Camera adaptive method suitable for multiple scenes

Abstract

The invention relates to the technical field of camera intelligent control, in particular to a camera self-adaption method suitable for multiple scenes. A camera adaptive method suitable for multiple scenes comprises the following steps: S1, acquiring a video stream shot by a camera, and configuring at least 2N corresponding application scenes for the video stream of the camera, each application scene corresponding to an adaptive model; and S2, processing a video stream shot by the camera corresponding to the application scene by using the adaptive model corresponding to each application scene to obtain the fitness of the adaptive model for the camera in multiple scenes. According to the method, the high peak, the middle peak and the low peak of an application scene are marked, the processing weight is intelligently adjusted, the switching barrier factor is evaluated in combination with historical data, and LSTM prediction and visual display are utilized, so that the robustness and the adaptive ability of the system are enhanced, and efficient operation in a variable environment is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of camera intelligent control, and in particular to a camera self-adaptation method applicable to multiple scenarios. Background Art

[0002] The advantage of the camera adaptive method is that it can achieve stable image quality and efficient monitoring effects under different lighting, weather and environmental conditions through intelligent adjustment of parameters and algorithm optimization. The disadvantage of the camera adaptive method in multiple scenes is that the image quality and stability decrease under extreme conditions, and the intelligent adjustment algorithm has misjudgment and response delay. The advantage of the existing solution of the camera adaptive method in multiple scenes is that it can optimize the image quality under different conditions through intelligent adjustment. But the disadvantage is that the effect is not good in extreme environments and there are algorithm misjudgment and response delay. Especially under extreme environmental conditions, such as extremely high temperature, low temperature, strong wind, heavy rain, and dense fog, the self-adjustment ability of the adaptive method is often not enough to cope with these harsh environmental challenges, resulting in serious degradation of image quality or even failure to work properly. In addition, the performance of sensors in extreme environments will also be affected, further exacerbating the problems of image quality and stability. Therefore, how to improve the adaptive ability and robustness of cameras in extreme environments is still a technical problem that needs to be solved urgently. Summary of the invention

[0003] In order to overcome the disadvantage of poor adaptability in extreme environments, the present invention provides a camera adaptation method applicable to multiple scenarios.

[0004] The technical implementation scheme of the present invention is: a camera adaptation method applicable to multiple scenes, comprising the following steps: S1: Obtain the video stream captured by the camera, and configure at least 2N corresponding application scenarios for the video stream of the camera, each application scenario corresponds to an adaptive model; S2: Using the adaptive model corresponding to each application scenario, the video stream captured by the camera corresponding to the application scenario is processed to obtain the adaptability of the adaptive model for multiple camera scenarios; S3: Visualize the adaptability of the adaptive model for multiple camera scenarios.

[0005] Preferably, the method obtains a video stream captured by a camera, and configures at least 2N corresponding application scenarios for the video stream of the camera, each application scenario corresponds to an adaptive model, including: a method for constructing an adaptive model is to first obtain the number of application scenarios, and then sort the scene switching order of the application scenarios according to time sequence, construct two identical roulette wheels, and the capacity of each roulette wheel is equally divided into K sectors of the same area, and finally the application scenes are placed in the first roulette wheel in time sequence, and the camera switching obstacle factor data in each application scene is extracted at the same time, and the switching obstacle factor data is placed in the second roulette wheel in the order in which the corresponding application scenes are placed in the first roulette wheel, and the first roulette wheel and the second roulette wheel are analyzed.

[0006] Preferably, the method of first obtaining the number of application scenarios includes: obtaining at least 2N application scenarios existing in T working days in the environment where the camera is located, dividing the more than 2N application scenarios according to the duration and idle time of each scene when it is not switched once, and similarly dividing the duration and idle time of switching obstruction, using the first division as the data of the first roulette wheel, and similarly, the second division as the data of the second roulette wheel.

