Airport luggage cross-domain tracking re-identification method, device and system with time constraint

Abstract

The invention discloses an airport luggage cross-domain tracking re-identification method, device and system with time constraint, and a prime number method comprises the steps: S1, collecting luggage image information, carrying out the ID numbering of the image information, and uploading the image information to a server for storage; s2, carrying out luggage state detection on the image information by utilizing a luggage detection model, if the detection result is that the luggage state is normal, entering the next step, otherwise, sending out an alarm prompt that the luggage state is abnormal; s3, judging whether luggage retrieval is carried out or not, if yes, judging whether the luggage retrieval mode is online retrieval or not, and if yes, carrying out the step S4 to start online retrieval. The retrieval efficiency can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method, device and system for cross-domain tracking and re-identification of airport luggage with time constraints. Background Art

[0002] The full-process tracking of checked baggage is an important measure for airports to improve the quality of passenger services, the level of ground service support and service efficiency. The RFID (Radio Frequency Identification) technology currently used in airport checked baggage tracking is based on node supervision, that is, in the entire system, the places where RFID identification devices are deployed constitute the identification and tracking nodes of the system. The baggage tracking and supervision between these nodes is still carried out manually, that is, when abnormal baggage conditions (such as baggage falling, detention, damaged suitcases, etc.) occur at any location between the nodes, they cannot be discovered in time, and staff need to spend a lot of time to investigate afterwards.

[0003] At present, the cameras deployed on the airport baggage transportation routes have basically achieved full coverage of the transportation routes. Therefore, the use of machine vision technology to track the baggage throughout the entire process can quickly locate anomalies in the baggage transportation process and further improve staff efficiency.

[0004] To locate and track the luggage during the transportation process, it is necessary to complete the target detection and positioning tracking of the luggage. Among them, since multiple cameras are used to collect images of the luggage transportation route, the key algorithm for positioning and tracking is the image re-recognition algorithm. Image re-recognition is mainly based on the characteristics of the object to be identified, the model for feature extraction is designed, the feature vector of the object to be identified is established, and then the images collected by different cameras are matched and retrieved. However, in the entire luggage transportation line, due to the parameters, position, lighting and other reasons of the camera, it is very difficult to retrieve and match the luggage images collected by different cameras.

[0005] At the same time, since there are a large number of cameras on the automatic baggage transportation line, when using the re-recognition method for image matching, if the images collected by all cameras in the entire system are used as retrieval objects, the system baggage retrieval efficiency will be very low. Summary of the invention

[0006] In order to solve at least one of the above problems, the present invention provides a method, device and system for cross-domain tracking and re-identification of airport baggage with time constraints, which can effectively improve the retrieval efficiency.

[0007] The technical solution adopted by the present invention is:

[0008] In a first aspect, the present application discloses a method for cross-domain tracking and re-identification of airport baggage with time constraints, comprising:

[0009] Collecting luggage image information, assigning ID numbers to the image information, and uploading it to the server for storage;

[0010] Using the baggage detection model to detect the baggage status of the image information, if the detection result is that the baggage status is normal, proceed to the next step, otherwise, an alarm prompt of abnormal baggage status is issued;

[0011] Determine whether to perform baggage retrieval, and if so, determine whether the baggage retrieval method is online retrieval, and if so, perform online retrieval;

[0012] The online search includes:

[0013] Determine the starting search position;

[0014] Define a search step length of a single search to lock the search area of ​​the single search, wherein the search step length of the single search includes at most R cameras, R<15;

[0015] Performing a first search, determining a first search area according to a defined search step, and the first search area includes the starting search position;

[0016] Call the re-identification model to extract features of the target baggage and the first collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the first collected image comes from at most R cameras in the determined first search area;

[0017] If the match fails, move a defined search step from the first search area to the entrance and perform a secondary search as the secondary search area. If the secondary search fails, move a defined search step to the entrance again, and continue searching towards the entrance until the match is successful.

[0018] As an optional technical solution, if the detection result of the luggage status detection is that the luggage status is abnormal, the image of the abnormal luggage is cropped, rotated and the color is changed.

[0019] As an optional technical solution, the starting search position is s 0 Indicates that 0 Determined according to the following formula:

[0020] s 0 =(t 2 -t 1 )×v

[0021] Among them, t1 is the time when the baggage to be retrieved enters the entrance of the baggage sorting system, t 2 is the current retrieval time, and v is the conveying speed of the baggage sorting system.

