Lost property recovery method, system, device and storage medium

Abstract

The present invention discloses a lost property recovery method, system, device, and storage medium. The key technical aspects of the method are: extracting features from a received lost property image to obtain a first feature vector; obtaining video data containing the lost property and at least one person, and extracting a second feature vector for each person in the video data; calculating the closeness between the lost property and the corresponding person based on the first feature vector and each second feature vector; obtaining the lost property's location and the positions of each person based on the video data, and calculating a positional relationship metric between the lost property and the corresponding person based on the lost property's location and the positions of each person; and calculating a corresponding comprehensive relationship metric based on the corresponding closeness and positional relationship metric for each person, with the person with the largest comprehensive relationship metric being identified as the owner. The present invention can proactively locate the owner, reduce the storage space occupied by the lost property, reduce the owner's losses, and avoid resource waste.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a lost property recovery method, system, device and storage medium. Background Art

[0002] In real life, it is very easy for items we carry with us to be lost. For example, headphones, umbrellas, water cups, etc. may be left behind after being placed carelessly and become lost items. Generally, lost items will be sent to the lost and found office.

[0003] However, a large number of lost items at the lost and found office are not claimed by the owners, resulting in an increasing number of lost items. More and more space is needed to store lost items, and the owners have to spend money to buy the lost items again, resulting in a waste of resources. Summary of the Invention

[0004] The purpose of the present invention is to provide a lost property recovery method, system, device and storage medium, which can actively find the owner, reduce the storage space occupied by the lost property, reduce the loss of the owner, and avoid waste of resources.

[0005] A first aspect of the present invention provides a lost property recovery method, comprising:

[0006] Performing feature extraction on the received lost property image to obtain a first feature vector;

[0007] Acquire video data containing lost property and at least one person, and extract a second feature vector of each person in the video data;

[0008] Calculating the closeness between the lost object and the corresponding person based on the first feature vector and each second feature vector;

[0009] Obtaining the location of the lost object and the locations of each person according to the video data, and calculating a positional relationship measurement value between the lost object and the corresponding person according to the location of the lost object and the locations of each person;

[0010] According to the closeness and position relationship measurement values ​​corresponding to each of the characters, a corresponding comprehensive relationship measurement value is calculated, and the character with the largest comprehensive relationship measurement value is determined as the owner.

[0011] In some embodiments, calculating the closeness between the lost item and the corresponding person based on the first feature vector and each second feature vector includes:

[0012] Substitute the first eigenvector and each second eigenvector into a compactness calculation formula to calculate the corresponding compactness. The compactness calculation formula is:

[0013]

[0014] Wherein, X represents the first eigenvector, and P i represents the second eigenvector, the ||X|| represents the norm of the first eigenvector, the ||P i || represents the norm of the second eigenvector.

[0015] In some embodiments, the calculating of the position relationship metric between the lost object and the corresponding person based on the lost object location and the positions of each person includes:

[0016] Substitute the lost item location and the location of each person into the position relationship metric value calculation formula to calculate the corresponding position relationship metric value. The position relationship metric value calculation formula is:

[0017]

[0018] Among them, (x0, y0) represents the location of the lost item, Indicates the position of the i-th person.

[0019] In some embodiments, the calculating of the corresponding comprehensive relationship metric value based on the closeness and position relationship metric values ​​corresponding to each of the characters includes:

[0020] Substitute each of the closeness and position relationship metrics into the comprehensive relationship metric value calculation formula to calculate the corresponding comprehensive relationship metric value. The comprehensive relationship metric value calculation formula is:

[0021] S i =β*cosine_sim(X,P i )-α*D i ,

[0022] Wherein, α represents the position weight, β represents the feature weight, and D i Indicates the closeness corresponding to the i-th person, the cosine_sim(X, P i ) represents the position relationship measurement value corresponding to the i-th person.

[0023] In some embodiments, obtaining the location of the lost item and the locations of each person based on the video data includes:

[0024] Recognize the outline of the lost object and the outlines of each person according to the video data;

[0025] Determine the lost property center according to the lost property outline, and use the lost property center as the lost property location;

[0026] The corresponding character center is determined according to each character outline, and each character center is used as the corresponding character position.