[0007] Preferably, the scene switching order of the application scenarios is then sorted in chronological order to construct two identical roulette wheels, each of which has a capacity equally divided into K sectors of equal area, including: firstly, the application scenarios are sorted in chronological order, and T / M sequences are constructed based on T working days, where M represents the number of time periods divided in each working day, and the capacity of each sequence is 1 / MT / M; the application scenarios are put into the sequence in sequence according to the sorting results, and the application scenario type, duration, and number of occurrences are recorded in the sequence; the duration and number of occurrences are normalized and then multiplied; the application scenarios in the sequence are screened to obtain the application scenario with the largest product of duration and number of occurrences, until all application scenarios are screened.

[0008] Preferably, the screening of the application scenarios in the sequence includes: adding one to the number of occurrences of the application scenario each time, recording the duration according to the actual duration of each occurrence, and finally screening out the application scenario with the largest product of duration and number of occurrences within the maximum sequence range.

[0009] Preferably, the application scenarios are finally placed into the first roulette wheel in order of time, and the camera switching obstacle factor data in each application scenario is extracted, and the switching obstacle factor data is placed into the second roulette wheel in the order in which the corresponding application scenarios are placed into the first roulette wheel, and the first roulette wheel and the second roulette wheel are analyzed, including: the data in the first roulette wheel is marked as high peak, middle peak, and low peak in sequence according to the sequence results, and the processing time is high peak < middle peak < low peak; the camera switching obstacle factor data includes, firstly, obtaining the first switching error data of the camera in the same application scenario and the switching error data between different scenarios in the historical application scenario, and secondly taking the inverse ratio of the sorting result in the sequence as the second error switching data of the same application scenario, and finally multiplying the first switching error data, the switching error data, and the second error switching data of the camera, the size of the area of ​​each partial sector in the roulette wheel is the normalized value of the product data, and then creating a roulette wheel for each application scenario in the first roulette wheel.

[0010] Preferably, the adaptive model corresponding to each application scenario is used to process the video stream captured by the camera corresponding to the application scenario to obtain the fitness of the adaptive model for multiple camera scenarios, including: determining the fitness of each application scenario in the first roulette wheel according to the sequence results, and dividing the fitness into a first fitness, a second fitness, and a third fitness in descending order according to the product result of the number of scene occurrences and the duration; and secondly dividing the second roulette wheel.

[0011] Preferably, the second roulette wheel is divided, including: the size of each sector in the second roulette wheel is a normalized value, and the size of each value represents the size of the fitness; according to the scenario applied in the first roulette wheel, the roulette wheel corresponding to the application scenario is randomly selected, and the fitness in the corresponding roulette wheel is adjusted according to the scene switching corresponding to the selection result or the adaptation result in the same scene, including the first fitness, the second fitness, the third fitness in the first roulette wheel and the fitness of the second roulette wheel, that is, the sector area, and the model is used to monitor the overall operation of the system.

[0012] Preferably, the use of the model to monitor the overall operation of the system includes: using the LSTM model to predict the operation process of the adaptive model; verifying the prediction results of the LSTM model based on a test data set; and using a fuzzy inference model to output the modification direction of the adaptive model based on the prediction results of the LSTM model.

[0013] Preferably, the adaptability of the adaptive model to multiple camera scenarios is visualized, including: collecting the roulette wheel rotation process after the roulette wheel is constructed, recording the camera adaptation process and adaptation results after each rotation process, and visually displaying the roulette wheel rotation results and the camera adaptation results.

