Multi-sequence gait retrieval method, device, electronic device and medium

Abstract

The present application provides a multi-sequence gait retrieval method, device, electronic device and medium proposed in an embodiment of the present application, wherein the method can obtain multiple character features of a target person; wherein the character features include clothing features, angle features and gait features; based on the clothing features and / or angle features, the multiple character features are divided into multiple target categories; based on at least part of the gait features of each target category, a target reference gait feature is calculated; based on the target reference gait feature, the target gait feature is retrieved to obtain a gait retrieval result; the reference gait feature is more comprehensive and can more accurately represent the walking posture of the person in reality, thereby more accurately performing gait feature retrieval.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and more specifically, to a multi-sequence gait retrieval method, device, electronic device, and medium. Background Art

[0002] With increasing attention to public safety, video surveillance cameras are becoming ubiquitous in cities, and various recognition algorithms are finding widespread application in areas such as safe production, intelligent security, and intelligent traffic management. Traditional identity recognition methods require the use of additional information, such as high-definition facial images. However, because facial recognition involves sensitive facial image information and is limited by factors such as close proximity, angle deviation, lighting, and partial occlusion, the performance of identity recognition systems in surveillance scenarios is severely impacted. To address this issue, research is developing gait recognition systems that utilize walking posture as identification information for medium- and long-range identification.

[0003] However, identity recognition is performed only through a gait sequence, which cannot accurately represent the walking posture of the person. Summary of the Invention

[0004] In view of this, the purpose of the present application is to provide a multi-sequence gait retrieval method, device, electronic device and medium, which can perform gait feature retrieval based on multiple gait features, thereby more accurately identifying the identity information of the target person.

[0005] The present invention provides a multi-sequence gait retrieval method, comprising:

[0006] Acquire multiple character features of the target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle of the character image;

[0007] Based on clothing features and / or angle features, the plurality of character features are divided into a plurality of target categories; wherein the clothing features and / or angle features in the character features of each target category meet the preset classification conditions corresponding to the target category;

[0008] Calculating a target reference gait feature of the target person based on at least part of the gait feature of each target category of the target person;

[0009] According to the target reference gait feature of the target person, the target gait feature is retrieved from a pre-established gait feature library to obtain a gait retrieval result; the similarity between the target gait feature and the target reference gait feature meets a preset similarity condition.

[0010] In some embodiments, in the multi-sequence gait retrieval method, classifying the multiple person features into multiple target categories based on clothing features and / or angle features includes:

[0011] Determining at least one target classification strategy from a plurality of preset classification strategies based on attributes of a plurality of character features; wherein the attributes of the plurality of character features include one of the following: shooting locations of the plurality of character images, shooting times of the plurality of character images; the character images are images from which the character features are extracted;

[0012] According to the target classification strategy, the multiple character features are divided into multiple categories.

[0013] In some embodiments, in the multi-sequence gait retrieval method, the plurality of person features are divided into a plurality of target categories according to the target classification strategy, which are at least one of the following:

[0014] Classifying clothing features among the plurality of character features, determining clothing categories according to the classification results, and classifying the plurality of character features into a plurality of target categories according to the clothing categories;

[0015] Alternatively, the angle features among the plurality of character features are classified, angle categories are determined according to the classification results, and the plurality of character features are classified into a plurality of target categories according to the angle categories;

[0016] Alternatively, the angle features among the plurality of character features are first classified, and angle categories are determined based on the classification results. The plurality of character features are then classified into a plurality of first categories based on the angle categories. The clothing features of each first category are then classified, and clothing categories are determined based on the classification results. The character features of each first category are then classified into at least one target category based on the clothing categories.

[0017] Alternatively, the clothing features among the multiple character features are first classified, the clothing category is determined based on the classification results, the multiple character features are divided into multiple first categories based on the clothing category, and then the angle features of each first category are classified, the angle category is determined based on the classification results, and the character features of each first category are divided into at least one target category based on the angle category.