[0027] In some embodiments, obtaining video data containing the lost property and at least one person includes:

[0028] Obtaining the original video and the time when the lost property was picked up, wherein the original video is the video of the place where the lost property was picked up;

[0029] Subtract the preset time period from the lost property picking up time to obtain the interception time;

[0030] Starting from the interception time, sequentially selecting video frames along the time sequence, and performing lost object recognition on the video frames until no lost object is found in the selected video frames, thereby obtaining a first video clip containing the lost object;

[0031] Perform person recognition on the video frames in the first video clip, select the video frames with people, obtain a second video clip, and acquire video data containing the lost property and at least one person.

[0032] A second aspect of the present invention provides a lost property recovery system, comprising:

[0033] A first extraction module is used to extract features from the received lost property image to obtain a first feature vector;

[0034] a second extraction module, configured to obtain video data containing the lost item and at least one person, and extract a second feature vector of each person in the video data;

[0035] A first calculation module, configured to calculate the closeness between the lost object and the corresponding person based on the first feature vector and each second feature vector;

[0036] A second calculation module is used to obtain the position of the lost object and the positions of each person according to the video data, and calculate the position relationship measurement value between the lost object and the corresponding person according to the position of the lost object and the positions of each person;

[0037] The comprehensive calculation module is used to calculate the corresponding comprehensive relationship measurement value according to the closeness and position relationship measurement values ​​corresponding to each of the characters, and to select the character with the largest comprehensive relationship measurement value as the owner.

[0038] In some embodiments, the first computing module includes:

[0039] A compactness calculation unit is configured to substitute the first eigenvector and each second eigenvector into a compactness calculation formula to calculate corresponding compactness. The compactness calculation formula is:

[0040]

[0041] Wherein, X represents the first eigenvector, Pi represents the second eigenvector, ||X|| represents the norm of the first eigenvector, and ||P i || represents the norm of the second eigenvector.

[0042] A third aspect of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0043] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0044] The technical solution provided by the present invention has the following advantages and effects: by extracting the first eigenvector of the lost object image and the second eigenvector of each person, and calculating the closeness of the two, then calculating the position relationship measurement value between the lost object and each person in the video data, combining the closeness and the relationship measurement value to obtain a comprehensive relationship measurement value, and identifying the person with the highest comprehensive relationship measurement value as the owner, so as to actively find the owner, reduce the storage space occupied by the lost object, reduce the loss of the owner, and avoid waste of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic flow chart of the lost property recovery method provided by the present invention;

[0046] Figure 2 It is a structural block diagram of the lost property recovery system provided by the present invention;

[0047] Figure 3 It is a diagram of the internal structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] To facilitate understanding of the present invention, specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings.

[0049] Unless otherwise specified or defined, the "first, second..." used in this article is only used to distinguish names and does not represent a specific quantity or order.

[0050] Unless stated otherwise or defined otherwise, the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0051] It should be noted that, in this document, “fixed to” or “connected to” may mean directly fixing or connecting to an element, or indirectly fixing or connecting to an element.

[0052] like Figure 1As shown, this embodiment provides a lost property recovery method, including the following steps S1 to S5:

[0053] Step S1: extract features from the received lost property image to obtain a first feature vector.

[0054] In practical applications, lost items may be personal belongings such as headphones, mobile phones, water cups, umbrellas, etc. The lost item image is obtained by photographing the lost item with a camera. Images of the lost item from multiple different angles, such as the front, side, and back of the lost item, can be photographed. Then, a trained lost item recognition model is used to extract the first feature vector of the lost item. Specifically, the lost item recognition model can adopt the Faster R-CNN pre-trained model. The Faster R-CNN pre-trained model has been pre-trained using an object training set for a large number of trainings. The object training set includes a large number of photos of different objects from different angles. After the Faster R-CNN pre-trained model is trained until the model converges or reaches the number of iterations, the Faster R-CNN pre-trained model is tested using an object test set. The object test set includes a large number of photos of different objects from different angles. After the test results meet the expected results, the tested Faster R-CNN pre-trained model is used as the lost item recognition model.

[0055] Step S2: Obtain video data containing the lost property and at least one person, and extract a second feature vector of each person in the video data.

