are represented by saving the complete historical constraints, not by enumerating consecutive values. The space, time, and other unsolved variables and calibration parameters involved in the original constraints are saved for each M. When reading a set of M later, there must exist other unsolved variables that together satisfy these constraints. Therefore, a found solution is only recorded as a feasible instance, not replacing the complete historical constraints, nor freely combining the upper and lower limits obtained by each M.

[0041] Select the exhaust side video based on the airflow direction. For the same shooting position and background marker, if the lower limit of the subsequent transmittance range is higher than the previous upper limit, select the video in between as the smoke reduction segment; if the smoke area is no longer identified and the transmittance range includes a smoke-free reference value, select the video from the time the smoke appeared until that moment as the smoke dissipation segment.

[0042] Starting from the F corresponding to the selected segment, first read the upstream space M in the transport equation, then read the inflow, other discharge, and G in the conservation equation of that space. Trace along each smoke inlet opening to the adjacent upstream space; when encountering M, read the conservation equation of the previous time period and its unsolved quantities. Combine the smoke inlet videos associated with the unsolved quantities, the previous smoke discharge videos, and the initially selected videos into a segment group. The segment group is the set of videos that are linked by these equations and unsolved quantities and need to be solved together subsequently.

[0043] Record the examined spaces, time periods, openings, and quantities to be determined. When encountering the same quantity repeatedly, retain the connection but do not expand it again. Missing observations are not used as the endpoint for tracing. When tracing back to a saved historical constraint M, connect the complete historical constraint according to space and time; when tracing back to the starting point of a usable video, retain the non-negative starting M. When different segment groups use the same F or M, associate the segment groups, and the quantity to be determined continues to use the same value.

[0044] Retrieve local video from the corresponding camera according to the opening and acquisition time period of the segment group. Videos from the same camera and shooting position that are consecutive in time are connected according to the acquisition time. Multiple video streams are matched according to a unified time, and gaps are retained in the original time period. Record the camera and acquisition time period corresponding to the video that has not yet been acquired. Check the camera recording time period and shooting position record. Retrieve locally recorded videos that have not been transmitted back. After obtaining the content, check whether the opening is visible and whether the transmittance can be extracted. For the time period that was not shot, interrupted, or for which the transmittance cannot be obtained, retain the observation gaps and the quantities to be determined.

[0045] Supplementing the footage with images corresponding to the missing observation openings, recording the shooting location, exposure parameters, and new acquisition time, and extracting the new flue gas changes and transmittance ranges. The starting point M of the new time period uses the same unknown quantity as the previous ending point M, adding the new transport equation and observation constraints, and then extending the segment group along the original connections. Historical videos can be retrieved to fill corresponding historical gaps; supplementary shooting only extends from the actual acquisition time, while the original historical observation gaps are retained, resulting in an updated segment group.

[0046] S5: Combine the calculations of the same unknown quantity in each segment group, respectively prohibiting and allowing the addition of new flue gas within the range to be investigated, and obtain the solution results; In another preferred embodiment of the present invention, the solution results include: The transmittance range of the video clip is converted into a constraint on the amount of smoke using the aforementioned optical relationship, denoted as the observation constraint. Specifically, the upper and lower limits of the transmittance range are taken as their natural logarithms and their opposites, and the endpoints are swapped to limit the sum of smoke extinction in each space within the corresponding optical path. Each observation constraint corresponds to a ventilation opening, shooting position, and acquisition time.

[0047] Before solving each segment group individually, the original value range, historical accumulation constraints, and calibration parameter limitations for each unknown quantity are saved. Feasible values ​​for each segment group under conditions where new flue gas addition is prohibited are checked separately, and values ​​obtained from one segment group are not assigned to other segments. Subsequently, the observation constraints and flue gas transport relationships of all segment groups are aggregated. Only one unknown quantity is set for the transport volume at the same opening and time period, and for the accumulation volume at the same time in the same space. Related calibration parameters are also taken from common values ​​according to the applicable operating conditions. Unknown quantities and transport equations corresponding to missing observations continue to participate in the solution process.

[0048] In the joint solution process, the transport parameters are first used to limit the transport volume of each opening, and the optical parameters are used to limit the extinction of each optical path. These, along with the conservation equations, form a linear constraint that includes all feasible values ​​of the original constraints. The objective function is set to 0, and non-negative mass values ​​are obtained using linear programming. After obtaining the mass values, the mass is temporarily fixed, and the calibration parameters are used as the variables to be solved. These are substituted back into the original transport equations, observation constraints, and calibration constraints to continue solving for a set of common parameters. Only when the mass and parameters together satisfy all the original constraints is it recorded as a feasible solution.

