Airport arrival and departure flight information accurate display system for serving passengers

Abstract

The invention discloses an airport arrival and departure flight information accurate display system for providing service for passengers. The airport arrival and departure flight information accurate display system has the effects of accurate guidance of arrival passenger luggage extraction and departure passenger flight information combined display. For accurate guidance of luggage extraction of passengers entering a port, a pedestrian re-identification technology is innovatively introduced in the design to capture flight passenger states and obtain accurate drop-off time of a target flight, and the video capture calculation cost of arrival of the passengers in a luggage turntable area is effectively reduced through a pre-arrival time prediction algorithm; and an airport luggage system is linked, so that decision support is provided for workers to reasonably arrange luggage carrying, and the time for passengers to wait for luggage is shortened. For a departure passenger flight information combination display module, the design obtains flight information of a plurality of passengers watching a combination screen through a face recognition technology, optimizes and combines the flight information and flight information under an original rule, and sends the flight information and the flight information to display screen terminals in groups according to a passenger distribution angle. And flight information concerned by passengers is highlighted to the greatest extent.

Description

Technical Field

[0001] The present invention relates to the field of airport flight information display, and in particular to an accurate display system for airport inbound and outbound flight information serving passengers. Background Art

[0002] First, currently, if an incoming passenger has checked baggage, the way to obtain baggage claim information after getting off the plane is generally to follow the airport guidance signs to the baggage claim area, and then find the baggage carousel number corresponding to the flight on the flight display screen to obtain the claim information. The above baggage claim guidance process has problems such as unclear instructions, incoherent instructions, and long waiting time for baggage claim.

[0003] Second, with the continuous development of the civil aviation industry, the scale of civil aviation airports is getting larger and larger, the number of airports with a passenger capacity of more than 10 million is increasing, and the number of flights carried out per unit time is also increasing. In order to facilitate the outbound flight information inquiry service provided by the airport for the convenience of passengers, a combination of multiple display screens is set up to display more flights.

[0004] However, after the combined screen is adopted, passengers have to face multiple flight display screens when boarding. Each display screen has many pages of flights, and each page of flights is displayed for a short time, and it is not easy to find the flight information you want to pay attention to. This situation is very unfavorable to the convenience of passengers' travel and the improvement of airport service quality.

[0005] The existing combined screen display mode has two unsolvable problems: 1. The number of flights displayed will continue to increase with the growth of airport passenger throughput, and the difficulty and time spent by passengers in searching will increase significantly. In extreme cases, passengers may not be able to find their flights at all.

[0006] 2. If the passenger's flight is outside the display time range set by the combination screen, it is impossible for the passenger to query the flight information.

[0007] For example, an airport with an annual passenger throughput of 30 million will display outbound flights within four hours. During peak hours, there will be about 80 flight information that needs to be displayed. In order to display this flight information quickly and efficiently, the current airport flight information display system usually uses a combination of multiple display screens.

[0008] If four display screens are used for combined display, each display screen will display approximately 20 flights in total. Flight information generally needs to be displayed in Chinese and English. When the Chinese and English information are displayed simultaneously, it is more appropriate to display about 7 flights per screen. Therefore, each screen needs to be turned twice. If the flight information is displayed for 10 seconds each time, then the passenger will spend up to nearly 30 seconds browsing on each display screen. If the Chinese and English information are displayed separately, the number of flights that can be displayed on each screen can be slightly increased, for example, to 10, but the total number of page turns increases to 3 times, which takes more time. Adding more display screens can reduce the total number of flights displayed on each screen, but it will also take passengers more time to find the display screen where the flight is located.

[0009] The above display mode is generally a fixed passive display mode, which can only display flight information within a fixed time range. With the continuous increase of airports with a throughput of more than 30 million, the above-mentioned problems will become more and more common and significant. Summary of the invention

[0010] Therefore, in order to solve the above-mentioned shortcomings, the present invention provides an airport inbound and outbound flight information accurate display system serving passengers, which has the functions of accurate guidance of luggage claiming for inbound passengers and combined display of outbound passenger flight information.