[0014] Beneficial effects: The present invention realizes a camera adaptation method suitable for multiple scenarios through a multi-level design. First, by placing the application scenarios into the first roulette wheel in chronological order and marking them as high peak, medium peak and low peak, the system accurately locates the importance of different application scenarios, thereby giving priority to the scenes with high importance. Secondly, the switching barrier factor data in the second roulette wheel is divided into sectors by normalized values, and the processing weight of each application scenario is intelligently adjusted, so that the system can flexibly respond to different scenarios. In addition, combined with a variety of switching error data in historical data, the system more comprehensively evaluates the switching barrier factor of each application scenario and enhances robustness. The LSTM model is used to predict the next application scenario in advance, and the system prepares the corresponding adaptive model in advance, so as to respond quickly. By visually displaying the roulette wheel rotation process and the adaptive results, the user can intuitively understand the adaptive situation of the system in different application scenarios, which is convenient for monitoring and debugging. Finally, by combining the advantages of deep learning and fuzzy logic, the system not only performs well in complex and changeable environments, but also significantly enhances the adaptive ability to ensure that the system can work normally in various environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of a camera self-adaptation method applicable to multiple scenes of the present invention; Figure 2 A schematic diagram of the adaptive model construction process of the present invention; Figure 3 A schematic diagram of the process of constructing a roulette wheel according to the present invention. DETAILED DESCRIPTION

[0016] 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.

[0017] A camera adaptation method suitable for multiple scenes, such as Figure 1 As shown, the following steps are included: S1: Obtain the video stream captured by the camera, and configure at least 2N corresponding application scenarios for the video stream of the camera, each application scenario corresponds to an adaptive model; S2: Using the adaptive model corresponding to each application scenario, the video stream captured by the camera corresponding to the application scenario is processed to obtain the adaptability of the adaptive model for multiple camera scenarios; S3: Visualize the adaptability of the adaptive model for multiple camera scenarios.

[0018] The video stream captured by the camera is obtained, and at least 2N corresponding application scenarios are configured for the video stream of the camera, each application scenario corresponds to an adaptive model, including: the adaptive model is constructed by first obtaining the number of application scenarios, and then sorting the scene switching sequence of the application scenarios in chronological order, constructing two identical roulette wheels, each roulette wheel capacity is equally divided into K sectors of the same area, and finally the application scenes are placed in the first roulette wheel in chronological order, and the camera switching obstacle factor data in each application scenario is extracted at the same time, and the switching obstacle factor data is placed in the second roulette wheel in the order in which the corresponding application scenarios are placed in the first roulette wheel, and the first roulette wheel and the second roulette wheel are analyzed.

[0019] Further explanation is that the first roulette is understood as follows: assuming that the working environment of the camera is spring, summer, autumn and winter, the first roulette is placed in the order of spring, summer, autumn and winter. Similarly, the scenes of the working environment of the camera include working scenes with large passenger flow, general passenger flow, small passenger flow, zero passenger flow or windy and sandy scenes, rainy and snowy scenes, heavy fog scenes, hail scenes and the like. The specific number of multiple scenes is determined according to the actual working environment of the camera (the value of K is greater than or equal to the number of application scenes, and K is a positive integer greater than or equal to 2), and is not limited to or restricted to the four working scenes or passenger flow scenes and weather scenes cited as examples; the second roulette is understood as follows: assuming that the working environment of the camera is spring, summer, autumn and winter, the problems faced by the camera in each season of spring, summer, autumn and winter (i.e., switching barrier factor data) are also placed in the second roulette; the volume size (i.e., sector area) of the first roulette is consistent with that of the second roulette, and the size of the small volume (i.e., sector area) in each roulette after classification is consistent.

[0020] First, obtain the number of application scenarios, including: obtain at least 2N application scenarios that exist in T working days in the environment where the camera is located, divide more than 2N application scenarios according to the duration and idle time of each scene when it is not switched once, and similarly divide the duration and idle time of switching obstruction, and use the first division as the data of the first roulette wheel, and similarly, the second division as the data of the second roulette wheel.

[0021] To further explain, assuming there are four application scenarios, namely wind and sand, thunder, rain and snow, the roulette wheel is divided into four equal volumes, starting at 0:00 and ending at 12:00. The duration of the wind and sand application scenario is 0:00-2:00, with a idle time of 2:00-3:00. The duration of the thunder application scenario is 3:00-6:00, with no idle time. The duration of the rain application scenario is 6:00-7:00, with a idle time of 7:00-9:00. The duration of the snow application scenario is 9:00-12:00, with no idle time. The data division of the second roulette wheel is based on the data division of the first roulette wheel, but the division content is different from that of the first roulette wheel.