[0018] In some embodiments, in the multi-sequence gait retrieval method, classifying clothing features among multiple character features and determining clothing categories based on the classification results includes:

[0019] Performing cluster analysis on the clothing features among the plurality of character features according to a preset cluster analysis algorithm to obtain at least one cluster;

[0020] Each cluster is considered as a clothing category.

[0021] In some embodiments, in the multi-sequence gait retrieval method, classifying angle features among multiple character features includes:

[0022] Inputting the plurality of character features into a trained viewpoint classification model;

[0023] The angle features in the plurality of character features are processed by the perspective classification model to obtain a perspective label for each angle feature; wherein, angle features with the same perspective label have the same angle category.

[0024] In some embodiments, in the multi-sequence gait retrieval method, calculating the target reference gait features of the target person based on at least part of the gait features of each target category of the target person includes:

[0025] Calculating a class reference gait feature for each target class based on at least a portion of the gait features of the target person;

[0026] Based on the class reference gait features of each target class, a target reference gait feature of the target person is calculated.

[0027] In some embodiments, the multi-sequence gait retrieval method retrieves target gait features from a pre-established gait feature library based on target reference gait features of the target person; including:

[0028] Calculating the similarity between the gait features that meet the screening conditions in the gait feature library and the target reference gait features of the target person respectively;

[0029] Target gait features whose similarity meets the preset similarity conditions are again screened out from the gait features that meet the screening conditions.

[0030] In some embodiments, a multi-sequence gait retrieval device is further provided, comprising:

[0031] An acquisition module is used to acquire multiple character features of a target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle in the character image;

[0032] a classification module, configured to classify the plurality of character features into a plurality of target categories based on clothing features and / or angle features; wherein the clothing features and / or angle features of the character features in each target category satisfy a preset classification condition corresponding to the target category;

[0033] a calculation module, configured to calculate a target reference gait feature of the target person based on at least a portion of the gait features of each target category of the target person;

[0034] The retrieval module is used to retrieve the target gait feature from a pre-established gait feature library based on the target reference gait feature of the target person to obtain a gait retrieval result; the similarity between the target gait feature and the target reference gait feature meets the preset similarity condition.

[0035] In some embodiments, an electronic device is also provided, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the multi-sequence gait retrieval method are performed.

[0036] In some embodiments, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the multi-sequence gait retrieval method are executed.

[0037] Embodiments of the present application provide a multi-sequence gait retrieval method, apparatus, electronic device, and medium. The multi-sequence gait retrieval method can obtain multiple character features of a target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle in the character image; based on the clothing features and / or angle features, the multiple character features are divided into multiple target categories; wherein the clothing features and / or angle features in the character features in each target category meet the preset classification conditions corresponding to the target category; based on at least part of the gait features of each target category of the target person, a target reference gait feature of the target person is calculated; based on the target reference gait features of the target person, a target gait feature of the target person is retrieved from a pre-established gait feature library to obtain a gait retrieval result; the similarity between the target gait feature of the target person and the target reference gait feature meets the preset similarity conditions; in this way, the gait features under different clothing and different angles are integrated to obtain a more comprehensive reference gait feature. This reference gait feature can more accurately represent the walking posture of the person in reality, thereby more accurately retrieving the gait feature and performing identity recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 A flowchart of the multi-sequence gait retrieval method according to an embodiment of the present application is shown;

[0040] Figure 2 A flow chart of a method for classifying a plurality of character features into a plurality of target categories based on clothing features and / or angle features according to an embodiment of the present application is shown;

[0041] Figure 3 A flow chart of a method for calculating target reference gait features of a target person according to an embodiment of the present application is shown;

[0042] Figure 4 A flow chart of a method for retrieving target gait features from a pre-established gait feature library according to an embodiment of the present application is shown;

[0043] Figure 5 A schematic structural diagram of a multi-sequence gait retrieval device according to an embodiment of the present application is shown;