[0056] Specifically, obtaining video data containing the lost property and at least one person includes:

[0057] Obtaining the original video and the time when the lost property was picked up, wherein the original video is the video of the place where the lost property was picked up;

[0058] Subtract the preset time period from the lost property picking up time to obtain the interception time;

[0059] Starting from the interception time, sequentially selecting video frames along the time sequence, and performing lost object recognition on the video frames until no lost object is found in the selected video frames, thereby obtaining a first video clip containing the lost object;

[0060] Perform person recognition on the video frames in the first video clip, select the video frames with people, obtain a second video clip, and acquire video data containing the lost property and at least one person.

[0061] Specifically, the original video can be the surveillance time or mobile phone video. After the lost property is picked up, the surveillance video of the place where the lost property was picked up can be retrieved. The time when the lost property was picked up can be learned by asking the person who picked up the lost property. In order to avoid misidentifying the person who picked up the lost property as the owner in the subsequent identification process, and to reduce the subsequent identification of the person who picked up the lost property, feature vector extraction, closeness calculation and position relationship measurement value calculation, a preset time period is set, such as 30 seconds, 1 minute, 1 minute and 30 seconds, 2 minutes, etc. For example, if the lost property is picked up in the cafeteria at 10 o'clock on the same day, the preset time period is 1 minute, then the interception time is 9:59, and video frames are selected from 9:59 onward in the surveillance video of the cafeteria. The video frames are recognized by the lost property recognition model. If the lost property is not recognized in the selected video frames, the selection is stopped, and multiple video frames containing the lost property are obtained to form a first video clip. Then, person recognition is performed on the video frames of the first video clip to obtain multiple video frames containing the lost property and the person to form a second video clip. For person recognition and feature vector extraction, a person recognition model can be used for feature extraction and recognition. The person recognition model can also be obtained after a large amount of training using the Faster R-CNN pre-trained model. Before training, a large number of person images of different persons at different angles are collected, and then the person images are divided into a person training set and a person test set. The Faster R-CNN pre-trained model is trained with the person training set until the model converges or reaches the number of iteration rounds. The Faster R-CNN pre-trained model is tested with the person test set. If the test results meet the expected results, the tested Faster R-CNN pre-trained model is used as the person recognition model.

[0062] Step S3: Calculate the closeness between the lost item and the corresponding person based on the first feature vector and each second feature vector.

[0063] Specifically, calculating the closeness between the lost item and the corresponding person based on the first feature vector and each second feature vector includes:

[0064] Substitute the first eigenvector and each second eigenvector into a compactness calculation formula to calculate the corresponding compactness. The compactness calculation formula is:

[0065]

[0066] Wherein, X represents the first eigenvector, and P i represents the second eigenvector, the ||X|| represents the norm of the first eigenvector, the ||P i || represents the norm of the second eigenvector.

[0067] In actual applications, before the lost item is lost, the owner usually carries the lost item with them. The lost item recognition model identifies the first eigenvector of the lost item, and the person recognition model identifies the corresponding second eigenvector. The closeness calculation formula is used to calculate the closeness of the first eigenvector and the second eigenvector. When the lost item is carried by the owner, the closeness of the lost item's second eigenvector and the first eigenvector is higher than the closeness of the second eigenvector and the first eigenvector of other persons. Therefore, by calculating the closeness of the first eigenvector and the second eigenvector, the accuracy of identifying the lost item from various persons can be improved.

[0068] Step S4: obtaining the position of the lost object and the positions of each person according to the video data, and calculating a position relationship measurement value between the lost object and the corresponding person according to the position of the lost object and the positions of each person.

[0069] Specifically, obtaining the location of the lost item and the locations of each person according to the video data includes:

[0070] Recognize the outline of the lost object and the outlines of each person according to the video data;

[0071] Determine the lost property center according to the lost property outline, and use the lost property center as the lost property location;

[0072] The corresponding character center is determined according to each character outline, and each character center is used as the corresponding character position.

[0073] In practical applications, a lost object recognition model and a person recognition model are used to identify each frame of the video data, obtain the lost object outline and each person outline in each frame of the image, extract multiple first coordinates of the lost object outline and multiple second coordinates of each person outline, and the coordinate calculation method of the lost object center includes:

[0074]

[0075] Among them, (x i ,y i ) represents the i-th first coordinate, n represents the number of first coordinates, and (x0, y0) represents the center of the lost object, that is, the location of the lost object.