[0049] When common parameters cannot be assigned to candidate qualities, the remaining parameter intervals are bisected in turn, and the solution is continued according to the segmentation level. If the linear constraint corresponding to a certain parameter range has been determined to be unsolvable, that range is excluded; only when all allowed ranges are excluded is the constraint deemed infeasible. If there are still unexcluded ranges and no values ​​satisfying the original constraints have been obtained, it is recorded as an incomplete decision.

[0050] When adding new flue gas internally is prohibited, the internal addition amount in each space within the investigation range during the judgment period is set to 0, while the external input amount and initial accumulation amount are retained. When adding new flue gas internally is permitted, the above restriction of equal to 0 is lifted, restoring the non-negative value range, while other constraints remain the same. Two joint solutions are performed separately. If a feasible solution exists when it is prohibited, the corresponding value is saved as the updated feasible value set, and the investigation range is judged as undetermined; if the constraints are infeasible when prohibited but a feasible solution exists when permitted, internal smoke generation is determined. If the constraints are infeasible when permitted, the judgment is not completed, or the current air volume and temperature / humidity exceed the applicable range of the flue gas calibration record, it is judged as undetermined.

[0051] For the areas identified as having internal smoke production, observation constraints are temporarily removed group by group in sequence. Each time the original value range, historical accumulated constraints, and calibration parameter limitations saved before the individual solution are reloaded, the previously removed states are retained, and the value ranges that were previously tightened by the removed observation constraints are not reused. Transport equations and relationships using the same unknown variable are retained. When the same observation constraint is referenced by other segments that have not been removed, those references remain effective.

[0052] Under the condition that no new flue gas is allowed inside, the solution is recombined. If the constraints are still determined to be infeasible, the set of observation constraints is removed; if a feasible solution is obtained again or the determination is not completed, the set of observation constraints is restored. After processing each set, the last retained segment set and its constraints are saved. Removing observation constraints does not delete the original video; the original video is still saved according to the camera and acquisition time period.

[0053] The determination that the area under investigation contains internal smoke production or is undetermined is taken as the solution result, and the determination time period, the status of the two joint solutions, and the last retained fragment group are associated. When a feasible solution is obtained, the feasible value group is saved at the same time; when it is determined to be infeasible, the exclusion record of the corresponding parameter range is saved; when the determination is not completed, the parameter range that has not yet been excluded is retained.

[0054] S6: Based on the solution results and the segment group, the range to be investigated is divided into segments. The range where internal smoke production is determined to exist is recorded as the source range, and the source range or the range to be investigated is output.

[0055] In a preferred embodiment of the present invention, the scope to be investigated is divided according to the solution results and the segment group segmentation, including: Read the solution results and their associated fragment groups. If the solution result indicates the presence of internal smoke generation, record the corresponding area to be investigated as the source area, and save the equipment, connected spaces, and determination time period within it; if the solution result is indefinite, retain the area to be investigated, and do not directly specify the source based on the local smoke concentration.

[0056] Based on the ventilation openings and shooting positions in the segment group, the effective video time periods on both sides of each internal opening are collected, and it is verified whether each side covers the entire judgment time period. Internal openings that are continuously recorded on both sides and have no observation gaps are used as the dividing points. Taking the interior of the equipment and the connected spaces as the basic objects, these openings are temporarily regarded as the boundaries between ranges, and objects with other internal openings that are still connected are grouped into the same sub-range. Openings with recording gaps are not used as separate dividing boundaries.

[0057] After partitioning, all flue gas transport relationships, historical accumulation constraints, calibration parameter limitations, and observation constraints involved in the current determination are retained for the original area to be investigated. The same transport rate continues to be used on both sides of the opening of the sub-area, and the same quantity to be determined continues to be used for the accumulation rate at the same time. The partitioning only determines the spatial range that needs to be examined separately, and does not change the flue gas flow across the sub-area to unrelated external inputs.

