Airport high-definition array camera adaptive target detection algorithm selection method and system

Abstract

The application discloses an airport high-definition array camera adaptive target detection algorithm selection method and system in the technical field of target detection, and aims to solve the problem that target detection cannot adaptively select a suitable target detection algorithm according to scene information in the prior art. The method comprises the following steps: acquiring scene information; generating a matching number according to the scene information and in combination with a matching number generation rule; obtaining an algorithm serial number according to the matching number and in combination with a preset matching number and algorithm serial number correlation; and obtaining a matched target detection algorithm according to the algorithm serial number and in combination with a preset algorithm serial number and target detection algorithm correlation. The application is suitable for target detection, can adaptively select a target detection algorithm most suitable for a current scene for a high-definition array camera, has high timeliness, comprehensive evaluation dimensions, significantly improves detection efficiency and precision, reduces labor cost, and provides a basic guarantee for subsequent intelligent monitoring scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, in particular to an airport high-definition array camera adaptive target detection algorithm selection method and system. BACKGROUND

[0002] The development of science and technology inevitably brings many security risks, among which the security risks of airport runway traffic cannot be ignored. In order to reduce and effectively avoid these security risks, intelligent monitoring technology is essential. At present, airports mainly collect real-time video image data through high-definition array cameras to ensure data quality, and then use target detection algorithms to detect the position of specific targets from video images, thereby providing a basic guarantee for intelligent identification and early warning of security risks. Therefore, as the primary step of airport video monitoring, target detection is directly related to the quality of the entire monitoring system, and plays a crucial role.

[0003] Target detection is an important problem in the field of computer vision, which has also experienced two periods of development, traditional target detection and deep learning-based target detection. Existing target detection algorithms have become relatively mature.

[0004] However, due to the complexity of the scene, changes in lighting, similar colors, object occlusion and other problems, target detection has always been a hot and difficult topic in the field of computer vision. Each algorithm has its own exclusive field of expertise, some detect quickly, some detect accurately, some are suitable for detecting small targets, and some are suitable for detecting dense scenes.

[0005] Therefore, in the face of such an environment where safety factors are extremely important, how to adaptively select the corresponding appropriate target detection algorithm according to the environment, time period and other scene information of the high-definition array camera, so that the algorithm can exert its maximum ability, thereby improving the accuracy and efficiency of target detection and reducing labor costs, is a technical problem that needs to be solved at present. SUMMARY

[0006] The purpose of the present application is to overcome the shortcomings of the prior art and provide an airport high-definition array camera adaptive target detection algorithm selection method and system, which solves the problem that the appropriate target detection algorithm cannot be selected according to the scene information when performing target detection.

[0007] To solve the above technical problems, the present application is realized by adopting the following technical scheme:

[0008] On the one hand, the present application provides an airport high-definition array camera adaptive target detection algorithm selection method, comprising:

[0009] Obtaining scene information of a location where a high-definition array camera of an airport is located, wherein the scene information comprises meteorological information, location information and day and night information;

[0010] Generating a matching number according to the scene information and in combination with a matching number generation rule;

[0011] Obtaining an algorithm serial number according to the matching number and in combination with a preset association between the matching number and the algorithm serial number;

[0012] Obtaining a matched target detection algorithm according to the algorithm serial number and in combination with a preset association between the algorithm serial number and the target detection algorithm.

[0013] Further, after obtaining the matched target detection algorithm according to the algorithm serial number and in combination with the preset association between the algorithm serial number and the target detection algorithm, the method further comprises:

[0014] Determining whether the matched target detection algorithm is unique;

[0015] If yes, taking the unique matched target detection algorithm as a target detection algorithm to be selected;

[0016] If no, comparing and screening the matched target detection algorithms to take an optimal target detection algorithm as the target detection algorithm to be selected.