[0011] The present invention is implemented by constructing an accurate display system for airport inbound and outbound flight information serving passengers; comprising an accurate guidance unit for luggage claiming of inbound passengers and a combined display unit for flight information of outbound passengers; The precise guidance unit for baggage claiming of inbound passengers can obtain video streams from the airport video surveillance system interface, obtain flight information from the flight display system, and connect to the broadcasting system to control the zoning control of the inbound area; Obtain the video stream of the video surveillance at the bridge, extract the characteristic information of the passengers in the picture; associate the passenger characteristics with the flight information; and establish the passenger characteristic library group of the flight; If the flight has started to disembark passengers, the display screen on the staff side will display the start of disembarkation, and the staff will be reminded by voice. At the same time, the passenger estimated arrival time sorting model is called to calculate the estimated arrival time of the flight passengers at the baggage claim area, and sort the passenger feature library groups of each flight; Obtain the video stream of the video surveillance at the must-pass point to the baggage claim area, and extract the characteristic information of the passengers in the picture through the multi-passenger recognition algorithm based on deep learning; compare the passenger characteristics with the characteristic library grouping based on the priority comparison algorithm sorted by the expected arrival time. If the number of matches exceeds the threshold, it is determined that the passenger of the flight is about to arrive at the baggage claim area. The large screen at the entrance of the baggage claim area will display the baggage carousel number of the flight in real time, and control the broadcasting system to broadcast the baggage claim information of the flight in sections; The system displays on the staff side display that the flight passengers have arrived at the baggage claim area and reminds the staff through voice; The outbound passenger flight information combined display screen unit system includes a combined screen terminal and a camera. A high-definition camera is arranged in front of the combined screen terminal to obtain the image in front of the screen in real time. When the passenger is in front of the display screen, the corresponding camera is used to obtain the passenger's facial information and transmit it to the flight display system to obtain the flight information corresponding to all the faces. The system integrates these flight information with the flights screened under the original rules and sends them to the combined screen display terminal. The first combined screen gives priority to sending facial recognition flights and some flights screened by the original display rules, and the remaining flights are displayed in order on other combined screens.

[0012] Further, the multi-passenger identification algorithm based on deep learning is performed as follows: (1) Obtaining video stream: The video analysis service platform is connected to the video surveillance system through the streaming media server; (2) Preprocessing: Convert the streaming media into images through the image frame extraction algorithm. In order to process each part of the image separately, the image segmentation algorithm is used in this system to segment the image into background, face, and human body area outside the face; (3) Image optimization: To solve the problem of low imaging resolution of old cameras affecting passenger visual recognition analysis, the present invention is based on a model of cross-modal pedestrian re-identification, which includes an effective and lightweight grayscale enhancement exploration, a dual-stream network structure, and a feature-constrained all-modality center triplet loss. Grayscale enhancement exploration is used to enable the network to better mine modal shared information and improve its robustness to modal-specific information; by removing the color information of the visible light modality, the network can pay more attention to modal shared information. At the same time, in order to strengthen feature constraints at the feature level, the all-modality center-triplet (AMCT) loss is used, which can effectively reduce the impact of modal differences and appearance differences through feature metrics.

[0013] (4) Algorithm recognition analysis; Passenger recognition application scenario: given a monitored passenger image, match it with the passenger image across devices; divided into two steps: feature extraction and similarity measurement; 1) Feature extraction: The passenger status from the bridge area to the baggage claim area is relatively continuous. The camera shoots from a high altitude and as close as possible to extract the passenger information; 2) Similarity measurement: Based on the dual-branch Siamese network as a deep neural network structure, it is used for similarity comparison tasks; the basic idea of ​​the dual-branch Siamese network is to share weights and parameters by inputting two input samples into two identical sub-networks. Each branch is a convolutional neural network (CNN) feature extractor, which is used to extract meaningful feature representations from the input image; the two branches are: category branch (ID Branch) and attribute branch (AttributeBranch); the category branch (ID Branch) distinguishes: conventional features such as hairstyle and clothing color; the attribute branch distinguishes: whether it is a backpack, hand luggage, whether it is carrying children, etc. The two branches perform deep association aggregation to obtain the characteristic values ​​of the passenger in this row. Finally, the goal of passenger identification is achieved by matching the feature vectors.