[0058] First, read the feasible value set when adding smoke to the entire space within the original investigation range, and verify that it still satisfies the current constraints. Then, sequentially select sub-ranges, setting all internal additions within that sub-range to 0 during the decision period, while keeping the internal additions of the remaining sub-ranges non-negative and unknown, and retaining the external input and initial accumulated values. Re-examine all segment groups according to the aforementioned joint solution process, solving not only the video corresponding to the investigated sub-range.

[0059] If prohibiting the addition of new flue gas within a sub-range is infeasible, and a feasible solution exists after lifting the prohibition, then that sub-range is recorded as the source range. After each inspection, the prohibition condition is lifted before inspecting the next sub-range; prohibition conditions from previous sub-ranges cannot be accumulated. If multiple sub-ranges meet the conditions, they are saved separately, without considering flue gas intensity.

[0060] If no suitable sub-range is found, the original source range is retained. For example, if there is a feasible solution when adding new flue gas to sub-range 1 is prohibited but adding it to sub-range 2 is allowed, and conversely, if there is also a feasible solution when adding it to sub-range 1 is prohibited but adding it to sub-range 2 is allowed, then the two sub-ranges can provide different source interpretations, and their original source ranges are retained. Sub-ranges not recorded as source ranges are not marked as smokeless or excluded from the possibility of smoke production within them.

[0061] Save the video clip, shooting location, acquisition time period, and solution results corresponding to each judgment. When the camera turns, zooms, or the video exits the analysis, the video that has already participated in the judgment is still saved according to the original acquisition information, and the original judgment continues to correspond to the original judgment time period. Check whether the ventilation opening is still in the frame according to the new shooting location. For subsequent time periods that were not shot or for which transmittance could not be extracted, retain the quantities to be determined, mark the corresponding propagation relationship as an unverified relationship, and record the start and end times.

[0062] When a video exits analysis but the camera retains local recording, the corresponding video is marked as pending retrieval; exiting analysis is not directly equated to stopping recording. After a new video restores coverage, its observation constraints are added to the extended decision time period, and joint solution and sub-range verification are performed again to update the range attribution. The decision records of the original video and the original time period are retained.

[0063] Outputs the verified source range and judgment time period. The source range is defined by the actual location of the device and the connected space. If it is determined to be untimely, the output will show the range to be investigated, along with the unverified relationship and the recorded gap time period, as well as the camera, acquisition time period, and recorded local storage location for the video that has not yet been retrieved.

[0064] This invention bases source determination on whether the entire process of smoke propagation is consistent. Smoke appearing near equipment could originate from internal smoke generation, external introduction, or the release of previously accumulated smoke. When the initial record is incomplete, these possibilities are initially retained. As the smoke weakens, becomes intermittent, or dissipates, it is examined whether external introduction and accumulated release can simultaneously explain the changes at various points. Only when internal smoke generation must be introduced to explain the change is the corresponding source confirmed. Evidence gaps caused by changes in camera rotation, zoom, and transmission remain as unknowns. Subsequent partial footage needs to be verified against the existing process to avoid misinterpreting observations as changes in smoke source and to gradually narrow down the range of sources that can be supported by evidence.

[0065] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A multi-channel streaming media data analysis method based on artificial intelligence, characterized in that, Includes the following steps: S1: Acquire multiple video feeds, ventilation data, flue gas calibration records, and flue gas recognition models. Input the multiple video feeds into the flue gas recognition model to extract flue gas changes and video clips. S2: Establish the transmission relationship based on ventilation data, and combine equipment and connected spaces where flue gas changes are not continuously recorded to form the scope to be investigated; S3: Based on the flue gas calibration records and propagation relationships, set the external input volume, accumulated volume, internal new volume, and ventilation opening delivery volume as quantities to be determined and associate them with video clips; S4: Tracing back the undemanded quantity along the propagation relationship, grouping the associated video clips into a clip group, retrieving the corresponding video and supplementing the shooting, and updating the clip group; S5: Combine the calculations of the same unknown quantity in each segment group, respectively prohibiting and allowing the addition of new flue gas within the range to be investigated, and obtain the solution results; S6: Based on the solution results and the segment group, the range to be investigated is divided. The range where internal smoke production is determined to exist is recorded as the source range, and the source range or the range to be investigated is output.