[0022] As an optional technical solution, the search method further includes offline search, and the offline search includes:

[0023] Start searching from the entrance according to the defined search step length;

[0024] Call the re-identification model to extract features of the target baggage and the second collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the second collected image comes from at most R cameras within a search step starting from the entrance;

[0025] If the match fails, move one search step away from the entrance and search again, and move the search towards the entrance until the match is successful.

[0026] As an optional technical solution, the re-identification model adopts an architecture that combines a Swin Transformer feature extractor and an FPN feature fusion module, wherein the output of each layer of the Swin Transformer feature extractor serves as the input of the FPN feature fusion module.

[0027] In a second aspect, the present application discloses a device for cross-domain tracking and re-identification of airport luggage with time constraints, comprising:

[0028] The acquisition module is used to collect luggage image information, and upload the image information to the server for storage after assigning ID numbers to the image information;

[0029] A baggage detection module, used to detect the baggage status of the image information using a baggage detection model, and if the detection result shows that the baggage status is normal, the judgment module is called; otherwise, an alarm prompt indicating that the baggage status is abnormal is issued;

[0030] A judgment module is used to determine whether to perform baggage retrieval. If so, it determines whether the baggage retrieval method is online retrieval. If so, the online retrieval module is called;

[0031] The online search module is used for:

[0032] Determine the starting search position;

[0033] Define a search step length of a single search to lock the search area of ​​the single search, wherein the search step length of the single search includes at most R cameras, R<15;

[0034] Performing a first search, determining a first search area according to a defined search step, and the first search area includes the starting search position;

[0035] Call the re-identification model to extract features of the target baggage and the first collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the first collected image comes from at most R cameras in the determined first search area;

[0036] If the match fails, move a defined search step from the first search area to the entrance and perform a secondary search as the secondary search area. If the secondary search fails, move a defined search step to the entrance again, and continue searching towards the entrance until the match is successful.

[0037] As an optional technical solution, the luggage detection module is further used to: when the detection result of the luggage status detection is that the luggage status is abnormal, perform cropping, rotation and color conversion operations on the image of the abnormal luggage.

[0038] As an optional technical solution, the starting search position is s 0 Indicates that 0 Determined according to the following formula:

[0039] s 0 =(t 2 -t 1 )×v

[0040] Among them, t 1 is the time when the baggage to be retrieved enters the entrance of the baggage sorting system, t 2 is the current retrieval time, and v is the conveying speed of the baggage sorting system.

[0041] As an optional technical solution, the re-identification model adopts an architecture combining a Swin Transformer feature extractor and an FPN feature fusion module, wherein the outputs of each layer of the Swin Transformer feature extractor are used as inputs of the FPN feature fusion module; the device further comprises an offline retrieval module, and when the judgment module determines that the retrieval method of the luggage is an offline retrieval, the offline retrieval module is called, and the offline retrieval module is used to:

[0042] Start searching from the entrance according to the defined search step length;

[0043] Call the re-identification model to extract features of the target baggage and the second collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the second collected image comes from at most R cameras within a search step starting from the entrance;

[0044] If the match fails, move one search step away from the entrance and search again, and move the search towards the entrance until the match is successful.

[0045] In a third aspect, the present application discloses a time-constrained airport baggage cross-domain tracking and re-identification system, comprising: the device as described in the second aspect above and a plurality of cameras arranged along the baggage conveyor belt between the check-in counter and the loading and unloading area.

[0046] The beneficial effects of the present invention are:

[0047] 1. It makes up for the lack of inter-node monitoring in the airport baggage transportation and sorting system. This application uses a multi-camera re-identification method based on machine vision in the full-line supervision of baggage transportation, establishes a baggage re-identification model that takes baggage as the identification object and considers multi-camera data collection, and completes the monitoring, tracking and retrieval of the entire process of airport baggage transportation.

[0048] 2. Improved efficiency in handling baggage anomalies. When searching for anomalies such as lost baggage during transportation, you only need to call the baggage image collected at the transportation entrance according to the baggage number, and determine the search method and search range according to the time the baggage entered the system, and quickly locate the baggage to be found.

[0049] 3. Baggage retrieval with time constraints ensures the efficiency of baggage retrieval under multi-camera conditions. The airport baggage transportation line is equipped with more than 100 cameras. In order to solve the problem of low retrieval efficiency caused by too many cameras, according to the characteristics of the fixed route and running speed of the baggage sorting and transportation system, the R-way camera near the estimated baggage location is extracted for retrieval during retrieval, which greatly improves the efficiency of baggage retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flowchart of the airport baggage cross-domain tracking and re-identification method with time constraints.

