Airport adaptive climate target detection system and design method thereof

Abstract

The invention discloses a method for designing an airport climate-adaptive target detection system, and aims to solve the problem that an airport monitoring system is poor in target detection effect under atrocious weather conditions. According to the system, input video streams are analyzed in real time through a weather classification algorithm, real-time weather conditions such as sunny days, foggy days and rainy days can be recognized, real-time weather forecast information is obtained through an API, and the real-time weather category of an airport is obtained through a weather classification algorithm module; according to the weather category, in combination with a preset image enhancement algorithm, selecting a matched algorithm to process the image; inputting the enhanced image into a target detection algorithm, and carrying out the recognition and detection of a target object in the airport; effective target data of image detection and data of the millimeter wave radar and the laser radar are subjected to fusion processing, and finally a target detection result is output. According to the method, an adaptive climate image enhancement mode is adopted, the airport scene target detection effect under various climate conditions is improved, and the airport governance efficiency and the troubleshooting capability of air traffic control potential safety hazards are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and in particular to an airport climate-adaptive target detection system and a design method thereof. Background Art

[0002] With the continuous development of the global economy and the gradual improvement of people's living standards, airports have become an indispensable key infrastructure in the process of modern urbanization, and their importance to cities is increasing. In the daily operation of airports, meteorological conditions play a vital role in ensuring flight safety, improving flight efficiency, and optimizing flight duty. At present, under good climatic conditions, the existing airport monitoring system can achieve a relatively stable target detection effect. However, severe weather conditions, such as fog, rain or snow, greatly interfere with the performance of traditional target detection algorithms, resulting in a significant decrease in the accuracy and efficiency of target detection.

[0003] Traditional airport monitoring systems mainly rely on sensors such as cameras and radars for target detection and tracking. However, these systems often find it difficult to maintain effective target detection capabilities when encountering weather changes, such as fluctuations in ambient light, decreased visibility, and increased noise. Therefore, developing a target detection system that can adapt to various climatic conditions can not only significantly improve the monitoring performance of airports in different climates, but also effectively overcome the limitations of existing technologies under extreme weather conditions. This is an important research direction in the current development of airport monitoring technology. Summary of the invention

[0004] To solve the above problems, the present invention provides an airport climate adaptive target detection system and a design method thereof, which solve the problems of low detection accuracy and poor detection quality of current airport target detection systems under complex meteorological conditions.

[0005] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions:

[0006] The system design of the present invention includes three stages:

[0007] In the first stage, the present invention provides a method for distinguishing airport surface weather, comprising:

[0008] Get the video frames captured by the high-resolution array camera in the airport scene;

[0009] Perform image weather identification on each acquired video frame image of the camera;

[0010] Obtain the weather type of each frame image according to the weather discrimination model;

[0011] Get weather information provided by external API;

[0012] Based on the image weather discrimination results obtained by the model and the API meteorological information, a weighted voting algorithm is used to obtain the weather category of the current airport scene. The preset weather categories are sunny, foggy, rainy, and snowy.

[0013] Specific weighted voting algorithm:

[0014] First, determine the weights: assign weights to each camera based on its historical accuracy data, device quality, location, and other factors. The weight data weights = [weight1, weight2, ..., weight n ] Assign weights to API weather information api .

[0015] Collect results: If there are n cameras in the airport, each camera outputs a weather category. The output result array of the camera is camera outputs =[T f1 ,T f2 ,……,T fn ], for API weather category there is api output .

[0016] Calculate weighted votes: Calculate the weighted votes for various weather categories based on the data collected by the mobile phone.

[0017] Select the final result: Select the weather category with the highest weighted votes as the final result of the current identified weather category.

[0018] In the second stage, the system will determine the corresponding image processing algorithm serial number based on the weather category identified in the first stage and the mapping relationship between the preset weather category and the image processing algorithm serial number;

[0019] According to the corresponding image processing algorithm serial number and in combination with the correspondence between the preset algorithm serial number and the specific image processing algorithm, an image processing algorithm suitable for complex meteorological conditions is selected from the algorithm pool.