[0022] Secondly, the scene switching order of the application scenarios is sorted in chronological order, and two identical roulette wheels are constructed. The capacity of each roulette wheel is equally divided into K sectors of equal area, including: firstly, the application scenarios are sorted in chronological order, and T / M sequences are constructed based on T working days (M represents the number of time periods divided in each working day). The capacity of each sequence is 1 / MT / M. The application scenarios are put into the sequence in sequence according to the sorting results. The application scenario type, duration, and number of occurrences are recorded in the sequence. The duration and number of occurrences are normalized and then multiplied. The application scenarios in the sequence are screened to obtain the application scenario with the largest product of duration and number of occurrences, until all application scenarios are screened.

[0023] The application scenarios in the sequence are screened, including: the number of occurrences of the application scenario is increased by one each time, the duration is recorded according to the actual duration of each occurrence, and finally the application scenario with the largest product of duration and number of occurrences is screened out within the maximum sequence range.

[0024] To further explain, the sequence is further understood as follows: assuming that within the specified T (T includes but is not limited to 1 hour, 1 day, 2 days, and the monitoring cycle of a camera) working days, the application scenarios of the camera include rain, snow, sand, rain, snow, sand, thunder, rain, sand, rain, snow, and rain, and the duration and number of occurrences of each application scenario are obtained; the specific analysis process is understood as setting the maximum volume of the sequence to 3, and the content in the volume is the application scenario type, duration, and number of occurrences. Now, the analysis process of 12 application scenarios is specifically illustrated by examples. The first round of processing [rain, 0:00-0:50, 1, snow, 1:00-1:20, 1, sand, 2:00-3:00, 1], the second round of processing [rain, 0:00-0:50, 1, rain, 3:20-4:00, 1, sand, 2:00-3:00, 1], In the third round of processing, the duration of snow in the fifth application scenario is 5:30-5:40, [rain, 90, 2, snow, 30, 2, sand, 2:00-3:00, 1]. It should be noted that in the first and second rounds of processing, for the convenience of understanding, the duration is based on a certain moment to a certain moment. At the same time, in the second round of processing, since the fourth application scenario is raining and the product of the duration of snow in the application scenario and the number of occurrences in the first round of processing is the smallest, the fourth application scenario replaces the second application scenario. It should also be noted that for the convenience of calculation, the result of simply multiplying the duration of 20 minutes and the number of occurrences 1 is used for comparison. In actual situations, as the duration and the number of occurrences increase, the duration and the number of occurrences are normalized and multiplied for the convenience of calculation and understanding. The processing methods of the remaining application scenarios are the same as those of the first five application scenarios.

[0025] Finally, the application scenarios are placed in the first roulette wheel in order of time, and the camera switching obstacle factor data in each application scenario is extracted, and the switching obstacle factor data is placed in the second roulette wheel in the order in which the corresponding application scenarios are placed in the first roulette wheel. The first roulette wheel and the second roulette wheel are analyzed, including: the data in the first roulette wheel are marked as peak, middle peak, and low peak (processing time peak < middle peak < low peak) in sequence according to the sequence results; the camera switching obstacle factor data includes, first, obtaining the first switching error data of the camera in the same application scenario and the switching error data between different scenarios in the historical application scenarios, and then taking the inverse ratio of the sorting result in the sequence as the second error switching data of the same application scenario, and finally multiplying the first switching error data, switching error data, and second error switching data of the camera, the size of the area of ​​each partial sector in the roulette wheel is the normalized value of the product data, and then creating a roulette wheel for each application scenario in the first roulette wheel.

[0026] A further explanation is that the peak, medium and low peaks are understood as follows: the peak represents the application scenario with the largest number of occurrences and the longest duration, and the processing effect is good under the existing technical means, so the processing time is set short, and the processing time of the medium and low peaks is lengthened respectively; the data in the first roulette wheel is marked as peak, medium and low peaks in sequence according to the sequence results (processing time peak < medium peak < low peak) and is further understood as when a certain type of scenario occurs frequently and lasts long enough, the adaptability to the scenario is greater under the existing technology, so the height of the peak also reflects the adaptability to a certain scenario (for example, application scenarios with large traffic) or a certain type of scenario (for example, weather-related application scenarios).