[0014] Further, the priority comparison algorithm module based on the expected time sorting is performed as follows: The priority comparison algorithm module consists of the following components: (1) Establishment of prediction model: Based on the grid method, the terminal area road network is divided into equal-sized and non-intersecting grids. Each grid has its own number, and the original trajectory points are mapped to the grid to obtain the trajectory represented by the grid sequence. Then, the spatiotemporal association is learned at the granularity of the spatial grid to perform end-to-end travel arrival time prediction. The model is mainly based on Deeptravel and is divided into a feature extraction layer and a prediction layer. In the feature extraction layer, the model extracts the embedded representation of spatiotemporal features respectively; the prediction layer uses a two-layer long short-term memory network, and simultaneously learns the features of the grid sequence in the forward and reverse directions for travel time prediction; (2) Establish basic data: a database of paths from each bridge to the baggage claim point; (3) Prediction module: According to the corridor bridge number, the path trajectory data is imported, and the expected arrival time of the passengers with captured characteristics is predicted through the prediction model, and the passenger characteristics are sorted according to the order of expected arrival time; (4) Pedestrian re-identification, that is, when re-identifying the video at the baggage claim point, the current time will be compared with the expected arrival time, starting with the time closest to the time; (5) After the passenger is successfully matched, the actual arrival time is recorded and fed back to the prediction model, and the model is retrained.

[0015] The present invention has the following advantages: The present invention provides an accurate display system for airport inbound and outbound flight information serving passengers, which has the functions of accurate guidance for luggage collection of inbound passengers and combined display of flight information of outbound passengers. For accurate guidance for luggage collection of inbound passengers, the present design provides passenger status information for luggage handling personnel. Based on the above-mentioned innovative pedestrian re-identification technology and associated with the corresponding flights, accurate passenger drop-off time, estimated time for passengers to arrive at the luggage collection area, and reminder of passengers about to arrive at the luggage carousel can be provided to luggage handling personnel. Baggage handling personnel can reasonably arrange the luggage handling time according to this information and shorten the time passengers wait for their luggage. After the outbound flight display guidance adopts the above method, the number of flights displayed on the combined screen will be greatly reduced, and the problem of passengers failing to query flights due to the reduction of flights can be avoided. Therefore, this part has the following advantages: 1. The flight information that may have taken more than 1 minute to search before now takes no more than 10 seconds, and the improvement effect is significant, which improves the travel experience of passengers. 2. Flight information could not be found for flights outside the time range before, but now it can be queried regardless of the flight time. 3. Face recognition flights use a non-sensing interactive method, so passengers do not feel anything, and there will be no complaints due to failed queries or errors. 4. After the number of flights is greatly reduced, the combined display screen can enlarge the font of flight information, making it easier for passengers to watch from a distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the process of baggage claim guidance and baggage claim chain optimization for the inbound passengers in this application; Figure 2 This is a schematic diagram of the execution of the multi-passenger identification algorithm based on deep learning in this application; Figure 3 This is a schematic diagram of the execution of the priority comparison algorithm module based on the expected arrival time sorting of the present application; Figure 4 This is a schematic diagram of the implementation of the combined display screen unit for outbound passenger flight information of this application; Figure 5 It is a schematic diagram of the implementation comparison of the combined display screen unit for outbound passenger flight information of this application. DETAILED DESCRIPTION

[0017] The following will be combined with the attached Figure 1-Figure 5 The present invention is described in detail, and the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The present invention provides an accurate display system for airport inbound and outbound flight information for passengers, including an accurate guidance unit for luggage claiming of inbound passengers and a combined display unit for flight information of outbound passengers; the following describes these two parts in detail respectively; First, in order to satisfy the passengers' wonderful travel experience, the airport strives to create a refined, standardized and regular service process, always insists on taking passengers' needs as the service orientation, further optimizes the inbound baggage service chain, improves the efficiency of baggage retrieval, and allows "passengers to wait less and baggage to run faster". At present, after the technology was applied to a certain airport, the satisfaction of inbound passengers increased by 30%, and the baggage retrieval time was shortened by 30%.

[0019] The existing inbound baggage claim guidance system has problems such as unclear instructions, incoherent instructions, and guidance information. To address the above problems, the present invention uses visual recognition, estimated arrival time prediction, video analysis technology and other technologies to provide an intelligent system for inbound passengers to provide accurate baggage claim guidance services through multi-algorithm fusion. The specific advantages of the present invention are as follows: (1) Currently, inbound passengers follow the airport’s guidance signs to the baggage claim area, and then use the flight display screen to find the baggage carousel number corresponding to their flight to obtain the baggage claim information. The baggage claim area has unclear and incoherent instructions.