[0020] Specifically, the system will adopt the following processing strategies according to different weather categories:

[0021] If the current weather category is sunny, select the corresponding sunny image processing algorithm in the algorithm pool;

[0022] In particular, for sunny days, the ambient light is sufficient, the image quality is high, and the object detection algorithm can achieve the best results on the original image, so no special image processing is performed. Therefore, the image processing algorithm corresponding to sunny days in the algorithm pool should directly return the original image without additional image preprocessing.

[0023] If the current weather category is foggy, the real-time defogging model will be called to defog the image.

[0024] Specifically, the system uses advanced defogging algorithms, such as defogging methods based on dark color prior or atmospheric scattering models, to process the original image to eliminate image blur and contrast reduction caused by fog. The defogging image can significantly improve the accuracy of target detection.

[0025] If the current weather category is rain, the system will call the rain removal model to remove rain streaks from the image. The system uses deep learning technologies such as convolutional neural networks to identify and remove raindrops and raindrops from the image. Rain removal can eliminate the interference of rain on the image, restore the clarity and integrity of the image, and provide high-quality image data for subsequent target detection.

[0026] If the current weather category is snow, the system will call the snow removal model to remove snow from the image. In snowy weather, snowflakes will block the camera's view, resulting in reduced image quality. The system uses a special snow removal algorithm to remove snow from the image, improve image clarity, and ensure that the target detection algorithm can still work properly under snowy conditions.

[0027] Further, after selecting and applying a specific image processing algorithm, the system further includes the following steps:

[0028] Effect evaluation: The system evaluates the effect of the selected image processing algorithm and determines the adaptability and efficiency of the algorithm under current weather conditions by comparing it with historical data. Evaluation indicators include processing speed, image quality improvement, and target detection accuracy.

[0029] Algorithm optimization: If the evaluation results show that the current algorithm is not sufficient to cope with extreme climate conditions, the system will start the algorithm adjustment mechanism. By calling the backup algorithm in the algorithm pool or adjusting the algorithm parameters, a more suitable image processing method for the current conditions is selected to ensure the optimization of system performance.

[0030] Configuration record: The system builds an algorithm configuration table, which records in detail the image processing algorithm and its performance indicators selected under each weather condition. This provides an important reference for future algorithm optimization and adjustment, and helps to continuously improve the adaptability and reliability of the system.

[0031] In the third stage of system design, the processed images obtained in the second stage are used for target detection and data fusion. The key steps of this stage include:

[0032] Image-based object detection: The system applies advanced object detection algorithms to the processed images output by the second stage. The algorithm can be based on a deep learning framework such as a convolutional neural network (CNN) or a regional convolutional neural network (R-CNN) to accurately identify and locate objects in the image. The output of this step is the pixel coordinates of the object in the image coordinate system and its bounding box information.

[0033] Get LiDAR point cloud data: At the same time, the system obtains relevant target point cloud data from the LiDAR sensor. LiDAR can provide high-precision spatial measurement and generate a three-dimensional point cloud that reflects the spatial position and shape characteristics of objects within the sensor's field of view.

[0034] Mapping of radar space and image coordinates: In order to achieve data fusion, the system establishes the mapping relationship between the radar space coordinate system and the image coordinate system according to the preset internal and external parameter calibration. Through the external parameter matrix (describing the rotation and translation between sensors) and the internal parameter matrix (describing the focal length and optical center position of the camera), the three-dimensional point cloud data of the lidar is accurately projected onto the two-dimensional image plane to generate a point cloud image corresponding to the image.

[0035] Post-fusion processing: After obtaining the target detection results in the image coordinate system and the projected point cloud data, the system performs post-fusion processing to improve the accuracy and robustness of target detection. The specific steps of post-fusion processing include:

[0036] Target association: Use spatial position and feature information to associate the targets detected in the image with the targets in the point cloud data. Methods such as multi-target tracking algorithms, Kalman filtering, or joint probabilistic data association (JPDA) can be used to ensure the accuracy of target matching.

