CN-121457999-B - Airport resource intelligent scheduling method based on multi-source perception and flight linkage

CN121457999BCN 121457999 BCN121457999 BCN 121457999BCN-121457999-B

Abstract

The invention discloses an airport resource intelligent scheduling method based on multi-source perception and flight linkage, which comprises the steps of obtaining multi-source perception data of an airport operation area, wherein the multi-source perception data at least comprise multi-view video data reflecting the state of passengers, obtaining individual emotion labels and emotion recognition confidence corresponding to the individual emotion labels through a multi-view emotion recognition model based on anonymous feature vectors and preliminary emotion probability distribution, obtaining flight operation data containing flight dynamic and event severity parameters in real time, binding passenger tracks with flights according to the space-time distribution of the anonymous feature vectors and the flight operation data to generate a plurality of flight-crowd clusters, and carrying out weighted aggregation on each flight-crowd cluster by combining the emotion labels of the individuals contained in the flight-crowd clusters and the emotion recognition confidence of the individual emotion labels to obtain an aggregate emotion index of the flight-crowd clusters. The method and the system improve airport service response efficiency and resource scheduling accuracy and reduce public security risks.

Inventors

GE QIANG
FANG LIANG
ZHANG TAO
ZHENG HONGFENG

Assignees

飞友科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260105

Claims (8)