[0027] For further explanation, assume that the switching obstacle factor data of the camera under different weather conditions are as follows: 1. Obtain historical data: Assume that in the historical data, the switching error data of the camera in different application scenarios are as follows: Wind and sand: The first switching error data is 0.8, and the switching error data between different scenes is 0.7. Thunder: The first switching error data is 0.7, and the switching error data between different scenes is 0.6. Rain: The first switching error data is 0.6, and the switching error data between different scenes is 0.5. Snow: The first switching error data is 0.5, and the switching error data between different scenes is 0.4. 2. Build the first roulette: Put the application scenarios into the first roulette in chronological order. Assume that the chronological order is: wind and sand, thunder, rain, snow. 3. Build the second roulette: Extract the camera switching obstacle factor data in each application scenario. And put the switching obstacle factor data into the second roulette in the order in which the corresponding application scenarios are put into the first roulette. 4. Analyze the first roulette: The data in the first roulette are marked as peak, middle peak, and low peak according to the sequence results (processing time peak < middle peak < low peak). Assume that after analysis, the marking results are: Rain: peak. Wind and sand: middle peak. Snow: middle peak. Thunder: low peak. 5. Analyze the second roulette: Calculate the switching barrier factor data for each application scenario: Wind and sand: The first switching error data is 0.8, the switching error data between different scenarios is 0.7, the second error switching data is 0.7, and the product data is 0.8*0.7*0.7=0.392. Thunder: The first switching error data is 0.7, the switching error data between different scenarios is 0.6, the second error switching data is 0.6, and the product data is 0.7*0.6*0.6=0.252. Rain: The first switching error data is 0.6, the switching error data between different scenarios is 0.5, the second error switching data is 0.5, and the product data is 0.6*0.5*0.5=0.15. Snow: The first switching error data is 0.5, the switching error data between different scenes is 0.4, the second error switching data is 0.4, and the product data is 0.5*0.4*0.4=0.08. Normalized values: Wind and sand: 0.392 / (0.392+0.252+0.15+0.08)≈0.45.

[0028] Thunder: 0.252 / (0.392+0.252+0.15+0.08)≈0.29.

[0029] Raining: 0.15 / (0.392+0.252+0.15+0.08)≈0.17.

[0030] Snowing: 0.08 / (0.392+0.252+0.15+0.08)≈0.09.

[0031] This design achieves: 1. Accurate positioning: By placing application scenarios in the first wheel in chronological order and marking them as peak, mid-peak, and low peak, the importance of different application scenarios can be accurately located. This ensures that the system gives priority to scenarios with high importance when processing different application scenarios. 2. Intelligent adjustment: The switching barrier factor data in the second wheel is divided into sectors by normalized values, and the processing weight of each application scenario is intelligently adjusted. This ensures that the system can flexibly adjust the processing strategy according to actual conditions when processing different application scenarios. 3. Robustness enhancement: By combining the first switching error data, the switching error data between different scenarios, and the second error switching data in the historical data, the switching barrier factor of each application scenario is more comprehensively evaluated. This improves the robustness of the system and makes it more robust when facing complex and changing application scenarios. Through this design, the system can not only better adapt to a variety of scenarios, but also optimize its own performance through intelligent decision-making and prediction mechanisms, ensuring that even if the next application scenario is not the currently selected application scenario, the system can still flexibly adjust and reduce misjudgment and response delays.

[0032] The adaptive model corresponding to each application scenario is used to process the video stream captured by the camera corresponding to the application scenario to obtain the fitness of the adaptive model for multiple camera scenarios, including: determining the fitness of each application scenario in the first roulette wheel according to the sequence results, and dividing the fitness into the first fitness, the second fitness, and the third fitness in descending order according to the product result of the number of scene occurrences and the duration; and then dividing the second roulette wheel.