[0037] Information fusion: For the associated target, the data from the image and point cloud are fused. The image provides rich texture and color information, while the point cloud provides accurate depth and shape information. The fusion process can use weighted averaging, Bayesian estimation, or deep fusion network to integrate multi-source data into a more comprehensive target description.

[0038] Optimization of detection results: Through fusion, the confidence of target detection is improved, and false detection and missed detection caused by the limitations of a single sensor are eliminated. Especially under complex meteorological conditions, the fused results can effectively resist noise and interference, and improve the reliability of detection.

[0039] Output fusion results: Finally, the system outputs the fused target detection results, including the target's location, size, category, confidence level, etc. This information can be used for subsequent airport security monitoring, navigation, decision support and other applications.

[0040] Through the above method, the system can adaptively adjust the image processing strategy according to real-time weather changes, ensuring that a high level of target detection performance can be maintained under various complex climate conditions. This adaptive capability is crucial for airport security monitoring and can effectively reduce safety hazards caused by weather factors.

[0041] An airport climate-adaptive target detection system comprising:

[0042] Information acquisition unit: used to obtain real-time airport scene pictures and meteorological factor information obtained by API;

[0043] Weather classification unit: used to obtain the real-time weather category of the airport according to the meteorological factor information and in combination with the weather classification algorithm;

[0044] Image enhancement unit: used for selecting a matching image enhancement algorithm to process the image according to the weather category and in combination with a preset image enhancement algorithm;

[0045] Target detection unit: used to input the enhanced image into the target detection algorithm to identify and detect target objects in the airport.

[0046] Data fusion unit: used to fuse the valid target data output by the target detection algorithm with the data of the millimeter wave radar and / or lidar;

[0047] Result output unit: used to output the enhanced and accurate target detection results.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. Adaptive image processing: The present invention improves the accuracy and reliability of target detection by identifying weather types in real time and automatically selecting the image processing algorithm that best suits the current climate conditions.

[0050] 2. Rapid response mechanism: The multi-stage processing strategy allows the system to quickly adjust the algorithm when the weather changes, ensuring optimal performance under extreme weather conditions and enhancing the adaptability of the system.

[0051] 3. Data-driven optimization: By building an algorithm configuration table and continuous effect evaluation, the system can be optimized based on historical data, providing a solid foundation for future algorithm upgrades and adjustments.

[0052] 4. Multi-source data fusion improves detection performance: By fusing image and lidar data, the system overcomes the limitations of a single sensor, especially in complex meteorological conditions, and improves the accuracy and reliability of target detection.

[0053] 5. The system adopts a modular design, and each functional module is relatively independent, which is convenient for subsequent functional expansion and technical upgrades, such as the introduction of new sensor types or advanced algorithm models. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a system diagram of the present invention;

[0055] Figure 2 is a flow chart of an airport climate adaptive target detection system provided by Embodiment 1 of the present invention;

[0056] Figure 3 This is a flow chart of an airport climate adaptive target detection system provided by Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0057] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0058] Example 1

[0059] like Figure 1 As shown, an embodiment of the present invention provides a climate-adaptive target detection method in an airport scenario, which is applied in an airport scenario and includes the following steps:

[0060] S1. Obtain the videos captured by each surveillance camera in the airport scene.

[0061] The image is the video captured by each surveillance camera, which contains continuous frame images.

[0062] S2. Perform a first-stage weather classification on the acquired video frames to obtain the weather categories identified by each sub-camera, and perform a weighted vote with the weather API information results to obtain the weather category in the current scene. For weather classification, since weather changes are not immediate, considering the load of the system server, interval weather recognition is adopted.