2. The multi-channel streaming media data analysis method based on artificial intelligence according to claim 1, characterized in that, The process of extracting flue gas changes and video clips is as follows: Set black and white adjacent background markers in the background of the ventilation opening, capture smoke-free images, record the shooting position and exposure parameters of the smoke-free images and multiple video streams, and read the acquisition time of the multiple video streams. Acquire videos of labeled flue gas areas and train a flue gas recognition model. Input multiple videos into the flue gas recognition model to extract the flue gas areas. Based on the correspondence between pixel values, exposure parameters and brightness in the camera calibration data, as well as the calibration error, the pixel values ​​of the background markers in the smokeless image and multiple video streams are converted into a range of brightness values. Select a smoke-free image based on the shooting location. Within the brightness range, divide the black-and-white brightness difference of the corresponding background marker outside the smoke area in the smoke-free image by the black-and-white brightness difference in the multiple video streams to obtain the ratio. Multiply the black-and-white brightness difference of the background marker in the multiple video streams by the ratio, and then divide it by the corresponding black-and-white brightness difference in the smoke-free image to obtain the transmittance range. Arrange the flue gas areas and transmittance ranges according to the time of collection to form flue gas changes, extract video clips of the corresponding time periods and associate them with ventilation openings and shooting positions; The periods when the background is obscured by non-smoke objects, the video is interrupted, or the transmittance range cannot be obtained are marked as observation missing, and the corresponding ventilation openings and start and end times are recorded.

3. The multi-channel streaming media data analysis method based on artificial intelligence according to claim 2, characterized in that, The scope of investigation includes: Read the equipment layout, ventilation opening location and site boundary in the ventilation data, record the external space that is connected to the equipment through the ventilation opening as the connected space, associate the ventilation opening with the equipment interior or connected space on both sides, and establish the propagation relationship; Based on the changes in flue gas, the internal space of the equipment or connected space corresponding to the ventilation opening where flue gas appears is included in the scope of investigation, and the ventilation opening connecting the scope of investigation to the outside is found along the propagation relationship. Select video clips from both sides according to the ventilation opening and shooting position, and record the time period that is not covered by video clips and is missing from observation from the first identification of smoke to the current collection time as the recording gap; For ventilation openings with recording gaps on either side of the site boundary, incorporate the connected equipment interiors or adjacent spaces that are not included in the scope of investigation into the scope of investigation. Repeat the search, inspection, and merging until there are no recording gaps on either side of the found ventilation opening or it is located at the site boundary. Mark propagation relationships with gaps in the records at the merging points and site boundaries as unverified relationships, and associate the start and end times of the gaps in the records; mark propagation relationships in ventilation data where the connection points or airflow directions are not identified as unverified relationships.

4. The multi-channel streaming media data analysis method based on artificial intelligence according to claim 3, characterized in that, Setting up the desired quantity and associating it with video clips includes: The time period from the first identification of flue gas to the current collection time is recorded as the judgment time period. According to the equipment interior, connected space and collection time period, the external input amount, the accumulated amount, the internal new amount and the ventilation opening delivery amount are set as non-negative undetermined quantities. The discharge and inflow volumes on both sides of the same ventilation opening during the same sampling period use the same required quantity, and the accumulated volume at the end of the previous sampling period and the beginning of the next sampling period use the same required quantity. Based on the space volume, airflow direction, air volume, temperature and humidity in the ventilation data, and combined with the flue gas conservation relationship, establish the flue gas transport relationship; Acquire measurement data of smoke inlet, smoke outlet, and accumulated release during commissioning, along with corresponding transmittance ranges. Combine this with the flue gas transport relationship to calibrate the ranges of transport parameters and flue gas optical parameters, thus forming a flue gas calibration record. Associate the quantities to be determined with video clips based on ventilation openings and collection time periods; when observations are missing, retain the corresponding quantities to be determined and flue gas transport relationships. During the judgment period, no new flue gas is allowed to be added within the area to be investigated. The amount of external input and the initial accumulation are retained. Based on the flue gas transport relationship, the range of transport parameters and flue gas optical parameters in the flue gas calibration record and the shooting position, the transmittance range corresponding to the value of the quantity to be determined is calculated. The calculated transmittance range and the transmittance range of the corresponding video segment are used together as constraints to solve the flue gas transport relationship. The values ​​of the unknown quantity that satisfy the constraints are recorded as the feasible value set.