[0051] Figure 2 This is the architecture diagram of the airport baggage cross-domain tracking and re-identification device with time constraints.

[0052] Figure 3 This is the architecture diagram of the airport baggage cross-domain tracking and re-identification system with time constraints.

[0053] Figure 4 It is a schematic diagram of the structural deployment of the system monitoring device.

[0054] Figure 5 2 is a schematic diagram of the overall process of re-identification query in the embodiment.

[0055] Figure 6 This is the architecture diagram of the baggage re-identification model.

[0056] Figure 7 This is the architecture diagram of the baggage detection model. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components and method steps of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] Example

[0059] like Figure 1 As shown, this embodiment discloses a method for cross-domain tracking and re-identification of airport luggage with time constraints, including: S1, collecting luggage image information, and uploading the image information to the server for storage after ID numbering; S2, using the luggage detection model to detect the luggage status of the image information, if the detection result is that the luggage status is normal, then proceed to the next step, otherwise, issue an alarm prompt for abnormal luggage status; S3, determine whether to perform luggage retrieval, if so, determine whether the luggage retrieval method is online retrieval, if so, proceed to step S4 to start online retrieval. Specifically, when abnormal signals such as luggage falling or stranded are received, it is determined that luggage retrieval is required.

[0060] The step S4 online search comprises:

[0061] S401, determining a starting search position;

[0062] S402, defining a search step length of a single search to lock a search area of ​​the single search, wherein the search step length of the single search includes at most R cameras, R<15, and specifically, in this embodiment, R can be 12;

[0063] S403, performing a first search, determining a first search area according to a defined search step, and the first search area includes the start search position;

[0064] S404, calling the re-identification model to extract features of the target baggage and the first collected image, and calculating the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the first collected image comes from at most R cameras in the determined first search area;

[0065] S405. If the match fails, move a defined search step from the first search area to the entrance and perform a secondary search as the secondary search area. If the secondary search fails, move a defined search step to the entrance again, and continue searching toward the entrance until the match succeeds.

[0066] As an optional implementation manner, if the detection result of the luggage status detection is that the luggage status is abnormal, the image of the abnormal luggage is cropped, rotated and color-changed.

[0067] As an optional implementation, the starting search position is s 0 Indicates that 0 Determined according to the following formula:

[0068] s 0 =(t 2 -t 1 )×v

[0069] Among them, t 1 is the time when the baggage to be retrieved enters the entrance of the baggage sorting system, t 2 is the current retrieval time, and v is the conveying speed of the baggage sorting system.

[0070] As an optional implementation, the search method further includes offline search, and the offline search includes:

[0071] Start searching from the entrance according to the defined search step length;

[0072] Call the re-identification model to extract features of the target baggage and the second collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the second collected image comes from at most R cameras within a search step starting from the entrance;

[0073] If the match fails, move one search step away from the entrance and search again, and move the search towards the entrance until the match is successful.

[0074] As an optional implementation, the re-identification model adopts an architecture that combines a Swin Transformer feature extractor and an FPN feature fusion module, wherein the output of each layer of the Swin Transformer feature extractor serves as the input of the FPN feature fusion module.

[0075] like Figure 2 As shown, this embodiment also discloses a device for re-identifying airport luggage across domains with time constraints, including:

[0076] The acquisition module is used to collect luggage image information, and upload the image information to the server for storage after assigning ID numbers to the image information;

[0077] A baggage detection module, used to detect the baggage status of the image information using a baggage detection model, and if the detection result shows that the baggage status is normal, the judgment module is called; otherwise, an alarm prompt indicating that the baggage status is abnormal is issued;

[0078] A judgment module is used to determine whether to perform baggage retrieval. If so, it determines whether the baggage retrieval method is online retrieval. If so, the online retrieval module is called;

[0079] The online search module is used for:

[0080] Determine the starting search position;

[0081] Define a search step length of a single search to lock the search area of ​​the single search, wherein the search step length of the single search includes at most R cameras, R<15, and specifically, in this embodiment, R can be 12;

[0082] Performing a first search, determining a first search area according to a defined search step, and the first search area includes the starting search position;

[0083] Call the re-identification model to extract features of the target baggage and the first collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the first collected image comes from at most R cameras in the determined first search area;

[0084] If the match fails, move a defined search step from the first search area to the entrance and perform a secondary search as the secondary search area. If the secondary search fails, move a defined search step to the entrance again, and continue searching towards the entrance until the match is successful.