[0063] Specifically, for the n cameras at the airport, the n cameras are numbered in the order of 1-n. If the time when the system starts running is t1, the weather discrimination module will obtain the video frame images of all cameras at time t1, and store the video frame images in the video frame list figures, where figures = [f1, f2...f3]. The discrimination model first performs weather discrimination on f1 and obtains the weather category T of the frame image f1 at time t1. f1, then perform weather identification on the frame image f2 of the sub-camera numbered 2, and obtain the weather category T of the frame image f2 at time t1 f2 After all sub-cameras have been identified, the sub-camera weather category array T at time t1 is obtained = [T f1 ,T f2 ,……,T fn ], get the weather category T of the API information at time t1 api For any T, its actual value should be one of the following values: Sunny, Rain, Foggy, Snow. According to the T value at time t1, according to the system preset sub-camera weighting W = [w1, w2, ... w n ], the weight of api information w api .

[0064] A weighted voting algorithm is used to calculate the weather category at time t1.

[0065] Specific voting process:

[0066] Initialize the scorer: Create a score counter for each possible weather category (, initialize the score of each weather category to zero, i.e. S(T) = 0 for all T∈{Sunny,Rain,Foggy,Snow}.

[0067] Next, update the score according to the weather category and weight of each camera: S(T f1 )=S(T fi )+w i

[0068] At the same time, considering the weather category T provided by the external API api and its weight w api , further update the score: S(T api )=S(T api )+w api

[0069] Finally, by comparing the scores of all weather categories, the weather category with the highest score T is selected max As the final weather judgment result: T max =argmax_{T∈{Sunny,Rain,Foggy,Snow}}S(T)

[0070] S3, the final weather category T obtained according to S2 max , adjust the candidate algorithm of the current system image processing module. In particular, if the current T max= Sunny. The image processing module does not process the image frames of the video stream, but directly passes the image data to the target detection module. If the current identified weather type is not Sunny, the image processing module will max To call the relevant candidate algorithms in the algorithm pool.

[0071] In case 1, the algorithms in the algorithm pool include the defogging algorithm, corresponding to the weather tag T = Foggy; the rain streak removal algorithm, corresponding to the weather tag T = Rain; and the snowflake removal algorithm, corresponding to the weather tag T = Snow. max The specific value of is used to match the corresponding algorithm label. When the match is successful, the algorithm under the corresponding label will be called.

[0072] Only when T max The system will adjust the corresponding candidate algorithm only when the value of changes.

[0073] The observer mode is used to switch the algorithms in the algorithm pool, and the observed object is T in S2. max , the observer is the image processing module in S3, using the push method, if at t1 time T max The value of T maxt1 , T at the next weather determination time t2 max The value of T maxt2 , only when T maxt1 ≠T maxt2 , will the new T max The value is passed to the observer role image processing module. At this time, the image processing module is based on the new T max Candidate algorithms corresponding to value switching.

[0074] The specific steps of image processing module to process images are as follows:

[0075] A: For the image processing module, there should be two inputs. Input 1 is the video frame sequence of the airport camera, and input 2 should be the weather discrimination type T obtained by S2 voting weighting. max .

[0076] B: According to T max The determined weather category activates the corresponding candidate algorithm. The candidate algorithm first converts the image data in the video frame into a numerical image array matrix, and then processes the image to obtain a processed image array matrix.

[0077] (Iterative optimization, comparison with fixed scene images, and comparison based on subsequent target detection data)

[0078] C: Directly pass the image array matrix to the object detection module in the next stage.

[0079] S4: The target detection module receives the image array matrix output by the image processing module as input parameters

[0080] S5: Radar point cloud processing, the radar system independently monitors the target position and processes data;

[0081] Acquisition of data source: Millimeter wave radar data acquisition: obtain the distance, speed and angle information of the target; LiDAR point cloud acquisition: obtain three-dimensional point cloud data of the environment.

[0082] Radar point cloud data preprocessing: Use voxel filtering or statistical filtering to remove noise points and outliers in the point cloud data; downsampling: downsample the point cloud data to reduce the amount of calculation and increase the processing speed.