[0044] Figure 6 A schematic structural diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of illustration and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps without logical context can be reversed or implemented simultaneously. In addition, those skilled in the art, under the guidance of the contents of this application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0046] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0047] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0048] With increasing attention to public safety, video surveillance cameras are becoming ubiquitous in cities, and various recognition algorithms are finding widespread application in areas such as safe production, intelligent security, and intelligent traffic management. Traditional identity recognition methods require the use of additional information, such as high-definition facial images. However, because facial recognition involves sensitive facial image information and is limited by factors such as close proximity, angle deviation, lighting, and partial occlusion, the performance of identity recognition systems in surveillance scenarios is severely impacted. To address this issue, research is developing gait recognition systems that utilize walking posture as identification information for medium- and long-range identification.

[0049] However, identifying a person based solely on a single gait sequence cannot accurately represent their walking posture. This is because in practice, a person's walking posture is affected by complex and variable internal and external factors. For example, whether a person is carrying or not carrying a load, wearing winter or summer clothing, having hands in pockets or not, walking alone or holding hands, all affect their walking posture. Consequently, identifying a person based on a single gait sequence has low accuracy.

[0050] Based on this, an embodiment of the present application proposes a multi-sequence gait retrieval method, device, electronic device and medium, wherein the multi-sequence gait retrieval method can obtain multiple character features of a target person; wherein the character features include clothing features, angle features and gait features; the angle features represent the shooting angle in the character image; based on the clothing features and / or angle features, the multiple character features are divided into multiple target categories; wherein the clothing features and / or angle features in the character features in each target category meet the preset classification conditions corresponding to the target category; based on at least part of the gait features of each target category of the target person, the target reference gait features of the target person are calculated; based on the target reference gait features of the target person, the target gait features of the target person are retrieved from a pre-established gait feature library to obtain a gait retrieval result; the similarity between the target gait features of the target person and the target reference gait features meets the preset similarity conditions; in this way, the gait features under different clothing and different angles are integrated to obtain a more comprehensive reference gait feature, which can more accurately represent the walking posture of the person in reality, thereby more accurately retrieving the gait features and performing identity recognition.

[0051] Please refer to Figure 1 , Figure 1 The flowchart of the multi-sequence gait retrieval method according to an embodiment of the present application is shown; specifically, the multi-sequence gait retrieval method includes the following steps S101-S104:

[0052] S101, obtaining multiple character features of a target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle in the character image;

[0053] S102, classifying the plurality of character features into a plurality of target categories based on clothing features and / or angle features; wherein the clothing features and / or angle features of the character features in each target category meet preset classification conditions corresponding to the target category;

[0054] S103, calculating a target reference gait feature of the target person based on at least part of the gait features of each target category of the target person;

[0055] S104. According to the target reference gait feature of the target person, the target gait feature of the target person is retrieved from a pre-established gait feature library to obtain a gait retrieval result; the similarity between the target gait feature of the target person and the target reference gait feature meets a preset similarity condition.

[0056] The multi-sequence gait retrieval method described in the embodiment of the present application first classifies gait features according to clothing and angle, and then fuses gait features of different categories to obtain a more comprehensive reference gait feature. This reference gait feature covers gait features under the influence of multiple factors and can more accurately represent the walking posture of a person in reality, thereby more accurately retrieving gait features and performing identity recognition.

[0057] In step S101, a plurality of character features of a target person are obtained; wherein the character features include clothing features, angle features, and gait features; and the angle features represent the shooting angle of view in the character image.

[0058] Here, the target person's multiple features are extracted from multiple images of the person. The images include location attributes and time attributes. For example, the multiple images of the person were captured by camera 1 on road A; therefore, the images have location attributes. Each of the multiple images of the person has a capture time, therefore, the images have time attributes.

[0059] In this way, the character features extracted based on the character image also have time attributes and location attributes.

[0060] The character features include clothing features, angle features, and gait features. That is, the clothing features, angle features, and gait features correspond to each other and together represent the characteristics of the target person.