[0085] As an optional implementation manner, the luggage detection module is further used for: when the detection result of the luggage status detection is that the luggage status is abnormal, performing cropping, rotation and color conversion operations on the image of the abnormal luggage.

[0086] As an optional implementation, the starting search position is s 0 Indicates that 0 Determined according to the following formula:

[0087] s 0 =(t 2 -t 1 )×v

[0088] Among them, t 1 is the time when the baggage to be retrieved enters the entrance of the baggage sorting system, t 2 is the current retrieval time, and v is the conveying speed of the baggage sorting system.

[0089] As an optional implementation, the re-identification model adopts an architecture combining a Swin Transformer feature extractor and an FPN feature fusion module, wherein the outputs of each layer of the Swin Transformer feature extractor are used as inputs of the FPN feature fusion module; the device further comprises an offline retrieval module, and when the judgment module determines that the retrieval method of the luggage is an offline retrieval, the offline retrieval module is called, and the offline retrieval module is used to:

[0090] Start searching from the entrance according to the defined search step length;

[0091] The re-identification model is called to extract features of the target baggage and the second collected image, and the cosine similarity of the extracted features is calculated. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the second collected image comes from at most R cameras within a search step starting from the entrance; wherein the value of the target threshold can be set according to actual conditions, for example, in this embodiment, the target threshold can be set to 95%, that is, when the calculated similarity reaches or exceeds 95%, the match is considered successful;

[0092] If the match fails, move one search step away from the entrance and search again, and move the search towards the entrance until the match is successful.

[0093] like Figure 3 As shown, the present application discloses a time-constrained airport baggage cross-domain tracking and re-identification system, comprising: the above-mentioned device and a plurality of cameras arranged along the baggage conveyor belt between the check-in counter and the loading and unloading area.

[0094] In order to better understand the present application, the present application is further described below.

[0095] In this embodiment, a baggage cross-domain tracking and identification system based on machine vision is designed based on the existing camera acquisition conditions on the current airport baggage transportation line, and the baggage is tracked in the whole area during the airport baggage transportation and sorting process. The image acquisition of the baggage is completed at the entrance of the baggage transportation and sorting system, and then the baggage is detected and tracked and located across cameras through the target detection model and the re-identification model. On this basis, due to the predictability of the baggage transportation route and time, time constraints are added when tracking the baggage, which reduces the calculation intensity of baggage retrieval and positioning, and ensures the efficiency of baggage positioning.

[0096] The following will be combined Figures 4 to 6 This embodiment is described in detail.

[0097] The system's image acquisition relies on the airport's existing security equipment and monitoring devices for the baggage sorting system. The structure is deployed as follows: Figure 4 The entire re-identification process can be roughly divided into the following three parts: baggage query image library generation; baggage calibration and anomaly monitoring; baggage re-identification retrieval. Figure 5 The specific plan is as follows:

[0098] 1. Baggage query image database generation: First, at the entrance of the baggage sorting system, such as the check-in gate, use the camera at the check-in gate to collect baggage images, assign ID numbers to the baggage images, and then upload the images and information such as the time of entering the system to the baggage query image database, which is stored in the server.

[0099] 2. Baggage calibration and abnormality inspection. Since the collected images are high-resolution images, the baggage detection uses the yolov5s6 model to calibrate and detect the luggage in the image. This step is mainly to identify and calibrate the luggage image area in the collected image, and at the same time warn of abnormal conditions of the luggage.

[0100] During the training phase, baggage data is labeled into two categories: “normal” and “abnormal”. At the same time, in order to balance the amount of the two types of data, operations such as cropping, rotation, and color change are performed on the baggage labeled as “abnormal” to increase the amount of data of this type and reduce the impact on model performance caused by this.

[0101] Baggage detection model Figure 7 As shown, the model is divided into four parts:

[0102] (1) Input: The image pre-processing is completed in the input part, including image enhancement and initialization of four sets of anchor frames;

[0103] (2) Feature extraction part (Backbone): The New CSP-Darknet53 architecture is used to extract useful features from the input image. The image is downsampled by 8, 16, 32, and 64 times in the 3rd, 5th, 7th, and 9th layers of the 12-layer network structure, and feature maps of sizes of 160×160×256, 80×80×512, 40×40×768, and 20×20×1024 are obtained respectively;

[0104] (3) Feature fusion part (Neck): The model uses the structure of an improved feature pyramid FPN (Feature Pyramid Network) network + a path aggregation network PAN (Path Aggregation Network) to complete the feature fusion operation of the extracted multi-scale features, and obtains four fusion feature maps with sizes of 160×160×256, 80×80×512, 40×40×768 and 20×20×1024 respectively;

[0105] (4) Detection part (Head): Target detection of objects of different sizes is performed on four fused feature maps of different sizes. The large feature map is responsible for detecting small objects, the larger feature map is responsible for detecting medium-sized objects, and the smallest feature map is responsible for detecting large objects.