[0033] Secondly, the second roulette wheel is divided, including: the size of each sector in the second roulette wheel is a normalized value, and the size of each value represents the size of the fitness; according to the application scenario in the first roulette wheel, the sector in the second roulette wheel corresponding to the application scenario is randomly selected, and the fitness in the corresponding roulette wheel (including the first fitness, the second fitness, the third fitness in the first roulette wheel and the fitness of the second roulette wheel) is adjusted according to the scene switching corresponding to the selection result or the adaptation result in the same scene (including the first fitness, the second fitness, the third fitness in the first roulette wheel and the fitness of the second roulette wheel), and the model is used to monitor the overall operation of the system.

[0034] The model is used to monitor the overall operation of the system, including: using the LSTM model to predict the operation process of the adaptive model; verifying the prediction results of the LSTM model based on the test data set; and using the fuzzy inference model to output the modification direction of the adaptive model based on the prediction results of the LSTM model.

[0035] Further explanation is that by combining the advantages of deep learning and fuzzy logic, the powerful feature learning ability of deep learning is utilized, and the advantages of fuzzy logic in processing uncertainty and fuzzy information are utilized; the combination of LSTM prediction and fuzzy reasoning improves the robustness of the system and performs more robustly in the face of complex and changing application scenarios; the entire framework is designed to automatically adjust the parameter settings of the camera according to the prediction results, thereby enhancing the system's adaptability and enabling the system to perform well in different environments; the loss function formula involved in verifying the prediction results is: ; in, Sample size, It is The actual value of the samples, is the predicted value.

[0036] A further explanation is that the LSTM model is used to predict the next application scenario in advance and prepare the corresponding adaptive model in advance. In this way, even if the next scenario changes, the system can respond quickly.

[0037] The adaptability of the adaptive model for multiple camera scenarios is visualized, including: collecting the roulette wheel rotation process after the roulette wheel is built, recording the camera adaptation process and adaptation results after each rotation process, and visually displaying the roulette wheel rotation results and camera adaptation results.

[0038] A further explanation is that by visually displaying the wheel rotation process and adaptive results, we can intuitively understand the adaptive status of the system in different application scenarios. This makes it easier to monitor and debug the system to ensure that it can work normally in various environments.

[0039] The above is a detailed introduction to the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A camera adaptation method applicable to multiple scenes, characterized by: The following steps are involved: S1: Obtain the video stream captured by the camera, and configure at least 2N corresponding application scenarios for the video stream of the camera, each application scenario corresponds to an adaptive model; S2: Using the adaptive model corresponding to each application scenario, the video stream captured by the camera corresponding to the application scenario is processed to obtain the adaptability of the adaptive model for multiple camera scenarios; S3: Visualize the adaptability of the adaptive model for multiple camera scenarios.

2. A camera adaptation method applicable to multiple scenes according to claim 1, characterized in that: The method obtains a video stream captured by a camera, and configures at least 2N corresponding application scenarios for the video stream of the camera, each application scenario corresponding to an adaptive model, including: a method for constructing an adaptive model is to first obtain the number of application scenarios, then sort the scene switching sequence of the application scenarios according to time, construct two identical roulette wheels, each roulette wheel capacity is equally divided into K sectors of the same area, finally the application scenes are placed in the first roulette wheel according to time order, and the camera switching obstacle factor data in each application scenario is extracted at the same time, and the switching obstacle factor data is placed in the second roulette wheel according to the order in which the corresponding application scenarios are placed in the first roulette wheel, and the first roulette wheel and the second roulette wheel are analyzed.

3. A camera adaptation method applicable to multiple scenes according to claim 2, characterized in that: The method first obtains the number of application scenarios, including: obtaining at least 2N application scenarios existing in T working days in the environment where the camera is located, dividing the more than 2N application scenarios according to the duration and idle time of each scene when it is not switched once, and similarly dividing the duration and idle time of switching obstruction, using the first division as the data of the first roulette wheel, and similarly, the second division as the data of the second roulette wheel.