[0076] The coordinate calculation method of the character center includes:

[0077]

[0078] Among them, (x ij ,y ij ) represents the jth second coordinate of the i-th person, m represents the number of first coordinates, Represents the center of the i-th character, that is, the position of the i-th character.

[0079] Specifically, the calculating of the position relationship measurement value between the lost object and the corresponding person according to the lost object location and the positions of each person includes:

[0080] Substitute the lost item location and the location of each person into the position relationship metric value calculation formula to calculate the corresponding position relationship metric value. The position relationship metric value calculation formula is:

[0081]

[0082] Among them, (x0, y0) represents the location of the lost item, Indicates the position of the i-th person.

[0083] In actual applications, before the lost item is lost, the owner usually carries the lost item with them. The absolute distance between the lost item's location and the locations of various people is calculated using the position relationship measurement value calculation formula. When the lost item is carried by the owner, the distance between the lost item and the owner is the closest compared to the distance between the lost item and other people. Therefore, by calculating the position relationship measurement value between the lost item's location and the locations of various people, the accuracy of identifying the lost owner from various people can be further improved.

[0084] Step S5: Calculate the corresponding comprehensive relationship measurement value based on the closeness and position relationship measurement values ​​corresponding to each of the characters, and select the character with the largest comprehensive relationship measurement value as the owner.

[0085] Specifically, the calculating of the corresponding comprehensive relationship metric value according to the closeness and position relationship metric value corresponding to each of the characters includes:

[0086] Substitute each of the closeness and position relationship metrics into the comprehensive relationship metric value calculation formula to calculate the corresponding comprehensive relationship metric value. The comprehensive relationship metric value calculation formula is:

[0087] S i =β*cosine_sim(X,P i )-α*D i ,

[0088] Wherein, α represents the position weight, β represents the feature weight, and D i Indicates the closeness corresponding to the i-th person, the cosine_sim(X, P i ) represents the position relationship measurement value corresponding to the i-th person.

[0089] In practical applications, the comprehensive relationship metric combines both the closeness and location relationship metrics. The person with the highest comprehensive relationship metric indicates the closest connection to the lost item and the highest probability of being the owner. Therefore, the person with the highest comprehensive relationship metric is considered the owner, and the owner is contacted to retrieve the lost item. This reduces the storage space occupied by the lost item, reduces the owner's loss, and avoids resource waste. Lost items vary in both the loss scenario and size, and both affect the closeness and location weights. Therefore, the location and feature weights can be dynamically adjusted based on the actual prediction accuracy. For example, if the initial location weight is 0.5 and the feature weight is 0.5, if the actual prediction accuracy does not meet the expected accuracy, the location weight can be adjusted to 0.7 and the feature weight to 0.3 if the lost item size is within the preset volume. If the lost item size is larger than the preset volume, the location weight can be adjusted to 0.4 and the feature weight to 0.6, thereby improving the accuracy of locating lost items in different scenarios and different owner-recoveries.

[0090] The lost property recovery method of the present invention extracts the first eigenvector of the lost property image and the second eigenvector of each person, calculates the closeness of the two, and then calculates the position relationship measurement value between the lost property and each person in the video data. The closeness and relationship measurement value are combined to obtain a comprehensive relationship measurement value. The person with the highest comprehensive relationship measurement value is determined as the owner, so as to actively find the owner, reduce the storage space occupied by the lost property, reduce the owner's loss, and avoid resource waste.

[0091] like Figure 2 As shown, an embodiment of the present invention further provides a lost property recovery system, comprising:

[0092] A first extraction module 10 is used to extract features from the received lost property image to obtain a first feature vector;

[0093] A second extraction module 20 is configured to obtain video data containing the lost item and at least one person, and extract a second feature vector of each person in the video data;

[0094] A first calculation module 30, configured to calculate the closeness between the lost object and the corresponding person based on the first feature vector and each second feature vector;

[0095] A second calculation module 40 is configured to obtain the location of the lost object and the locations of each person according to the video data, and calculate a positional relationship metric between the lost object and the corresponding person according to the location of the lost object and the locations of each person;

[0096] The comprehensive calculation module 50 is used to calculate the corresponding comprehensive relationship measurement value according to the closeness and position relationship measurement values ​​corresponding to each of the characters, and select the character with the largest comprehensive relationship measurement value as the owner.