[0017] Further, the method of comparing and screening the matched target detection algorithms comprises:

[0018] Obtaining a main target detection category;

[0019] Obtaining average precision mean values of the target detection algorithms and average precision in the main target detection category according to the main target detection category and in combination with a preset algorithm evaluation index set, wherein the algorithm evaluation index set comprises the average precision mean values of the target detection algorithms, detection speed and average precision in various target detection categories;

[0020] Calculating comprehensive precision of the target detection algorithms according to a theoretical formula, wherein the theoretical formula comprises:

[0021] P = X * AP + Y * mAP

[0022] In the formula, P represents comprehensive precision, AP represents average precision in the main target detection category, mAP represents average precision mean, X represents a weight of average precision, and Y represents a weight of average precision mean;

[0023] Screening a target detection algorithm with maximum comprehensive precision from the target detection algorithms;

[0024] determining whether the target detection algorithm with the maximum comprehensive precision is unique;

[0025] if yes, the unique target detection algorithm is selected as the target detection algorithm to be selected;

[0026] if no, the target detection algorithm with the fastest detection speed is selected as the target detection algorithm to be selected according to a preset algorithm evaluation index set.

[0027] Further, the matching number generation rule specifically includes:

[0028] the scene information is spliced into a string and separated by a comma;

[0029] the string is converted into a fixed-length hash value by using a hash function;

[0030] the hash value is converted into an integer value;

[0031] the integer value is mapped to a matching number value range to generate a matching number.

[0032] Further, the method for determining the association between the algorithm serial number and the target detection algorithm includes:

[0033] various target detection algorithms to be used in an actual application scenario are obtained;

[0034] the various target detection algorithms are numbered to obtain the association between the algorithm serial number and the target detection algorithm.

[0035] Further, the method for determining the association between the matching number and the algorithm serial number includes:

[0036] various scene information applicable to a certain target detection algorithm is combined with a matching number generation rule to generate various matching numbers corresponding to the target detection algorithm;

[0037] the association between the algorithm serial number and the various matching numbers corresponding to the target detection algorithm is established according to the association between the algorithm serial number and the target detection algorithm;

[0038] the association between the algorithm serial number and the various matching numbers corresponding to each target detection algorithm is summarized to determine the association between the matching number and the algorithm serial number.

[0039] Further, after determining whether the matched target detection algorithm is unique, the method further includes:

[0040] Bind the airport high-definition array camera with the target detection algorithm to be selected, establish a binding relationship between the airport high-definition array camera and the target detection algorithm, aggregate the binding relationship between the airport high-definition array camera and the target detection algorithm, and construct an algorithm configuration table;

[0041] Start the airport high-definition array camera, and the airport high-definition array camera performs target detection according to the target detection algorithm bound in the algorithm configuration table, records the target detection result, and constructs a target detection result list, wherein the target detection result includes a target category, target coordinates, target confidence, and processing speed FPS;

[0042] Obtain airport comprehensive information, calculate pixel coordinate information of the target in the image, compare the pixel coordinate information of the real target in the image with the target detection result list, and construct a target detection comparison error list;

[0043] According to the target detection comparison error list, it is judged whether the comparison error exceeds a preset threshold, if yes, it is judged whether the migration training times are less than a preset threshold, and if yes, migration training is performed;

[0044] According to the target detection comparison error list, the original target detection algorithm model is retrained to obtain a new target detection algorithm model;

[0045] It is judged whether the new target detection algorithm model is better than the original target detection algorithm model, if yes, the new target detection algorithm model is covered to the original target detection algorithm model, otherwise manual intervention is applied.

[0046] Further, the pixel coordinate information of the target in the image is calculated, including:

[0047] According to the flight plan and air traffic control information in the airport comprehensive information, the number of flights and the passenger flow are obtained;

[0048] According to the radar information in the airport comprehensive information, the specific position information of the target is obtained;

[0049] According to the number of flights, the passenger flow and the specific position information of the target, the pixel coordinate information of the target in the image is calculated by using a coordinate conversion formula.

[0050] On the other hand, the application provides an airport high-definition array camera adaptive target detection algorithm selection system, comprising:

[0051] An information acquisition module is used to acquire scene information, wherein the scene information includes meteorological information, position information and day and night information;

[0052] A matching algorithm constructor is used to generate a matching number according to the scene information and a matching number generation rule.

[0053] An algorithm pool is configured to obtain an algorithm serial number according to the matching serial number and in combination with a preset association between the matching serial number and the algorithm serial number;

[0054] An algorithm selection module is configured to obtain a matched target detection algorithm according to the algorithm serial number and in combination with a preset association between the algorithm serial number and the target detection algorithm.