[0061] In the embodiment of the present application, the clothing features and gait features are pedestrian re-identification features (ReID features) extracted using Multiple Granularity Network (MGN).

[0062] The angle feature, or the viewing angle of the person image, is specifically the angle between the person's orientation in the image and the camera's optical axis, or the angle between the person's walking path in the image and a horizontal vector, where the horizontal vector is oriented from left to right. The angle ranges from [0, 360]. The orientation of a person in an image taken from different viewing angles varies, for example, from the front, side, or back. Accordingly, the same person's walking posture also varies from different viewing angles. Therefore, angle features can influence gait characteristics; gait characteristics vary from different viewing angles.

[0063] The clothing features refer to the clothing types of the person in the person image, such as skirt, pants, sportswear, high heels, sneakers, down jacket, etc. Clothing also affects the walking posture of the person. For example, the walking posture is different when wearing high heels and sneakers, and the walking posture is different when wearing a bulky down jacket and a light sportswear.

[0064] Based on this, in an embodiment of the present application, character features including clothing features, angle features and gait features are obtained to obtain gait features under multiple influencing factors. This gait feature is more representative and can more widely characterize the walking posture of the target person.

[0065] In step S102, the multiple character features are divided into multiple target categories based on clothing features and / or angle features; wherein the clothing features and / or angle features in the character features in each target category meet the preset classification conditions corresponding to the target category.

[0066] That is, the clothing features and / or angle features in each object category are similar.

[0067] For details, please refer to Figure 2 The method of classifying the plurality of character features into a plurality of target categories based on clothing features and / or angle features comprises the following steps S201-S202:

[0068] S201, determining at least one target classification strategy from a plurality of preset classification strategies based on attributes of a plurality of character features; wherein the attributes of the plurality of character features include one of the following: shooting locations of the plurality of character images, shooting times of the plurality of character images; the character images are images from which character features are extracted;

[0069] S202: Classify the multiple character features into multiple categories according to the target classification strategy.

[0070] In an embodiment of the present application, the preset classification strategies include the following four: classify the clothing first, and then classify the character features under each clothing by angle; classify the angle first, and then classify the clothing by the character features under each angle; classify only the clothing; and classify only the angle.

[0071] That is to say, classification according to clothing features and classification according to angle features can be performed separately or simultaneously.

[0072] At least one target classification strategy is determined from the preset classification strategies based on the time attributes and location attributes of the plurality of character features.

[0073] The time attribute is a time identifier, and the location attribute is a location identifier, such as a camera number.

[0074] In an embodiment of the present application, multiple character features are extracted from multiple character images. For example, when multiple character images are taken by a camera, it can be inferred that their shooting angles are relatively similar. In this case, only the clothing can be classified, or the clothing can be classified first and then the angle can be classified.

[0075] Alternatively, the multiple character images are taken by a camera of a company that has a unified dress code. The clothing styles in the multiple character images are rarely or even completely the same. In this case, only the angles can be classified, or the angles can be classified first and then the clothing can be classified.

[0076] Alternatively, the multiple character images are taken continuously by a camera on road A within half an hour. The clothing of the characters in the multiple character images is highly likely to remain unchanged, and only the angles may be classified.

[0077] That is, at least one target classification strategy may be determined based on the number of clothing categories and angle categories in the plurality of character features. The number of clothing categories and angle categories may be determined based on the location attributes and time attributes of the plurality of character features.

[0078] In some embodiments, the number of clothing categories and angle categories may also be determined or predicted according to other methods, such as roughly determining them manually based on images of people.

[0079] Based on this, in an embodiment of the present application, according to the target classification strategy, the multiple character features are divided into multiple target categories, which are at least one of the following:

[0080] Classifying clothing features among the plurality of character features, determining clothing categories according to the classification results, and classifying the plurality of character features into a plurality of target categories according to the clothing categories;

[0081] Alternatively, the angle features among the plurality of character features are classified, angle categories are determined according to the classification results, and the plurality of character features are classified into a plurality of target categories according to the angle categories;

[0082] Alternatively, the angle features among the plurality of character features are first classified, and angle categories are determined based on the classification results. The plurality of character features are then classified into a plurality of first categories based on the angle categories. The clothing features of each first category are then classified, and clothing categories are determined based on the classification results. The character features of each first category are then classified into at least one target category based on the clothing categories.