[0020] This application connects multiple systems such as the airport video surveillance system, flight management system, broadcasting system, and baggage management system. By using the existing bridge cameras and without adding new cameras, advanced visual recognition technology is used to capture passenger feature information through multi-dimensional recognition of passengers' facial features, clothing types, colors, behavior patterns, etc., and the backend system associates the corresponding flight of the passenger. On the route that passengers must take to reach the baggage claim area, the baggage carousel number of the passenger's flight is prominently displayed to passengers passing through the area through real-time comparison of passenger feature information, and the baggage carousel number of the passenger's flight is broadcast in the area in conjunction with the zone broadcast, which not only realizes the senseless guidance of baggage claim, but also realizes the location tracking and management of passengers.

[0021] (2) Pedestrian re-identification requires high computing resources, large scenes such as terminal buildings require large investments, and recognition efficiency is low. This design innovatively introduces a predicted arrival time sorting model, divides passengers into different groups according to flights, and sorts them by predicting the arrival time at the baggage claim area. When passengers enter the baggage claim area, they are prioritized to the groups with the highest sorting. Once any passenger in this group is successfully matched, the baggage claim information of the group's associated flights is accurately displayed, achieving a recognition effect in seconds. The present invention effectively improves the matching efficiency and directly improves the actual response speed of the system and the user experience.

[0022] (3) Baggage handling personnel are unable to perceive the status of passengers, which directly affects the baggage handling plan and the time passengers have to wait for their luggage.

[0023] This design provides baggage handlers with passenger status information. Based on the above innovative pedestrian re-identification technology and associated with the corresponding flight, it can provide baggage handlers with accurate drop-off time, estimated arrival time of passengers at the baggage claim area, and reminders of passengers arriving at the baggage carousel. Baggage handlers can reasonably arrange baggage handling time based on this information and shorten the waiting time for passengers.

[0024] See also Figure 1 , which shows the method flow of baggage claim guidance for inbound passengers and optimization of the baggage claim chain, to achieve accurate baggage claim guidance to the greatest extent possible and shorten the waiting time of passengers at the baggage carousel.

[0025] Combined with Figure 1 Explain the workflow of this unit: (1) Build a baggage claim sensorless guidance system (this system), obtain video streams from the video surveillance system interface, obtain flight information from the flight display system, and connect to the broadcasting system to control the arrival area zoning control.

[0026] (2) Obtain the video stream from the video surveillance at the boarding bridge, and extract the characteristic information of the passengers in the image through a variety of video analysis algorithms (clothing, body shape, and behavioral characteristics); associate the passenger characteristics with the flight information; and establish a passenger characteristic library group for the flight.

[0027] (3) When the video analysis determines that the flight has started to disembark, the staff side display screen will display the start of disembarkation of the flight, and the staff will be reminded by voice; at the same time, the passenger estimated arrival time sorting model is called to calculate the estimated arrival time of the flight passengers at the baggage claim area, and the passenger feature library groups of each flight are sorted.

[0028] (4) Obtain the video stream of the video surveillance at the point where you must pass to the baggage claim area, and extract the characteristic information of the passengers in the picture through the multi-passenger recognition algorithm based on deep learning; compare the passenger characteristics with the characteristic library grouping based on the priority comparison algorithm sorted by the expected arrival time. If the number of matches exceeds the threshold, it is determined that the passenger of the flight is about to arrive at the baggage claim area. The large screen at the entrance of the baggage claim area will display the baggage carousel number of the flight in real time, and control the broadcast partition to broadcast the flight claim information; (5) This unit displays on the staff side display that the passenger of the flight has arrived at the baggage claim area and reminds the staff through voice.

[0029] like Figure 1 The specific implementation process includes: Step 1: Capture the characteristic information of outbound passengers through the bridge camera; Step 2: The background associates the characteristic information in step 1 with the flight and establishes the passenger characteristic group of the flight; Step 3: When the number of passengers identified by the bridge video exceeds the threshold, it is determined that the flight has started to disembark; the display screen on the staff side shows that the flight has started to disembark, and there is a voice reminder; Step 4: At the point where passengers must pass before arriving at the baggage claim area, the camera extracts passenger information and compares it with the feature library. The comparison algorithm innovatively introduces a priority comparison algorithm based on the expected arrival time. The matching number threshold can be set, and the data is derived from the airport experience; Step 5: If multiple matches are found in the passenger feature database group of a certain flight, it means that multiple passengers of the flight are about to arrive at the baggage claim area. The large screen at the entrance of the baggage claim area will display the baggage carousel number of the flight in real time; and cooperate with the zone broadcast to broadcast the baggage carousel number corresponding to the flight number.