[0055] Further, the method further comprises:

[0056] An algorithm screening module is configured to determine whether the matched target detection algorithm is unique;

[0057] If yes, the matched unique target detection algorithm is taken as the target detection algorithm to be selected;

[0058] If no, each of the matched target detection algorithms is compared and screened, and the optimal target detection algorithm is taken as the target detection algorithm to be selected.

[0059] Compared with the prior art, the present application has the following beneficial effects:

[0060] 1. The present application can adaptively select the target detection algorithm most suitable for the current scene of the high-definition array camera by obtaining the scene information and matching it with the scene information applicable to the target detection algorithm, has high timeliness, comprehensive evaluation dimension, significantly improves the detection efficiency and precision, reduces the labor cost, and provides a basic guarantee for subsequent intelligent monitoring scenes;

[0061] 2. The present application can select the optimal selection that gives consideration to both the detection precision and the detection speed by screening the matched target detection algorithm according to the algorithm evaluation index, ensures that the target detection algorithm adopted is the most suitable selection at present, and has good practical application effect. DETAILED DESCRIPTION

[0062] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation to the present application. In the drawings:

[0063] Figure 1 is a flowchart of the airport high-definition array camera adaptive target detection algorithm selection method provided by the first embodiment of the present application;

[0064] Figure 2 is a flowchart of the airport high-definition array camera adaptive target detection algorithm selection method provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0065] The application will be further described below with reference to the drawings. The following examples are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0066] Example one:

[0067] The embodiment provides an airport high-definition array camera adaptive target detection algorithm selection method, which comprises the following steps:

[0068] S1: Obtain scene information of a position where an airport high-definition array camera is located, wherein the scene information comprises weather information, position information and day and night information.

[0069] For example, a certain scene information is cloudy, by the escalator and day. The source of the scene information can be a multi-source sensor arranged around the airport,

[0070] S2: Generate a matching number according to the scene information and in combination with a matching number generation rule.

[0071] The matching number generation rule specifically comprises:

[0072] A: Concatenate the scene information into a string and separate it with a comma.

[0073] Specifically, the weather information is weather, the position information is location, and the day and night information is (day or night), wherein the day and night information is represented by 0 and 1, 0 represents day, and 1 represents night; the above information is collected to obtain a string str=weather, location, day. For example, the above scene information (cloudy, by the escalator and day) is concatenated into str=cloudy, by the escalator, 0.

[0074] B: Convert the string into a fixed-length hash value by using a hash function.

[0075] Specifically, the hash function used in the embodiment is a SHA-256 hash function, but is not limited to this, and the specific limitation is not made. The above string (str=cloudy, by the escalator, 0) can be converted into a hash value e06cd9e92167ba069debf4a19c3a1cc3b754a0e43f111c789c49aeddc1fae3d8 by using the hash function.

[0076] C: Convert the hash value into an integer value.

[0077] For example, the first eight bytes of the hash value can be interpreted as a 64-bit unsigned integer, and the specific mode is as follows:

[0078] First, it is converted from hexadecimal number to binary number, and the result is: 11100000011011001101100111101001001000011011011101101011110110100011011110001111101000001110001000011100110011001101101010111101111010010000110000001110100001111100001000111111111101000100111000111000111001100011110101111;

[0079] Then, according to the division of each eight bits, eight eight-bit binary numbers are obtained: 11100000, 01101100, 11011001, 11101001, 00100011, 01101101, 11011010, 10111101;

[0080] Finally, it is converted from binary number to decimal number: 22410821724935109218189.

[0081] D: mapping the integer value into the matching number value range to generate the matching number.

[0082] Specifically, in the embodiment, the mapping of the integer value is completed by the modulo operation, but it is not limited thereto; the matching number value range is set to be: [1~weather number*location number*2].

[0083] S3: obtaining the algorithm serial number according to the matching number and the preset association between the matching number and the algorithm serial number.

[0084] Specifically, the determination method of the association between the matching number and the algorithm serial number in S3 includes:

[0085] Step 1: generating various matching numbers corresponding to a target detection algorithm according to various scene information applicable to the target detection algorithm and a matching number generation rule.

[0086] It can be understood that there are multiple scene information applicable to each target detection algorithm, and therefore there are multiple matching numbers generated accordingly.