[0097] As an optional implementation, the second extraction module includes the following units (not shown):

[0098] An acquisition unit, configured to acquire an original video and a time when the lost property was picked up, wherein the original video is a video of the place where the lost property was picked up;

[0099] A time determination unit, configured to subtract a preset time period from the lost property picking-up time to obtain a pick-up time;

[0100] A first selection unit is configured to sequentially select video frames from the interception time forward along the time sequence, perform lost object recognition on the video frames, and stop selecting when no lost object is found in the selected video frames, thereby obtaining a first video segment containing the lost object;

[0101] The second selection unit is used to perform person recognition on the video frames in the first video clip, select the video frames with people, obtain the second video clip, and obtain video data containing the lost property and at least one person.

[0102] As an optional implementation, the first calculation module includes the following units (not shown):

[0103] A compactness calculation unit is configured to substitute the first eigenvector and each second eigenvector into a compactness calculation formula to calculate corresponding compactness. The compactness calculation formula is:

[0104]

[0105] Wherein, X represents the first eigenvector, Pi represents the second eigenvector, ||X|| represents the norm of the first eigenvector, and ||P i || represents the norm of the second eigenvector.

[0106] In some embodiments, the second computing module includes the following units (not shown):

[0107] The position metric value calculation unit is used to substitute the lost property location and the positions of each person into the position relationship metric value calculation formula to calculate the corresponding position relationship metric value. The position relationship metric value calculation formula is:

[0108]

[0109] Among them, (x0, y0) represents the location of the lost item, Indicates the position of the i-th person.

[0110] In some embodiments, the second computing module further includes the following units (not shown):

[0111] A contour recognition unit, configured to recognize the contours of lost objects and people based on the video data;

[0112] A first position determining unit is configured to determine a lost property center according to the lost property outline, and use the lost property center as the lost property location;

[0113] The second position determining unit is configured to determine a corresponding character center according to each character outline, and use each character center as a corresponding character position.

[0114] In some embodiments, the comprehensive calculation module includes the following units (not shown):

[0115] The comprehensive metric value calculation unit is used to substitute each of the closeness and position relationship metric values ​​into the comprehensive relationship metric value calculation formula to calculate the corresponding comprehensive relationship metric value. The comprehensive relationship metric value calculation formula is:

[0116] S i =β*cosine_sim(X,P i )-α*D i ,

[0117] Wherein, α represents the position weight, β represents the feature weight, and D i Indicates the closeness corresponding to the i-th person, the cosine_sim(X, P i ) represents the position relationship measurement value corresponding to the i-th person.

[0118] Each module of the lost property recovery system described above may be implemented in whole or in part through software, hardware, or a combination thereof. Each of the modules and units described above may be embedded in or independent of a processor within a computer device in hardware form, or may be stored in a memory within the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0119] like Figure 3 As shown, an embodiment of the present invention discloses a computer device, including a memory and a processor, wherein the memory stores a computer program;

[0120] The computer device may be a server, and its internal structure diagram may be as follows: Figure 3As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements the lost property recovery method described in the above embodiments.

[0121] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0122] An embodiment of the present invention further discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the lost property recovery method described in the above embodiments.

[0123] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0124] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A lost property recovery method, characterized in that: include: Performing feature extraction on the received lost property image to obtain a first feature vector; Acquire video data containing lost property and at least one person, and extract a second feature vector of each person in the video data; Calculating the closeness between the lost object and the corresponding person based on the first feature vector and each second feature vector; Obtaining the location of the lost object and the locations of each person according to the video data, and calculating a positional relationship measurement value between the lost object and the corresponding person according to the location of the lost object and the locations of each person; Calculate the corresponding comprehensive relationship measurement value based on the closeness and position relationship measurement values ​​corresponding to each of the characters, and select the character with the largest comprehensive relationship measurement value as the owner; The step of calculating the corresponding comprehensive relationship measurement value based on the closeness and position relationship measurement values ​​corresponding to each of the characters includes: Substitute each of the closeness and position relationship metrics into the comprehensive relationship metric value calculation formula to calculate the corresponding comprehensive relationship metric value. The comprehensive relationship metric value calculation formula is: ， Among them, the represents the position weight, represents the feature weight, the X represents the first feature vector, the represents the second eigenvector, the represents the position relationship measurement value corresponding to the i-th person, Indicates the closeness corresponding to the i-th person; the position weight and feature weight are dynamically adjusted according to the accuracy of the actual prediction results.