1. An airport resource intelligent scheduling method based on multi-source perception and flight linkage is characterized by comprising the following steps: S1, acquiring a multi-view video stream of an airport public area, and performing real-time edge calculation on the video stream to obtain anonymous feature vectors and preliminary emotion probability distribution representing a passenger individual; s2, obtaining an individual emotion label and emotion recognition confidence corresponding to the individual emotion label through a multi-view emotion recognition model based on the anonymous feature vector and the preliminary emotion probability distribution; s3, acquiring flight operation data comprising flight dynamics and event severity parameters in real time, wherein the event severity parameters are composite indexes for quantifying the influence degree of abnormal events of flights; S4, binding passenger tracks and flights according to space-time distribution of anonymous feature vectors and flight operation data to generate a plurality of flight-crowd clusters; S5, aiming at each flight-crowd cluster, carrying out weighted aggregation by combining the emotion labels of the individuals contained in the flight-crowd clusters and the emotion recognition confidence level of the individual to obtain an aggregate emotion index of the flight-crowd cluster; s6, aiming at each flight-crowd cluster, dynamically calculating a dedicated emotion situation threshold value through a preset event kernel function according to event severity parameters of the bound flights and the remaining time from planned take-off; S7, comparing the aggregate emotion indexes with the exclusive emotion situation threshold, generating a hierarchical scheduling instruction aiming at the flight-crowd cluster when the aggregate emotion indexes exceed the exclusive emotion situation threshold, and scheduling airport resources according to the hierarchical scheduling instruction; Performing real-time edge calculation on the video stream to obtain anonymous feature vectors and preliminary emotion probability distribution representing the individual passengers, wherein the method specifically comprises the following steps of: operating a lightweight neural network model in an edge computing node deployed on the video acquisition equipment side; Carrying out face detection and upper body key point positioning on an input video stream frame by frame; Synchronously executing two processes on each detected passenger target in a memory, wherein the two processes are that irreversible anonymous feature vectors are extracted through the lightweight neural network model based on facial and upper body region images, and preliminary emotion probability distribution containing multiple emotion classification probabilities is output through an emotion classification head of the lightweight neural network model based on the same input; After the processing is finished, immediately discarding the original video frame and the intermediate image data, and only transmitting anonymous feature vectors and preliminary emotion probability distribution, wherein the lightweight neural network model is MobileFaceNet models, and the anonymous feature vectors are 512-dimensional real vectors subjected to L2 normalization; the step S2 specifically includes: Inputting anonymous feature vectors from at least two different perspectives of the same passenger into a multi-perspective emotion recognition model, and performing feature alignment and fusion through a time-space attention fusion layer to generate unified emotion feature representation; The unified emotion feature representation and the preliminary emotion probability distribution of the corresponding visual angle set are input into a double-branch structure of a multi-visual angle emotion recognition model together to be processed, wherein fine-grained expression features are extracted from facial micro-expression branches under the adjustment of regional visibility gating, and gesture dynamic features are extracted from upper body gesture time sequence branches through a time sequence convolution network; The method comprises the steps of carrying out fusion and classification processing on high-order features output by two branches, and outputting an individual emotion recognition result which comprises probability information of a plurality of emotion categories; And determining the emotion category with the highest probability as an individual emotion label, and quantifying the highest probability value as emotion recognition confidence.
2. The intelligent airport resource scheduling method based on multi-source awareness linked flights according to claim 1, wherein the event severity parameter is calculated based on at least one or more of scheduled delay or cancellation time of flights, gate change times, flight cancellation flags, or delay reason codes including baggage problems and mechanical failures.
3. The airport resource intelligent scheduling method based on multi-source awareness and flight linkage according to claim 1, wherein step S4 specifically comprises: constructing a space-time track of each anonymous feature vector on an airport map; calculating the association probability of each space-time track with each candidate flight based on the check-in cut-off time, the flight planning time and the boarding gate position information in the flight operation data; And dividing all the space-time tracks according to the associated probability of the space-time tracks and each flight by adopting a probability soft clustering algorithm, and dynamically dividing a track set which belongs to the same flight and has high space-time aggregation into a flight-crowd cluster, wherein the probability soft clustering algorithm is a Gaussian mixture model algorithm.
4. An airport resource intelligent scheduling method based on multi-source awareness and flight linkage according to claim 3, wherein said calculating the probability of association with each candidate flight for each space-time trajectory comprises: Calculating time matching degree based on the time stamp of the space-time track passing through the security check area and the check-in deadline of the candidate flight, wherein the smaller the time difference is, the higher the matching degree is; Calculating spatial proximity based on the final stay position of the space-time track in the waiting area and the boarding gate position of the candidate flight, wherein the smaller the physical distance is, the higher the proximity is; and carrying out weighted fusion on the time matching degree and the space proximity degree, and calculating to obtain the association probability of the space-time track and each candidate flight.
5. The airport resource intelligent scheduling method based on multi-source awareness and flight linkage according to claim 1, wherein step S5 specifically comprises: Obtaining emotion labels of all individuals forming the flight-crowd cluster and emotion recognition confidence degrees corresponding to the emotion labels, wherein each emotion label corresponds to one emotion classification probability distribution; weighting the emotion classification probability distribution of each individual according to a preset negative emotion weight vector, and calculating to obtain the negative emotion contribution degree of the individual; and carrying out weighted summation on the negative emotion contribution degrees of all the individuals according to the emotion recognition confidence degrees corresponding to the individuals, and taking the summation result as an aggregate emotion index of the flight-crowd cluster.
6. The airport resource intelligent scheduling method based on multi-source awareness and flight linkage according to claim 1, wherein step S6 specifically comprises: acquiring an emotion baseline threshold value preset for the flight-crowd cluster; acquiring real-time event severity parameters of flights bound by the flight-crowd cluster and the remaining time from planned take-off; Calculating according to the event severity parameter and the residual time from planned take-off through a preset event kernel function to obtain an event influence factor; And dynamically adjusting the emotion baseline threshold according to the event influence factors to obtain the exclusive emotion situation threshold of the flight-crowd cluster.
7. The intelligent airport resource scheduling method based on multi-source awareness and flight linkage of claim 6, wherein the preset event kernel function is Dynamically adjusting the emotion baseline threshold according to the event influence factors to obtain the exclusive emotion situation threshold of the flight-crowd cluster, wherein the method specifically comprises the following steps: ; Wherein, the Is a specific emotion situation threshold; is an emotion baseline threshold; Is a policy parameter; Remaining time for planned take-off; is an event severity parameter.
8. The intelligent airport resource scheduling method based on multi-source awareness and flight linkage according to claim 1, wherein the airport resource scheduling method according to the hierarchical scheduling instruction specifically comprises: Analyzing the hierarchical scheduling instruction, and acquiring the designated scheduling level and the target flight-crowd cluster identification; Determining at least one type of execution resources and specific actions corresponding to the scheduling level according to a preset level-resource mapping strategy, wherein the execution resources comprise an airport broadcasting system, an information display screen, a mobile service terminal, customer service personnel or security personnel; sending a control command or a task instruction to the determined execution resource, wherein the control command or the task instruction comprises the position information of a target flight-crowd cluster, the flight information and the specific action required by the target flight-crowd cluster; and receiving and recording the feedback state of the execution resource aiming at the control command or the task instruction so as to confirm that the scheduling action is executed.