4. A camera adaptation method applicable to multiple scenes according to claim 2, characterized in that: The second step is to sort the scene switching order of the application scenarios in chronological order, construct two identical roulette wheels, and divide the capacity of each roulette wheel into K sectors of equal area, including: firstly sorting the application scenarios in chronological order, constructing T / M sequences based on T working days, where M represents the number of time periods divided in each working day, and the capacity of each sequence is 1 / MT / M, and placing the application scenarios into the sequence in sequence according to the sorting results, recording the application scenario type, duration, and number of occurrences in the sequence, normalizing the duration and the number of occurrences and then multiplying them, screening the application scenarios in the sequence, and obtaining the application scenario with the largest product of duration and number of occurrences, until all application scenarios have been screened.

5. The camera adaptation method applicable to multiple scenes according to claim 4 is characterized in that: The screening of the application scenarios in the sequence includes: adding one to the number of occurrences of the application scenario each time, recording the duration according to the actual duration of each occurrence, and finally screening out the application scenario with the largest product of duration and number of occurrences within the maximum sequence range.

6. A camera adaptation method applicable to multiple scenes according to claim 2, characterized in that: Finally, the application scenarios are placed in the first roulette wheel in order of time, and the camera switching obstacle factor data in each application scenario is extracted, and the switching obstacle factor data is placed in the second roulette wheel in the order in which the corresponding application scenarios are placed in the first roulette wheel. The first roulette wheel and the second roulette wheel are analyzed, including: the data in the first roulette wheel are marked as high peak, middle peak, and low peak in sequence according to the sequence results, and the processing time is high peak < middle peak < low peak; the camera switching obstacle factor data includes, firstly, obtaining the first switching error data of the camera in the same application scenario and the switching error data between different scenarios in the historical application scenario, and then taking the inverse ratio of the sorting result in the sequence as the second error switching data of the same application scenario, and finally multiplying the first switching error data, the switching error data, and the second error switching data of the camera, the size of the area of ​​each partial sector in the roulette wheel is the normalized value of the product data, and then creating a roulette wheel for each application scenario in the first roulette wheel.

7. A camera adaptation method applicable to multiple scenes according to claim 1, characterized in that: The method uses the adaptive model corresponding to each application scenario to process the video stream captured by the camera corresponding to the application scenario to obtain the fitness of the adaptive model for multiple camera scenarios, including: determining the fitness of each application scenario in the first roulette wheel according to the sequence results, and dividing the fitness into a first fitness, a second fitness, and a third fitness in descending order according to the product result of the number of scene occurrences and the duration; and then dividing the second roulette wheel.

8. A camera adaptation method applicable to multiple scenes according to claim 7, characterized in that: The second roulette wheel is divided, including: the size of each sector in the second roulette wheel is a normalized value, and the size of each value represents the size of the fitness; according to the application scenario in the first roulette wheel, the sector in the second roulette wheel corresponding to the application scenario is randomly selected, and the fitness in the corresponding roulette wheel is adjusted according to the scene switching corresponding to the selection result or the adaptation result in the same scene, including the first fitness, the second fitness, the third fitness in the first roulette wheel and the fitness of the second roulette wheel, and the overall operation of the system is monitored by the model.

9. A camera adaptation method applicable to multiple scenes according to claim 8, characterized in that: The overall operation of the system is monitored by using the model, including: using the LSTM model to predict the operation process of the adaptive model; verifying the prediction results of the LSTM model according to the test data set; and using the fuzzy inference model to output the modification direction of the adaptive model based on the prediction results of the LSTM model.

10. A camera adaptation method applicable to multiple scenes according to claim 1, characterized in that: The adaptability of the adaptive model to multiple camera scenarios is visualized, including: collecting the roulette wheel rotation process after the roulette wheel is built, recording the camera adaptation process and adaptation results after each rotation process, and visually displaying the roulette wheel rotation results and the camera adaptation results.

Images