[0106] 3. Baggage re-identification and retrieval. The retrieval function is divided into real-time retrieval (i.e. online retrieval) and offline retrieval. Real-time retrieval is to query and locate the luggage during the luggage transportation process, and retrieve the luggage location through the collected real-time video data; offline retrieval is to retrieve the historical data of the luggage transportation video. Specifically, it can be retrieved through t 2 -t 1 The difference between the current retrieval time and the 2 The time t when the baggage to be retrieved enters the entrance of the baggage sorting system (check-in counter) 1 The difference between the two determines which retrieval method to use: when t 2 -t 1 >T, use offline search; when t 2 -t 1 ≤T, the online search method is used. T represents the time required for the baggage to be transported from the check-in counter to the loading and unloading area according to the conveying speed v set by the baggage sorting system.

[0107] Real-time retrieval uses a time-constrained method, that is, according to the difference between the current retrieval time and the time when the luggage to be retrieved enters the sorting system, the retrieval area is first positioned at the estimated position before and after a maximum of 12 cameras for video retrieval according to the transmission speed of the sorting system. If the luggage to be retrieved is matched for the first time, the retrieval stops, otherwise the retrieval area is moved to the entrance, and the retrieval and matching of multiple cameras in the retrieval area are performed until the luggage is retrieved, and the retrieval ends. The step size and size of the moving retrieval area during the retrieval process can be customized.

[0108] Offline retrieval adopts global tracking retrieval, that is, starting from the moment the luggage to be retrieved enters the sorting system, the retrieval area is set according to a maximum of 12 cameras, and historical videos in the retrieval area are retrieved. If the retrieval luggage is matched, the retrieval stops; the retrieval area is moved according to a custom step length until the final position of the luggage to be retrieved is located.

[0109] Baggage retrieval is based on the baggage re-identification model. It extracts features from multiple pieces of luggage in the video image and then calculates and sorts them by similarity. If the highest similarity value is greater than the target threshold, it is considered a successful match, thereby completing the baggage search and retrieval.

[0110] Baggage re-identification model Figure 6 As shown in the figure, the model uses a standard 4-stage Swin Transformer as a feature extractor, the input image size is 128×128, and the image patch size is set to 4×4. After 4 stages, feature maps of sizes 32×32×96, 16×16×192, 8×8×384 and 4×4×768 are obtained respectively. Next, these feature maps of different levels are sent to the FPN feature fusion module for feature fusion. In the training phase, the feature map of stage 4 is used for maximum pooling MaxPooling to obtain the features as the global features Global Feature to calculate the ID Loss loss function, and the fused feature vector is stretched into a 1-dimensional vector to calculate the Triplet Loss loss function. In addition, after the image passes through the Liner embedding layer, a 1-dimensional learnable vector side information embedding module SIE (Side Information Embedding) can be inserted to represent the invariant features of the camera information and position for each patch.

[0111] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by combining software with necessary hardware. Therefore, the technical solution according to the embodiment of the present application can be embodied in the form of a software product, which can be stored in a readable medium or on a network, and includes several instructions to enable an electronic device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiment of the present application.

[0112] The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. All technical solutions that fall within the scope defined by the claims of the present invention fall within the protection scope of the present invention.

Claims

1. The airport baggage cross-domain tracking and re-identification method with time constraints is characterized by: include: Collecting luggage image information, assigning ID numbers to the image information, and uploading it to the server for storage; Using the baggage detection model to detect the baggage status of the image information, if the detection result is that the baggage status is normal, proceed to the next step, otherwise, an alarm prompt of abnormal baggage status is issued; Determine whether to perform baggage retrieval, and if so, determine whether the baggage retrieval method is online retrieval, and if so, perform online retrieval; The online search includes: Determine the starting search position; Define a search step length of a single search to lock the search area of ​​the single search, wherein the search step length of the single search includes at most R cameras, R<15; Performing a first search, determining a first search area according to a defined search step, and the first search area includes the starting search position; Call the re-identification model to extract features of the target baggage and the first collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the first collected image comes from at most R cameras in the determined first search area; If the match fails, move a defined search step from the first search area to the entrance and perform a secondary search as the secondary search area. If the secondary search fails, move a defined search step to the entrance again, and continue searching towards the entrance until the match is successful.