[0083] Target recognition and labeling: Clustering algorithm: Use algorithms such as DBSCAN or Euclidean Cluster Extraction to segment targets in point cloud data; Feature extraction: Calculate the centroid, size and shape features of each cluster; Target tracking: Use Kalman filter or multi-target tracking algorithm to track and estimate the state of the target.

[0084] Radar coordinate conversion: According to the installation position and attitude of the radar, the target position in the radar coordinate system is converted to the world coordinate system.

[0085] S6. Data alignment: post-fuse the target detection results and radar point cloud data.

[0086] Data alignment is divided into time synchronization and space alignment.

[0087] Time synchronization: Adjust the timestamps of radar and camera data, using linear interpolation or a synchronization signal to ensure that both are on the same time base.

[0088] Spatial alignment: Based on the sensor’s installation parameters, a coordinate system transformation matrix is ​​established to achieve spatial alignment of radar and camera data.

[0089] Coordinate transformation matrix: where r ij is the element of the rotation matrix, t x ,t y ,t z is the translation vector.

[0090] S7. Data Association

[0091] The nearest neighbor algorithm Hungarian algorithm or the method based on joint probabilistic data association (JPDA) is used to match the targets detected by the camera and radar.

[0092] Association criterion: Use Mahalanobis distance to measure the similarity between targets:

[0093]

[0094] Where S is the covariance matrix.

[0095] S8. Result fusion strategy

[0096] Using Kalman filter fusion

[0097] State definition: set the state vector of the target Contains position and velocity information.

[0098] Status prediction: F is the state transfer matrix, and Q is the process noise covariance matrix.

[0099] Measurement Update: x k|k =x k|k-1 +K k (z k -Hx k|k-1 ), P k|k =(IK k H)P k|k-1 , where H is the observation matrix, which maps the state vector to the measurement space. R is the covariance matrix of the measurement noise. k is the fused observation, containing measurements from the camera and the radar.

[0100] Fused Observations: For the same target, the measurements from the camera and radar are stacked:

[0101] S9, Decision-level Fusion

[0102] Rule setting: Set fusion rules based on business needs. For example, if the camera detects a target but the radar does not, it may be a long-distance target and the confidence level can be lowered.

[0103] Bayesian Fusion: Update the confidence of target existence using Bayesian probability.

[0104] S10. False detection and missed detection processing

[0105] False detection filtering: Utilize the complementarity of the two sensors to eliminate false detection results of a single sensor.

[0106] Missed detection compensation: If one sensor does not detect a target but another sensor does, a decision is made based on the confidence level and historical trajectory whether to add the target.

[0107] Embodiment 2:

[0108] In Example 2, the present invention further optimizes the climate-adaptive target detection method in the airport scenario, and adds a dynamic evaluation mechanism for the image defogging effect and detection results. The evaluation mechanism can automatically switch the image processing algorithm in the algorithm pool or adjust the parameters of the relevant algorithm according to the results of the real-time evaluation to adapt to more complex meteorological conditions and improve the accuracy and robustness of target detection. Specifically, the following steps are included:

[0109] S1. Video data acquisition

[0110] Like Example 1, Example 2 also obtains video streams through multiple surveillance cameras in an airport scenario. The system obtains the videos currently captured by each surveillance camera, which contain continuous frame image data, providing a basis for subsequent processing.

[0111] S2. Weather category identification

[0112] The system still uses a weighted voting algorithm to combine the weather category judgment results of multiple cameras and external API meteorological data to obtain the main weather category T in the current airport scene. max Since weather changes do not appear immediately, the system reduces the load pressure on the server by means of interval recognition. The specific weighted voting method is consistent with that in Example 1.

[0113] S3. Image processing algorithm selection

[0114] T obtained from S2 max , the image processing module selects the corresponding image processing algorithm from the algorithm pool. max If it is a sunny day, no image preprocessing is performed and the image is directly passed to the target detection module. max In foggy, rainy or snowy days, the system will call the corresponding defogging, deraining or desnowing algorithms to pre-process the image and remove environmental interference to improve image clarity and contrast.