[0030] Among them, Figure 2 As shown, the deep learning-based multi-passenger identification algorithm involved is executed as follows: (1) Obtaining video stream: The video analysis service platform is connected to the video surveillance system through the streaming media server; (2) Preprocessing: Convert the streaming media into images through the image frame extraction algorithm. In order to process each part of the image separately, the image segmentation algorithm is used in this system to segment the image into background, face, and human body area outside the face; (3) Image optimization: To solve the problem of low imaging resolution of old cameras affecting passenger visual recognition analysis, the present invention is based on a model of cross-modal pedestrian re-identification, including an effective lightweight grayscale enhancement exploration, a dual-stream network structure, and a feature-constrained all-modality center triplet loss; grayscale enhancement exploration is used to enable the network to better mine modal shared information and improve its robustness to modal-specific information; by removing the color information of the visible light modality, the network can pay more attention to modal shared information. At the same time, in order to strengthen feature constraints at the feature level, the all-modality center-triplet (AMCT) loss is used, which can effectively reduce the impact of modal differences and appearance differences through feature metrics.

[0031] (4) Algorithm recognition analysis; Passenger recognition application scenario: given a monitored passenger image, match it with the passenger image across devices; divided into two steps: feature extraction and similarity measurement; 1) Feature extraction: The passenger status from the bridge area to the baggage claim area is relatively continuous. The cameras shoot from a high altitude and as close as possible to extract features from passenger information.

[0032] 2) Similarity measurement: Based on the dual-branch Siamese network, which is a deep neural network structure, it is used for similarity comparison tasks. The basic idea of ​​the dual-branch Siamese network is to share weights and parameters by inputting two input samples into two identical sub-networks. Each branch is a convolutional neural network (CNN) feature extractor, which is used to extract meaningful feature representations from the input image. The two branches are: category branch (ID Branch) and attribute branch (AttributeBranch); the category branch (ID Branch) distinguishes: conventional features such as hairstyle and clothing color; the attribute branch distinguishes: whether it is a backpack, hand luggage, whether it is carrying children, etc. The two branches perform deep association aggregation to obtain the characteristic values ​​of the passenger in this row; finally, by matching the feature vectors, the goal of passenger identification is achieved.

[0033] Among them, Figure 3 As shown, the priority comparison algorithm module involved based on the expected arrival time sorting is executed as follows: The problem of high computing resource requirements for pedestrian re-identification, large-scale scenes such as terminal buildings require large investments, and low recognition efficiency is addressed. This design innovatively introduces a predicted arrival time sorting model, divides passengers into different groups according to flights, and sorts them by predicting the arrival time at the baggage claim area. When passengers enter the baggage claim area, they are given priority to the groups with higher rankings. Once any passenger in this group is successfully matched, the baggage claim information of the group's associated flights is accurately displayed, achieving recognition results in seconds. The present invention effectively improves the matching efficiency and directly improves the actual response speed of the system and user experience. The priority matching algorithm is based on the predicted arrival time sorting model to sort the passenger feature groups of each flight, and prioritizes the comparison of passenger feature groups of flights whose current time is close to the predicted time, effectively avoiding the catastrophic problem of insufficient computing resources caused by full comparison. The priority matching algorithm module based on predicted arrival time sorting consists of the following components: (1) Prediction model establishment: The grid-based method divides the terminal area road network into equal-sized and non-intersecting grids. Each grid has its own number, and the original trajectory points are mapped to the grid to obtain a trajectory represented by a grid sequence. Then, the spatiotemporal association is learned at the granularity of the spatial grid to perform end-to-end travel arrival time prediction. The model is mainly based on Deeptravel and is divided into a feature extraction layer and a prediction layer. In the feature extraction layer, the model extracts the embedded representation of spatiotemporal features respectively. The prediction layer uses a two-layer long short-term memory network and learns the features of the grid sequence in the forward and reverse directions for travel time prediction.

[0034] (2) Establish basic data: a database of the paths from each bridge to the baggage claim point.

[0035] (3) Prediction module: According to the corridor bridge number, the path trajectory data is imported, and the estimated arrival time of the passengers with captured characteristics is predicted through the prediction model, and the passenger characteristics are sorted according to the order of estimated arrival time.