2. The airport baggage cross-domain tracking and re-identification method with time constraints according to claim 1 is characterized by: If the detection result of the luggage status detection is that the luggage status is abnormal, the image of the abnormal luggage is cropped, rotated and the color is changed.

3. The airport baggage cross-domain tracking and re-identification method with time constraints according to claim 1 is characterized in that: The starting search position is denoted by s0, and s0 is determined according to the following formula: s0=(t2-t1)×v Among them, t1 is the time when the luggage to be retrieved enters the entrance of the baggage sorting system, t2 is the current retrieval time, and v is the conveying speed of the baggage sorting system.

4. The method for cross-domain tracking and re-identification of airport baggage with time constraints according to claim 1 is characterized in that: The retrieval method also includes offline retrieval, and the offline retrieval includes: Start searching from the entrance according to the defined search step length; Call the re-identification model to extract features of the target baggage and the second collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the second collected image comes from at most R cameras within a search step starting from the entrance; If the match fails, move one search step away from the entrance and search again, and move the search towards the entrance until the match is successful.

5. The method for cross-domain tracking and re-identification of airport baggage with time constraints according to claim 1 or 4, characterized in that: The re-identification model adopts an architecture combining a Swin Transformer feature extractor and an FPN feature fusion module, wherein the output of each layer of the Swin Transformer feature extractor serves as the input of the FPN feature fusion module.

6. The device for re-identifying airport baggage across domains with time constraints is characterized by: include: The acquisition module is used to collect luggage image information, and upload the image information to the server for storage after assigning ID numbers to the image information; A baggage detection module, used to detect the baggage status of the image information using a baggage detection model, and if the detection result shows that the baggage status is normal, the judgment module is called; otherwise, an alarm prompt indicating that the baggage status is abnormal is issued; A judgment module is used to determine whether to perform baggage retrieval. If so, it determines whether the baggage retrieval method is online retrieval. If so, the online retrieval module is called; The online search module is used for: Determine the starting search position; Define a search step length of a single search to lock the search area of ​​the single search, wherein the search step length of the single search includes at most R cameras, R<15; Performing a first search, determining a first search area according to a defined search step, and the first search area includes the starting search position; Call the re-identification model to extract features of the target baggage and the first collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the first collected image comes from at most R cameras in the determined first search area; If the match fails, move a defined search step from the first search area to the entrance and perform a secondary search as the secondary search area. If the secondary search fails, move a defined search step to the entrance again, and continue searching towards the entrance until the match is successful.

7. The device for cross-domain tracking and re-identification of airport luggage with time constraints according to claim 6 is characterized in that: The luggage detection module is further used for: when the detection result of the luggage status detection is that the luggage status is abnormal, performing cropping, rotation and color conversion operations on the image of the abnormal luggage.

8. The device for cross-domain tracking and re-identification of airport luggage with time constraints according to claim 6 is characterized in that: The starting search position is denoted by s0, and s0 is determined according to the following formula: s0=(t2-t1)×v Among them, t1 is the time when the luggage to be retrieved enters the entrance of the baggage sorting system, t2 is the current retrieval time, and v is the conveying speed of the baggage sorting system.

9. The device for cross-domain tracking and re-identification of airport luggage with time constraints according to claim 6, characterized in that: The re-identification model adopts an architecture combining a Swin Transformer feature extractor and an FPN feature fusion module, wherein the outputs of each layer of the Swin Transformer feature extractor are used as inputs of the FPN feature fusion module; the device further comprises an offline retrieval module, and when the judgment module determines that the retrieval method of the luggage is an offline retrieval, the offline retrieval module is called, and the offline retrieval module is used to: Start searching from the entrance according to the defined search step length; Call the re-identification model to extract features of the target baggage and the second collected image, and calculate the similarity of the extracted features. If the similarity is greater than the target threshold, the match is successful, the camera number corresponding to the successfully matched image is output, and the search is stopped; wherein the second collected image comes from at most R cameras within a search step starting from the entrance; If the match fails, move one search step away from the entrance and search again, and move the search towards the entrance until the match is successful.

10. Airport baggage cross-domain tracking and re-identification system with time constraints, characterized by: include: An apparatus as claimed in any one of claims 6 to 9 and a plurality of cameras arranged along a baggage conveyor belt between a check-in counter and a loading and unloading area.

Images