5. The multi-channel streaming media data analysis method based on artificial intelligence according to claim 4, characterized in that, Composing and updating fragment groups includes: Read historical videos before the judgment period, extract the transmittance value range and set the quantity to be determined according to the collection period. Allow the addition of flue gas inside within the period covered by the historical video. Solve the problem by combining the flue gas transport relationship with the transmittance value range. Save all accumulated quantity values ​​that satisfy the constraints and the constraints between each accumulated quantity. Based on the changes in flue gas, select the exhaust video segments that record the weakening or disappearance of flue gas, trace back along the flue gas transport relationship to find the quantities to be calculated in the exhaust calculation, and form a segment group by combining the associated inlet video segments, the previous exhaust video segments, and the selected exhaust video segments. When tracing back to the accumulated quantity, continue reading the quantity to be determined from the previous acquisition period. When tracing back to the saved accumulated quantity constraints, inherit the corresponding accumulated quantity values ​​and the constraints between each accumulated quantity. When tracing back to the starting point of video recording, retain the starting accumulated quantity as the quantity to be determined and associate the segments that use the same quantity to be determined. Retrieve local video from cameras according to the ventilation openings and acquisition time periods within the segment group, connect video segments according to acquisition time, and record the cameras and acquisition time periods corresponding to video segments that have not yet been retrieved. Read the camera recording time period, check the untransmitted video and the missing observations, retain the quantities to be determined corresponding to the missing observations, supplement the shooting according to the ventilation openings corresponding to the missing observations, and extend the segment group according to the newly added acquisition time.

6. The multi-channel streaming media data analysis method based on artificial intelligence according to claim 5, characterized in that, The solution results include: The range of transmittance values ​​for video segments is defined as the constraint on the quantity to be determined, and the range of values ​​for the quantity to be determined before each segment group is solved individually is recorded. By combining the observation constraints and flue gas transport relationships of each segment group, the feasible value set can be solved; During the judgment period, the addition of new flue gas within the scope to be investigated is prohibited and allowed respectively, and all segment groups are solved together; when the addition of new flue gas within the scope is prohibited, there is a feasible solution, the feasible value group is updated and the scope to be investigated is determined to be undetermined; when the constraint is not feasible when the addition of new flue gas within the scope is prohibited, but when the addition of new flue gas within the scope is allowed, there is a feasible solution, and the scope to be investigated is determined to have internal smoke generation. The observation constraints corresponding to each segment group are temporarily removed, while the flue gas transport relationship and the unsolved quantities used by each segment group are retained. The solution is then re-solved from the recorded range of unsolved quantities. If the constraints are still infeasible when adding new flue gas is prohibited, they are removed. If a feasible solution is obtained again when adding new flue gas is prohibited, or if the determination is not completed, the observation constraints are restored. Read the air volume and temperature and humidity during the debugging period, record the corresponding value range as the calibration range, and obtain the current air volume and temperature and humidity; when it is not feasible to constrain the addition of internal smoke, the current air volume and temperature and humidity exceed the calibration range, or the judgment cannot be completed, the range to be investigated is judged as undetermined. The determination that the area to be investigated contains internal smoke production or is undetermined is taken as the solution result, and the last group of segments with observation constraints is associated with the solution result.

7. The multi-channel streaming media data analysis method based on artificial intelligence according to claim 6, characterized in that, Based on the solution results and the segment group breakdown, the scope to be investigated includes: Based on the solution results, the area to be investigated that is determined to have internal smoke production is recorded as the source area, and associated with the judgment time period; The source range is divided along the internal ventilation openings continuously recorded by the segment group within the judgment period to obtain sub-ranges, and the flue gas transport relationship of the original range to be investigated and the quantities to be determined that are used by each segment group are retained; Within the judgment period, sub-ranges are selected sequentially, allowing new smoke to be added within the remaining sub-ranges, and prohibiting and allowing new smoke to be added within the selected sub-ranges respectively, and then solving all segment groups together; The sub-range where adding new flue gas into the prohibited sub-range is not feasible under constraints, but where adding new flue gas into the allowed sub-range has a feasible solution, is denoted as the source range. Sub-ranges that meet the conditions are retained, and the original source range is retained if no sub-ranges meet the conditions. Save the segment group corresponding to the solution results. When the camera turns, zooms, or the video exits the analysis, the saved video segments are retained. For subsequent periods without shooting, the corresponding undetermined quantities are retained. The propagation relationship of the corresponding ventilation opening is marked as an unverified relationship. Output the source range and judgment period. If it is determined that the range to be investigated is not output on time, it will also output the camera and acquisition period corresponding to the unverified relationship and the video clip that has not yet been retrieved.

Images