[0036] (4) Pedestrian re-identification, that is, when re-identifying a person via video at the baggage claim point, the current time will be compared with the expected arrival time, starting with the time closest to the time.

[0037] (5) After the passenger is successfully matched, the actual arrival time is recorded and fed back to the prediction model, and the model is retrained.

[0038] At present, after trial operation, the technology has increased the satisfaction of incoming passengers by 30% and shortened the baggage retrieval time by 30%.

[0039] The following is a detailed description of the current situation of the combined display screen unit for outbound passenger flight information; this unit aims to address the two problems existing in the current airport outbound flight display guidance, improves the flight content sending rules of the combined screen, uses active face recognition technology to obtain the passenger flight that is viewing the combined screen, and then optimizes and merges the flights to be displayed under the original rules, and finally groups them according to the combined screen order and sends them to each combined display terminal. Figure 4 and Figure 5 Come and see; The main methods implemented in this section are as follows: 1. Optimize terminal display and add the earliest and latest scheduled departure time of the flight to the combination screen. When the flight display system background sends flight data to the combination screen terminal, it sorts them according to the scheduled departure time. Each combination screen display terminal will have the earliest and latest scheduled departure time of the flight. The terminal will then display the earliest and latest flight time of the clear flight. Passengers can locate the corresponding display terminal according to the flight time, thereby reducing the search time.

[0040] 2. Bind the combined screen to the boarding gate and set the boarding gate number associated with the combined display screen. When screening flights, the flight display system background only sends flights that board at several nearby boarding gates to the combined display screen. This operation will greatly reduce the total number of flights displayed, but it will cause flights at other boarding gates to not be displayed, so it needs to be combined with the following face recognition associated flight technology.

[0041] 3. Face recognition is associated with flights, and the flight information of identified passengers is displayed first. A high-definition camera is set in front of the combination screen terminal to obtain the image in front of the screen in real time, and the mature face recognition technology is used to capture the face information of several people in front. Then, by using the correspondence between the face and flight information left by the passenger during security check, the system can obtain the flight information of the passenger in front of the combination screen. The system integrates this flight information with the flights screened under the original rules and sends it to the combination screen display terminal. The first combination screen gives priority to sending face recognition flights and some flights screened by the original display rules, and the remaining flights are displayed in order on other combination screens.

[0042] After the above method is adopted for the display guidance of outbound flights, the number of flights displayed on the combination screen will be greatly reduced, and the problem of passengers being unable to check flights due to the reduction of flights can be avoided. Therefore, this part has the following advantages: 1. It used to take more than a minute to search for flight information, but now it takes no more than 10 seconds. The improvement is significant and has improved the travel experience of passengers.

[0043] 2. Previously, it was impossible to find flight information for flights outside the time range. Now, you can query the flight information at any time.

[0044] 3. Facial recognition flights use a seamless interaction method, so passengers are unaware of it and there are no complaints due to failed queries or errors.

[0045] 4. After the number of flights is greatly reduced, the combined display screen can enlarge the font of flight information, making it easier for passengers to watch from a distance.

[0046] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An accurate display system of airport arrival and departure flight information for passengers, characterized by: It includes an accurate guidance unit for baggage claiming for inbound passengers and a combined display unit for flight information for outbound passengers; The inbound passenger baggage claim precision guidance unit can obtain video streams from the airport video surveillance system interface, obtain flight information from the flight display system, and connect to the broadcast system for zone broadcasting; Obtain the video stream of the video surveillance at the bridge, and extract the characteristic information of passengers in the picture through the multi-passenger recognition algorithm based on deep learning; Associating the passenger characteristics with flight information; Establish the passenger feature database group of the flight; If the flight has started to disembark passengers, the display screen on the staff side will display the start of disembarkation, and the staff will be reminded by voice. At the same time, the passenger estimated arrival time sorting model is called to calculate the estimated arrival time of the flight passengers at the baggage claim area, and sort the passenger feature library groups of each flight; Obtain the video stream of the video surveillance at the must-go point to the baggage claim area and extract the characteristic information of the passengers in the picture; The priority comparison algorithm based on the estimated arrival time sorting compares the passenger's features with the feature library grouping. If the number of matches exceeds the threshold, it is determined that the passenger of this flight is about to arrive at the baggage claim area. The large screen at the entrance of the baggage claim area displays the baggage carousel number of this flight in real time, and the broadcasting system is linked to broadcast the baggage claim information of this flight in different zones. The system displays on the staff side display that the flight passengers have arrived at the baggage claim area and reminds the staff through voice; The outbound passenger flight information combined display screen unit system includes a combined screen terminal and a camera. A high-definition camera is arranged in front of the combined screen terminal to obtain the image in front of the screen in real time. When the passenger is in front of the display screen, the corresponding camera is used to obtain the passenger's facial information and transmit it to the flight display system to obtain the flight information corresponding to all the faces. The system integrates these flight information with the flights screened under the original rules and sends them to the combined screen display terminal. The first combined screen gives priority to sending facial recognition flights and some flights screened by the original display rules, and the remaining flights are displayed in order on other combined screens.