[0115] S4. Image processing effect evaluation mechanism

[0116] In Example 2, in order to ensure that the image processing algorithm can achieve the expected effect under complex climate conditions, the system adds an image processing effect evaluation mechanism. The mechanism mainly includes the following evaluation steps:

[0117] Image quality assessment: The system evaluates the quality of the processed image, including contrast, clarity, noise level and other indicators. Commonly used evaluation indicators include peak signal-to-noise ratio (PSNR), structural similarity (SSIM) and contrast measurement. For example, PSNR is used to measure the denoising effect of an image, while SSIM is used to evaluate the overall clarity of the image. For defogging images, the system can calculate the changes in SSIM and PSNR of the image before and after defogging to ensure that the quality of the processed image is significantly improved.

[0118] Object detection accuracy evaluation: The system performs object detection on the processed image and evaluates based on the accuracy and error of the detection results. The detection results include the bounding box position, size, category, and confidence of the object. The system evaluates the effectiveness of the current image processing algorithm by calculating and comparing with historical detection data. If the accuracy of object detection drops significantly, the system will mark the current algorithm as ineffective.

[0119] Real-time feedback mechanism: The system sets a set of threshold standards. If the image quality evaluation index or target detection accuracy is lower than the set threshold, the evaluation mechanism will trigger feedback in real time. The system automatically switches to other candidate algorithms in the algorithm pool based on the feedback information, or optimizes the processing effect by adjusting the parameters of the existing algorithm (such as the contrast gain of the defogging algorithm, the depth of the convolution layer of the deraining algorithm, etc.).

[0120] S5. Dynamic algorithm adjustment

[0121] When the image processing effect evaluation mechanism detects that the current algorithm is not effective, the system will trigger dynamic algorithm adjustment. The dynamic adjustment mechanism includes the following:

[0122] Algorithm switching: The system will automatically select a new image processing algorithm from the algorithm pool to adapt to the current climate conditions. For example, if the current defogging algorithm is based on the dark color prior model and the evaluation effect is not good, the system will switch to a defogging algorithm based on the atmospheric scattering model to try to improve image quality.

[0123] Parameter optimization: For the image processing algorithm currently in use, the system automatically adjusts the algorithm parameters. For example, in the dehazing algorithm, the contrast enhancement coefficient can be dynamically adjusted, or the number of filter layers and step size of the convolutional neural network can be adjusted in the deraining algorithm. The system continuously optimizes the algorithm parameters based on the evaluation feedback information to achieve the best processing effect.

[0124] Intelligent learning mechanism: The system records the results of each evaluation and forms a data model based on historical records. By comparing the processing effects under different meteorological conditions, the system gradually optimizes the algorithm pool and parameter configuration to provide data support for future changes in meteorological conditions.

[0125] S6. Target Detection and Data Fusion

[0126] In Example 2, the image after image processing effect evaluation and algorithm optimization is passed to the target detection module. The target detection module uses a deep learning algorithm to detect the target in the image and generate the pixel coordinates and bounding box information of the target. At the same time, the system obtains the three-dimensional point cloud data of the target from the laser radar (LiDAR) sensor, establishes the mapping relationship between the radar and the image coordinates, and performs post-fusion processing.

[0127] Target association: The system uses Kalman filtering or joint probabilistic data association (JPDA) methods to associate the targets detected in the image with the targets in the point cloud data. The association is based on the spatial position and feature information of the target to ensure the accuracy of target matching.

[0128] Information fusion: By fusing image and point cloud data, the system can describe the target more comprehensively and improve the accuracy and robustness of detection. Fusion methods include weighted averaging and Bayesian estimation to eliminate false detection and missed detection by a single sensor.