[0087] Step 2: establishing the association between the algorithm serial number and various matching numbers corresponding to the target detection algorithm according to the association between the algorithm serial number and the target detection algorithm.

[0088] It should be noted that the target detection algorithm and the algorithm serial number are in a one-to-one relationship, and the algorithm serial number and the matching number are in a one-to-many relationship.

[0089] Step 3: The association between the algorithm serial numbers corresponding to each target detection algorithm and various matching numbers is summarized to determine the association between the matching number and the algorithm serial number.

[0090] S4: According to the algorithm serial number, and in combination with the preset association between the algorithm serial number and the target detection algorithm, the matched target detection algorithm is obtained.

[0091] Specifically, the method for determining the association between the algorithm serial number and the target detection algorithm in S4 comprises:

[0092] Step I: Obtain various target detection algorithms to be used in the actual application scenario.

[0093] It should be noted that the various target detection algorithms to be used are algorithms that are excellent in at least one aspect, such as YOLO series, Faster-RCNN, etc.

[0094] Step II: Number the various target detection algorithms to obtain the association between the algorithm serial number and the target detection algorithm.

[0095] S5: Determine whether the matched target detection algorithm is unique;

[0096] If yes, the unique target detection algorithm matched is selected as the target detection algorithm to be selected;

[0097] If not, the matched target detection algorithms are compared and selected, and the optimal target detection algorithm is selected as the target detection algorithm to be selected.

[0098] It should be noted that because the same scene information may be applicable to multiple target detection algorithms, if multiple target detection algorithms are matched, they need to be selected to select the most suitable one, so as to ensure the accuracy of target detection to be the best.

[0099] The present application can adaptively select the target detection algorithm most suitable for the current scene through the above steps, thereby significantly improving the detection efficiency and accuracy, and providing a basic guarantee for subsequent intelligent monitoring scenarios.

[0100] Specifically, the method for comparing and selecting the matched target detection algorithms in S5 comprises:

[0101] a: Obtain the main target detection category.

[0102] It should be noted that each target detection algorithm has multiple target detection categories, such as (aircraft, bird, car, and person), and the same target detection algorithm has different detection accuracies for different target detection categories.

[0103] b: According to the main target detection category, and combined with the preset algorithm evaluation index set, the average precision mean of each target detection algorithm and the average precision of the main target detection category are obtained.

[0104] Among them, the algorithm evaluation index set includes: the average precision mean of each target detection algorithm, the detection speed and the average precision of various target detection categories.

[0105] Specifically, in the field of target detection, the average precision is used to evaluate the detection accuracy of a target detection algorithm for a single target detection category, and the average precision mean is used to evaluate the comprehensive detection accuracy of the target detection algorithm, which is obtained by averaging the average precisions of all target detection categories in the data set.

[0106] It should be noted that the algorithm evaluation index of the target detection algorithm can be calculated and measured by the prior art after the algorithm training is completed, and the specific manner is not described here; the preset can be directly called in the use process, and real-time evaluation is not required.

[0107] c: According to the theoretical formula, the comprehensive precision of each target detection algorithm is calculated, and the theoretical formula includes:

[0108] P=X*AP+Y*mAP

[0109] In the formula, P represents the comprehensive precision, AP represents the average precision of the main target detection category, mAP represents the average precision mean, X represents the weight of the average precision, and Y represents the weight of the average precision mean.

[0110] In this embodiment, the weight X of the average precision is set to 0.6, and the weight Y of the average precision mean is set to 0.4, but it is not limited thereto, and other proportions can also be used, which can be adjusted according to actual conditions, and is not specifically limited.

[0111] d: In the each target detection algorithm, the target detection algorithm with the maximum comprehensive precision is selected.

[0112] It can be understood that the comprehensive precision not only focuses on the detection accuracy of the main target detection category, but also takes into account the detection accuracy of other target detection categories to some extent, which is used as a screening basis to select the target detection algorithm that best meets the actual use requirements.

[0113] e: Determine whether the target detection algorithm with the maximum comprehensive precision is unique.

[0114] If yes, the screened unique target detection algorithm is selected as the target detection algorithm to be selected;

[0115] If no, according to the preset algorithm evaluation index set, the target detection algorithm with the fastest detection speed is screened out as the target detection algorithm to be selected.