Description

Airport resource intelligent scheduling method based on multi-source perception and flight linkage Technical Field The invention relates to the technical field of computer vision and intelligent operation and maintenance, in particular to an airport resource intelligent scheduling method based on multi-source perception and flight linkage. Background With the continuous increase of civil aviation traffic, abnormal flight events of airports easily cause negative emotion aggregation of passengers in a waiting area, and potential public safety and service risks are formed. At present, airport management mainly relies on passive manual inspection and unidirectional broadcasting, and has the problems of delayed response and inaccurate scheduling. Meanwhile, a business system (such as an AOC) and a security monitoring system are mutually independent to form a data island, so that a manager cannot grasp the key situation of where and how a specific flight passenger is in real time. In the technical aspect, emotion recognition research based on computer vision has progressed, but when the method is applied to such complex public scenes of airports, the method still faces significant challenges, firstly, the existing method is insufficient in robustness of wearing actual conditions such as masks, side faces and illumination changes, recognition performance is unstable, secondly, most of the technology is focused on single emotion classification, the capability of modeling the space-time evolution dynamics of crowd emotion is lacking, group emotion aggregation is difficult to early warn, finally, the method is most critical, the existing scheme cannot realize deep and real-time intelligent coupling with business logic such as flight operation, and early warning threshold cannot be dynamically adjusted according to event severity (such as delay time), so that early warning mechanism is stiff, and closed-loop management from accurate perception to active scheduling intervention is difficult to support. Therefore, when the current airport is dealing with the risk of emotion driving, a series of problems of incomplete perception, difficult situation research and judgment, lack of basis for decision making and inaccurate linkage exist. A comprehensive technical scheme capable of integrating robust emotion perception, group situation modeling, business data linkage and self-adaptive decision is needed to improve the security active defense and operation guarantee capability of an airport. Disclosure of Invention In order to solve the technical problems in the background technology, the invention provides an airport resource intelligent scheduling method based on multi-source perception and flight linkage. The invention provides an airport resource intelligent scheduling method based on multi-source perception and flight linkage, which comprises the following steps: S1, acquiring multi-source perception data of an airport operation area, wherein the multi-source perception data at least comprises multi-view video data reflecting the state of a passenger; s2, obtaining an individual emotion label and emotion recognition confidence corresponding to the individual emotion label through a multi-view emotion recognition model based on the anonymous feature vector and the preliminary emotion probability distribution; s3, acquiring flight operation data comprising flight dynamics and event severity parameters in real time, wherein the event severity parameters are composite indexes for quantifying the influence degree of abnormal events of flights; S4, binding passenger tracks and flights according to space-time distribution of anonymous feature vectors and flight operation data to generate a plurality of flight-crowd clusters; S5, aiming at each flight-crowd cluster, carrying out weighted aggregation by combining the emotion labels of the individuals contained in the flight-crowd clusters and the emotion recognition confidence level of the individual to obtain an aggregate emotion index of the flight-crowd cluster; s6, aiming at each flight-crowd cluster, dynamically calculating a dedicated emotion situation threshold value through a preset event kernel function according to event severity parameters of the bound flights and the remaining time from planned take-off; And S7, comparing the aggregate emotion indexes with the exclusive emotion situation threshold, generating a hierarchical scheduling instruction aiming at the flight-crowd cluster when the aggregate emotion indexes exceed the exclusive emotion situation threshold, and scheduling airport resources according to the hierarchical scheduling instruction. Preferably, the calculating the real-time edge of the video stream to obtain anonymous feature vectors and preliminary emotion probability distribution representing the individual passenger specifically includes: operating a lightweight neural network model in an edge computing node deployed on the video acquisition equipm