[0129] S7. Effect evaluation and dynamic optimization

[0130] In Example 2, the system further combines the target detection results with the evaluation mechanism to establish a continuous evaluation mechanism for the target detection effect. The specific evaluation contents include:

[0131] Target detection accuracy: Evaluate the accuracy of the current algorithm by comparing the detection results with historical detection records. If the false detection rate or missed detection rate is higher than the threshold, the system will trigger dynamic optimization.

[0132] Robustness of detection results: Under complex meteorological conditions, the system checks the stability and consistency of detection results to ensure the adaptability of the algorithm in severe weather.

[0133] Confidence of fused data: The system combines the camera and radar data to calculate the confidence of the fusion result. If the confidence is low, the system will adjust the weight distribution of data fusion to ensure that the final output target detection result is reliable.

[0134] Result output and recording;

[0135] Finally, the system outputs the target detection results after dynamic evaluation and optimization, including the location, size, category and confidence of the target. The system stores the processing algorithm, evaluation results and optimization records under each weather condition in the algorithm configuration table. Through long-term data accumulation, the system can continuously improve the algorithm library and improve future adaptability and response speed.

[0136] Beneficial effects of embodiment 2

[0137] Compared with Example 1, Example 2 has improvements in the following aspects:

[0138] Adaptive optimization capability: Through real-time evaluation and feedback mechanism, the system can automatically select the best algorithm, or adjust the algorithm parameters to ensure the target detection effect under complex meteorological conditions. In foggy weather (horizontal visibility is 200-500 meters), the detection rate of Example 1 drops to 70%, while Example 2 maintains above 85% by dynamically adjusting parameters.

[0139] Improve detection accuracy and robustness: Under severe weather conditions, the system's dynamic optimization capability improves detection accuracy and reduces false detections and missed detections caused by the inadequacy of a single algorithm. In the same test environment, the false detection rate of Example 1 averaged 10% and the missed detection rate was 15%, while Example 2 reduced these two indicators to 3% and 5%, respectively.

[0140] Continuously improving intelligent learning mechanism: The system continuously optimizes and expands the algorithm pool through the accumulation and analysis of evaluation data, providing data support for future algorithm upgrades.

[0141] Scalability of modular design: The system adopts a modular design. The independence of each module ensures that the system can adapt to more types of sensors or algorithm models, which is conducive to future technology updates and expansions.

[0142] Embodiment 1 is mainly to meet basic needs, suitable for target detection tasks under standard meteorological conditions, and can better guarantee the basic monitoring needs in the airport environment. Embodiment 2: Based on Embodiment 1, feedback and adaptive adjustment capabilities are added, and the algorithm can be dynamically optimized, which is more suitable for high-precision target detection needs under complex and changeable meteorological conditions, and the system requirements are more complex and flexible.

[0143] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above technical features.

Claims

1. A design method for an airport climate-adaptive target detection system, characterized in that: The specific steps include: Step 1: Obtain the real-time scene image of the airport and the meteorological factor information obtained by API; Step 2: According to the meteorological factor information and in combination with the weather classification algorithm, the real-time weather category of the airport is obtained; The weather API information is weighted and voted together to get the weather category of the current scene. As the weather changes are not immediate, the weather is identified at intervals in consideration of the system server load. Step 3: According to the weather category obtained in step 2 and in combination with the preset image enhancement algorithm, a matching image enhancement algorithm is selected to process the image; Step 4: Input the enhanced image into the target detection algorithm to identify and detect target objects in the airport.