[0116] It can be understood that, by taking the detection speed as the screening basis, the target detection algorithm with the best efficiency can be screened out, so that the detection speed is also taken into account while ensuring the detection accuracy.

[0117] The present application can screen the target detection algorithm according to the algorithm evaluation index through the above steps, and the optimal selection that takes into account both the detection accuracy and the detection speed can be selected, so as to ensure that the target detection algorithm adopted is the most suitable selection at present, and the actual application effect is good.

[0118] Specifically, the method of comparing and screening the matched target detection algorithms in S5 comprises:

[0119] Step i: starting the obtained target detection algorithms in parallel, respectively detecting the video stream to obtain the corresponding detection results.

[0120] It should be noted that the video stream is derived from a high-definition array camera arranged at the airport.

[0121] Step ii: analyzing the detection results to obtain the number and coordinate information of the detected targets.

[0122] Step iii: obtaining the number and coordinate information of the targets detected by the radar.

[0123] Step iv: comparing the number and coordinate information of the detected targets with the number and coordinate information of the detected targets respectively to obtain the error rate of each target detection algorithm.

[0124] Step v: screening the target detection algorithm with the smallest error rate from the target detection algorithms as the target detection algorithm to be selected.

[0125] It should be noted that the above screening method takes the detection results of the target detection algorithm as the basis for judgment, which can ensure that the accuracy of the finally selected target detection algorithm is the highest, but it needs to obtain the information detected by the radar and occupies more resources, and the use is limited, which can be selectively used according to the actual situation.

[0126] Embodiment two:

[0127] The embodiment provides an airport high-definition array camera adaptive target detection algorithm selection method, which is different from the embodiment one in that the following steps are further included.

[0128] S6: binding the airport high-definition array camera with the target detection algorithm to be selected, establishing a binding relationship between the airport high-definition array camera and the target detection algorithm, collecting the binding relationship between each airport high-definition array camera and the target detection algorithm, and constructing an algorithm configuration table.

[0129] S7: starting the airport high-definition array camera, and performing target detection on the airport high-definition array camera according to the target detection algorithm bound in the algorithm configuration table, recording a target detection result, and constructing a target detection result list, wherein the target detection result includes a target category, a target coordinate, a target confidence and a processing speed FPS.

[0130] S8: obtaining airport comprehensive information, calculating pixel coordinate information of a real target in an image, comparing the pixel coordinate information of the real target in the image with the target detection result list, and constructing a target detection comparison error list.

[0131] It should be noted that the target detection comparison error list mainly records the target category and target coordinate of the missed detection and false detection of the airport high-definition array camera at the t time. The airport comprehensive information includes flight plans, radar information and flight control information.

[0132] Specifically, the pixel coordinate information of the target in the image calculated in S8 includes:

[0133] S81: obtaining the number of flights and the passenger flow according to the flight plans and the air control information in the airport comprehensive information.

[0134] S82: obtaining specific position information of the target according to the radar information in the airport comprehensive information.

[0135] S83: obtaining the pixel coordinate information of the target in the image by using a coordinate conversion formula according to the number of flights, the passenger flow and the specific position information of the target.

[0136] S9: determining whether the comparison error exceeds a preset threshold value according to the target detection comparison error list, and if yes, determining whether the number of migration training times is less than a preset threshold value, and if yes, performing migration training.

[0137] S10: retraining an original target detection algorithm model according to the target detection comparison error list, and obtaining a new target detection algorithm model.

[0138] S11: judging whether the new target detection algorithm model is better than the original target detection algorithm model, if yes, covering the original target detection algorithm model with the new target detection algorithm model, otherwise, applying manual intervention.

[0139] It can be understood that the manual intervention is that a staff member judges whether the algorithm needs to be replaced, and if yes, the algorithm configuration table is updated.

[0140] Embodiment three:

[0141] The embodiment provides an airport high-definition array camera adaptive target detection algorithm selection system, which comprises:

[0142] An information acquisition module is configured to acquire scene information, wherein the scene information comprises meteorological information, position information and day-night information.

[0143] A matching algorithm constructor is configured to generate a matching number according to the scene information and in combination with a matching number generation rule.