2. According to claim 1, a precise display system for airport arrival and departure flight information serving passengers, characterized in that; The deep learning-based multi-passenger identification algorithm is performed as follows: (1) Obtaining video stream: The video analysis service platform is connected to the video surveillance system through the streaming media server; (2) Preprocessing: Convert streaming media into images through image frame extraction algorithm; In order to process each image part separately, the image segmentation algorithm is used in this system to segment the image into background, face, and human body area outside the face. (3) Image optimization: To solve the problem of the low resolution of the old camera imaging affecting the visual recognition analysis of passengers, the present invention is based on a cross-modal pedestrian re-identification model, which includes an effective lightweight grayscale enhancement exploration, a dual-stream network structure, and a feature-constrained full-modal center triplet loss. Grayscale enhancement exploration is used to enable the network to better mine modal shared information and improve its robustness to modal-specific information. The color information of the visible light modality is removed so that the network can pay more attention to modal shared information. At the same time, in order to strengthen feature constraints at the feature level, the full-modal center triplet loss is used, which can effectively reduce the impact of modal differences and appearance differences through feature metrics. (4) Algorithm recognition analysis; Passenger recognition application scenario: given a monitored passenger image, match it with the passenger image across devices; divided into two steps: feature extraction and similarity measurement; 4.1) Feature extraction: The passenger status from the bridge area to the baggage claim area is relatively continuous. The camera shoots from a high altitude and as close as possible to extract the passenger information; 4.2) Similarity measurement: Based on the dual-branch Siamese network, which is a deep neural network structure, it is used for similarity comparison tasks. The basic idea of ​​the dual-branch Siamese network is to share weights and parameters by inputting two input samples into two identical sub-networks. Each branch is a convolutional neural network (CNN) feature extractor, which is used to extract meaningful feature representations from the input image. The two branches are: category branch (ID Branch) and attribute branch. The category branch (ID Branch) distinguishes conventional features such as hairstyle and clothing color. The attribute branch is used to distinguish whether the passenger is carrying a backpack, hand luggage, or children. The two branches perform deep correlation aggregation to obtain the characteristic values ​​of the passenger in the row. Finally, the goal of passenger identification is achieved by matching the characteristic vectors.

3. The accurate display system of airport arrival and departure flight information for passengers according to claim 1, characterized in that; The priority comparison algorithm module based on the expected arrival time sorting is performed as follows: The priority comparison algorithm module consists of the following components: (1) Prediction model establishment: The terminal area road network is divided into equal-sized and non-intersecting grids based on a grid-based method. Each grid is numbered and the original trajectory points are mapped to the grid to obtain a trajectory represented by a grid sequence. Then, the spatiotemporal association is learned at the granularity of the spatial grid to perform end-to-end travel arrival time prediction. The model is mainly based on Deeptravel and is divided into a feature extraction layer and a prediction layer. In the feature extraction layer, the model extracts the embedded representation of spatiotemporal features respectively. The prediction layer uses a two-layer long short-term memory network and simultaneously learns the features of the grid sequence in the forward and reverse directions for travel time prediction. (2) Establish basic data: a database of paths from each bridge to the baggage claim point; (3) Prediction module: According to the corridor bridge number, the path trajectory data is imported, and the expected arrival time of the passengers with captured characteristics is predicted through the prediction model, and the passenger characteristics are sorted according to the order of expected arrival time; (4) Pedestrian re-identification, that is, when re-identifying the video at the baggage claim point, the current time will be compared with the expected arrival time, starting with the time closest to the time; (5) After the passenger is successfully matched, the actual arrival time is recorded and fed back to the prediction model, and the model is retrained.

Images