2. The method for designing an airport climate-adaptive target detection system according to claim 1, characterized in that: The weather classification algorithm in step 2 includes: identifying the current weather conditions based on the brightness, contrast and clarity parameters in the real-time scene image; combining with the preset weather classification model; specifically, for the n cameras in the airport, the n cameras are numbered in the order of 1 to n. If the time when the system starts running is t1, the weather discrimination module will obtain the video frame images of all cameras at time t1, and store the video frame images in the video frame list figures, where figures = [f1, f2...f3]. The discrimination model first performs weather discrimination on f1 to obtain the weather category T of the frame image f1 at time t1. f1 Then, the weather classification is performed on the frame image f2 of the sub-camera numbered 2 to obtain the weather category T of the frame image f2 at time t1. f2 ; After all sub-cameras are identified, the sub-camera weather category array T at time t1 is obtained = [T f1 , T f2 , ..., T fn ], get the weather category T of the API information at time t1 api For any T, its actual value should be one of the following values: Sunny, Rain, Foggy, Snow. According to the T value at time t1, according to the system preset sub-camera weighting W = [w1, w2, ... w n ], the weight of api information w api ; A weighted voting algorithm is used to calculate the weather category at time t1. Specific voting process: Initialize the scorer: For each possible weather category, create a score counter and initialize the score of each weather category to zero, that is, S(T) = 0 for all T∈{Sunny, Rain, Foggy, Snow}. , Update the score according to the weather category and weight of each camera: S(T f1 )=S(T fi )+w i At the same time, considering the weather category T provided by the external API api and its weight w api , further update the score: S(T api )=S(T api )+w api Finally, by comparing the scores of all weather categories, the weather category with the highest score T is selected max As the final weather judgment result: T max =argmax_{T∈{Sunny, Rain, Foggy, Snow}}S(T).

3. The method for designing an airport climate-adaptive target detection system according to claim 1, characterized in that: The method for selecting the image enhancement algorithm in step 3 includes: searching a preset correspondence table between weather categories and image enhancement algorithms according to the weather category; obtaining an image enhancement algorithm suitable for the current weather conditions from the correspondence table; and selecting the final weather category T obtained according to step S2. max , adjust the candidate algorithm of the current system image processing module; if the current T max = Sunny. The image processing module does not process the image frames of the video stream, but directly passes the image data to the target detection module. If the current identified weather type is not Sunny, the image processing module will max To call the relevant candidate algorithms in the algorithm pool.

4. The method for designing an airport climate-adaptive target detection system according to claim 3, characterized in that: The image enhancement processing includes but is not limited to contrast enhancement, color correction, blur removal and noise reduction processing, and the algorithms of the weather image processing module include gUNet, SFNet, MixDehazeNet and DDPM.

5. The method for designing an airport climate-adaptive target detection system according to claim 1, characterized in that: The target detection algorithm includes but is not limited to a convolutional neural network (CNN) model based on deep learning, YOLO, Faster R-CNN or other target detection algorithms suitable for the airport environment.

6. The method for designing an airport climate-adaptive target detection system according to claim 1, characterized in that: It also includes fusing the effective target data output by the target detection algorithm with the data of the millimeter wave radar and / or lidar; based on the results of the fusion processing, outputting the enhanced and accurate target detection results.

7. The method for designing an airport climate-adaptive target detection system according to claim 6, characterized in that: The data fusion processing includes: synchronizing the target detection result with the radar data in time; Matching is performed on spatial coordinates to improve the reliability of detection results.

8. The method for designing an airport climate-adaptive target detection system according to claim 1, characterized in that: Also includes: According to the target detection results, the airport surface targets are classified and counted; Record abnormal targets and the time and location of their appearance.

9. An airport climate-adaptive target detection system, characterized in that: include: Information acquisition unit: used to obtain real-time airport scene pictures and meteorological factor information obtained by API; Weather classification unit: used to obtain the real-time weather category of the airport according to the meteorological factor information and in combination with the weather classification algorithm; Image enhancement unit: used for selecting a matching image enhancement algorithm to process the image according to the weather category and in combination with a preset image enhancement algorithm; Target detection unit: used to input the enhanced image into the target detection algorithm to identify and detect target objects in the airport.

10. The airport climate-adaptive target detection system according to claim 9, characterized in that: Also includes: Data fusion unit: used to fuse the valid target data output by the target detection algorithm with the data of the millimeter wave radar and / or lidar; Result output unit: used to output the enhanced and accurate target detection results.

Images