[0144] An algorithm pool is configured to obtain an algorithm serial number according to the matching number and in combination with a preset matching number-algorithm serial number association relationship.

[0145] An algorithm selection module is configured to obtain a matched target detection algorithm according to the algorithm serial number and in combination with a preset algorithm serial number-target detection algorithm association relationship.

[0146] An algorithm screening module is configured to judge whether the matched target detection algorithm is unique, if yes, taking the matched unique target detection algorithm as a target detection algorithm to be selected, and if no, comparing and screening the matched target detection algorithms to take an optimal target detection algorithm as the target detection algorithm to be selected.

[0147] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 An apparatus for implementing a function specified in one or more flows or one or more blocks in the flowcharts and / or block diagrams. Figure 1 An apparatus for implementing a function specified in one or more flows or one or more blocks in the flowcharts and / or block diagrams.

[0148] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The flow or flows and / or blocks Figure 1 The flow or flows and / or blocks

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 The flow or flows and / or blocks Figure 1 The flow or flows and / or blocks

[0150] The above description is only preferred embodiments of the present application, it should be pointed out that for those skilled in the art, without departing from the technical principles of the present application, a number of improvements and modifications can be made, these improvements and modifications should also be considered as the protection scope of the present application.

Claims

1. A method for selecting adaptive target detection algorithms for airport high-definition array cameras, characterized in that, include: Acquire scene information of the location of the airport's high-definition array camera, wherein the scene information includes: meteorological information, location information, and day / night information; Based on the scenario information and the matching number generation rules, a matching number is generated; Based on the matching number and the preset association between the matching number and the algorithm sequence number, the algorithm sequence number is obtained; Based on the algorithm sequence number and combined with the preset association between the algorithm sequence number and the target detection algorithm, the matched target detection algorithm is obtained. After obtaining the matched target detection algorithm based on the algorithm sequence number and the preset association between the algorithm sequence number and the target detection algorithm, the method further includes: Determine whether the matched target detection algorithm is unique; If so, the unique target detection algorithm that matches will be selected as the target detection algorithm. If not, the matched target detection algorithms are compared and selected, and the optimal target detection algorithm is selected as the target detection algorithm to be chosen. After determining whether the matched target detection algorithm is unique, the process also includes: Bind the airport HD array camera to the target detection algorithm to be selected, establish the binding relationship between the airport HD array camera and the target detection algorithm, summarize the binding relationship between each airport HD array camera and the target detection algorithm, and construct an algorithm configuration table; The airport HD array camera is started. The airport HD array camera performs target detection according to the target detection algorithm bound in the algorithm configuration table, records the target detection results, and constructs a target detection result list. The target detection results include: target category, target coordinates, target confidence, and processing speed (FPS). The process involves acquiring comprehensive airport information, calculating the pixel coordinates of the target in the image, comparing the pixel coordinates of the target in the image with the target detection result list, and constructing a target detection comparison error list. The comprehensive airport information includes flight plans, air traffic control information, and radar information. Based on the target detection comparison error list, determine whether the comparison error exceeds a preset threshold. If so, determine whether the number of transfer training attempts is less than a preset threshold, and then perform transfer training. Based on the target detection comparison error list, the original target detection algorithm model is retrained to obtain a new target detection algorithm model. Determine whether the new object detection algorithm model is superior to the original object detection algorithm model. If so, replace the original object detection algorithm model with the new one; otherwise, request manual intervention.

2. The method for selecting adaptive target detection algorithms for airport high-definition array cameras according to claim 1, wherein the method for comparing and filtering the matched target detection algorithms includes: Obtain the main target detection category; Based on the main target detection category and combined with the preset algorithm evaluation index set, the average accuracy of each target detection algorithm and the average accuracy in the main target detection category are obtained. The algorithm evaluation index set includes: the average accuracy of each target detection algorithm, detection speed, and average accuracy in various target detection categories. The overall accuracy of each target detection algorithm is calculated based on a theoretical formula, which includes: ， In the formula, P Indicates overall accuracy. AP This represents the average accuracy across the main target detection categories. mAP This represents the mean of the precision values. X The weights representing the average precision. Y Weights representing the mean of average precision; The target detection algorithm with the highest overall accuracy is selected from all the target detection algorithms. Determine whether the target detection algorithm with the highest overall accuracy is unique; If so, the unique target detection algorithm selected will be used as the target detection algorithm to be chosen; If not, then based on the preset algorithm evaluation index set, select the target detection algorithm with the fastest detection speed as the target detection algorithm to be selected.