2. The lost property recovery method according to claim 1, wherein: The calculating the closeness between the lost object and the corresponding person according to the first feature vector and each second feature vector includes: Substitute the first eigenvector and each second eigenvector into a compactness calculation formula to calculate the corresponding compactness. The compactness calculation formula is: , Wherein, X represents the first eigenvector, and P i represents the second eigenvector, the represents the norm of the first eigenvector, represents the norm of the second eigenvector.

3. The lost property recovery method according to claim 1, wherein: The calculating of the position relationship measurement value between the lost object and the corresponding person according to the lost object location and the positions of each person includes: Substitute the lost item location and the location of each person into the position relationship metric value calculation formula to calculate the corresponding position relationship metric value. The position relationship metric value calculation formula is: , Among them, (x0, y0) represents the location of the lost item, Indicates the position of the i-th person.

4. The lost property recovery method according to any one of claims 1 to 3, characterized in that: The obtaining of the location of the lost item and the location of each person according to the video data includes: Recognize the outline of the lost object and the outlines of each person according to the video data; Determine the lost property center according to the lost property outline, and use the lost property center as the lost property location; The corresponding character center is determined according to each character outline, and each character center is used as the corresponding character position.

5. The lost property recovery method according to any one of claims 1 to 3, characterized in that: The obtaining of video data containing the lost item and at least one person includes: Obtaining the original video and the time when the lost property was picked up, wherein the original video is the video of the place where the lost property was picked up; Subtract the preset time period from the lost property picking up time to obtain the interception time; Starting from the interception time, sequentially selecting video frames along the time sequence, and performing lost object recognition on the video frames until no lost object is found in the selected video frames, thereby obtaining a first video clip containing the lost object; Perform person recognition on the video frames in the first video clip, select the video frames with people, obtain a second video clip, and acquire video data containing the lost property and at least one person.

6. Lost property recovery system, characterized in that: include: A first extraction module is used to extract features from the received lost property image to obtain a first feature vector; a second extraction module, configured to obtain video data containing the lost item and at least one person, and extract a second feature vector of each person in the video data; A first calculation module, configured to calculate the closeness between the lost object and the corresponding person based on the first feature vector and each second feature vector; A second calculation module is used to obtain the position of the lost object and the positions of each person according to the video data, and calculate the position relationship measurement value between the lost object and the corresponding person according to the position of the lost object and the positions of each person; A comprehensive calculation module is used to calculate the corresponding comprehensive relationship measurement value according to the closeness and position relationship measurement values ​​corresponding to each of the characters, and select the person with the largest comprehensive relationship measurement value as the owner; Calculating the corresponding comprehensive relationship metric value based on the closeness and position relationship metric values ​​corresponding to each of the characters includes: substituting each of the closeness and position relationship metric values ​​into a comprehensive relationship metric value calculation formula to calculate the corresponding comprehensive relationship metric value, wherein the comprehensive relationship metric value calculation formula is: , wherein the represents the position weight, represents the feature weight, the X represents the first feature vector, the P i represents the second eigenvector, the Indicates the closeness corresponding to the i-th person, Represents the position relationship measurement value corresponding to the i-th person; the position weight and feature weight can be dynamically adjusted according to the accuracy of the actual prediction results.

7. The lost property recovery system according to claim 6, wherein: The first calculation module includes: A compactness calculation unit is configured to substitute the first eigenvector and each second eigenvector into a compactness calculation formula to calculate corresponding compactness. The compactness calculation formula is: , Wherein, X represents the first eigenvector, and P i represents the second eigenvector, the represents the norm of the first eigenvector, represents the norm of the second eigenvector.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images