3. The adaptive target detection algorithm selection method for airport high-definition array cameras according to claim 1, characterized in that, The matching number generation rules specifically include: The scene information is concatenated into a string and separated by commas; The string is converted into a fixed-length hash value using a hash function; Convert the hash value to an integer value; The integer value is mapped to the range of values ​​for the matching number to generate a matching number.

4. The adaptive target detection algorithm selection method for airport high-definition array cameras according to claim 1, characterized in that, The method for determining the association between the algorithm number and the target detection algorithm includes: Acquire various target detection algorithms used in real-world application scenarios; The various target detection algorithms are numbered to obtain the correlation between the algorithm number and the target detection algorithm.

5. The adaptive target detection algorithm selection method for airport high-definition array cameras according to claim 1, characterized in that, The method for determining the association between the matching number and the algorithm sequence number includes: By combining the various scenario information applicable to a certain target detection algorithm with the matching number generation rules, various matching numbers corresponding to that target detection algorithm are generated. Based on the association between algorithm serial numbers and target detection algorithms, establish the association between the algorithm serial number corresponding to the target detection algorithm and various matching numbers; Summarize the correlation between the algorithm number corresponding to each object detection algorithm and various matching numbers to determine the correlation between the matching number and the algorithm number.

6. The adaptive target detection algorithm selection method for airport high-definition array cameras according to claim 1, characterized in that, The pixel coordinates of the target in the image are calculated, including: The number of flights and passenger traffic are obtained from the flight plans and air traffic control information in the airport's integrated information system. The target's specific location information is obtained based on radar information from the airport's integrated information system. Based on the number of flights, passenger flow, and the specific location information of the target, the pixel coordinates of the target in the image are calculated using coordinate conversion formulas.

7. An adaptive target detection algorithm selection system for airport high-definition array cameras, characterized in that, include: Information acquisition module: used to acquire scene information, including: meteorological information, location information and day / night information; Matching algorithm builder: used to generate a matching number based on the scene information and the matching number generation rules; Algorithm pool: used to obtain the algorithm sequence number based on the matching number and the preset association between the matching number and the algorithm sequence number; Algorithm selection module: used to obtain the matched target detection algorithm based on the algorithm number and the preset association relationship between the algorithm number and the target detection algorithm; Algorithm filtering module: Used to determine whether the matched target detection algorithm is unique; If so, the unique target detection algorithm that matches will be selected as the target detection algorithm. If not, the matched target detection algorithms are compared and selected, and the optimal target detection algorithm is selected as the target detection algorithm to be chosen. Algorithm configuration table construction module: used to bind the airport HD array camera with the target detection algorithm to be selected, establish the binding relationship between the airport HD array camera and the target detection algorithm, summarize the binding relationship between each airport HD array camera and the target detection algorithm, and construct the algorithm configuration table; The target detection result list construction module is used to start the airport HD array camera. The airport HD array camera performs target detection according to the target detection algorithm bound in the algorithm configuration table, records the target detection results, and constructs a target detection result list. The target detection results include: target category, target coordinates, target confidence, and processing speed (FPS). The target detection comparison error list construction module is used to acquire comprehensive airport information, calculate the pixel coordinate information of the target in the image, compare the pixel coordinate information of the target in the image with the target detection result list, and construct the target detection comparison error list; wherein, the comprehensive airport information includes flight plan, air traffic control information and radar information; Transfer training judgment module: It is used to determine whether the comparison error exceeds the preset threshold according to the target detection comparison error list. If so, it determines whether the number of transfer training times is less than the preset threshold, and then performs transfer training. Target detection algorithm model acquisition module: used to retrain the original target detection algorithm model according to the target detection comparison error list to obtain a new target detection algorithm model; Object detection algorithm model update module: Used to determine whether the new object detection algorithm model is better than the original object detection algorithm model. If so, the new object detection algorithm model will replace the original object detection algorithm model; otherwise, manual intervention will be requested.

Images