[0083] Alternatively, the clothing features among the multiple character features are first classified, the clothing category is determined based on the classification results, the multiple character features are divided into multiple first categories based on the clothing category, and then the angle features of each first category are classified, the angle category is determined based on the classification results, and the character features of each first category are divided into at least one target category based on the angle category.

[0084] Specifically, the clothing features among the multiple character features are classified, and the clothing category is determined according to the classification results, including:

[0085] Performing cluster analysis on the clothing features among the plurality of character features according to a preset cluster analysis algorithm to obtain at least one cluster;

[0086] Each cluster is considered as a clothing category.

[0087] In an embodiment of the present application, unsupervised clustering is performed on the clothing features among multiple character features. The unsupervised clustering method here can adopt the DBSCAN algorithm. DBSCAN is a density-based clustering method that automatically completes the clustering operation according to the distribution of ReID features in high-dimensional space.

[0088] After clustering is complete, the characters with the same clothing category will be assigned the same label, and the characters with different clothing categories will be assigned different labels. In other words, the characters with the same clothing will be assigned the same label, and the characters with different clothing will be assigned different labels.

[0089] The perspective classification model uses the ResNet18 model as the backbone network, and labels the training set data with angle categories of every 30 degrees (class-0: [0,30), class-1: [30,60), class-2: [60,90), ...), and trains the perspective classification model through the training set to obtain a trained perspective classification model.

[0090] The viewing angle categories may also be 15 degrees per category, or 60 degrees per category, and so on.

[0091] In step S103 , a target reference gait feature of the target person is calculated based on at least part of the gait features of each target category of the target person.

[0092] That is to say, the target reference gait features of the target person can be calculated based on all the gait features in each target category, or the gait features in each target category can be screened twice, and the target reference gait features of the target person can be calculated based on the gait features screened twice.

[0093] The secondary screening, for example, can eliminate abnormal clothing features that are far away from the center point in each cluster during clothing clustering, so as to eliminate the gait features corresponding to the abnormal clothing features.

[0094] For details, please refer to Figure 3 , based on at least part of the gait features of each target category of the target person, calculating the target reference gait features of the target person, including the following steps S301-S302:

[0095] S301, calculating a category reference gait feature of each target category of the target person based on at least part of the gait feature of the target category;

[0096] S302: Calculate target reference gait features of the target person based on the category reference gait features of each target category.

[0097] That is to say, the gait features of each target category are first fused once to obtain the category reference gait features of the target category, and then the category reference gait features of each target category are fused twice to obtain the target reference gait features that cover the walking posture under multiple influencing factors.

[0098] In some embodiments, in step S301, when the first target category is determined based on clothing features and angle features, calculating the category reference gait features of each target category of the target person based on at least part of the gait features of the target person includes:

[0099] Fuse the gait features of the same angle and the same clothing category in the target category to obtain the first fused gait feature corresponding to the angle;

[0100] The first step features from different angles are fused to obtain the category reference gait features of the target category.

[0101] That is to say, in a fusion process, we can go one step further and first fuse the gait features of the same angle and the same clothing category to obtain the first fused gait features corresponding to the angle; then fuse the first gait features of different angles to obtain the category reference gait features of the target category.

[0102] In the embodiment of the present application, the reason why the gait features of the same angle are first fused, then the gait features of the same category are fused, and finally the gait features of different categories are fused is to ensure that the influence of various types of influencing factors on the gait features is evenly reflected in the target reference gait features of the target person.

[0103] For example, if there are 10 character features, 9 of which are wearing high heels and 1 is wearing sneakers, if they are directly fused, the high heels will have too much influence on the target reference gait feature, and the target reference gait feature basically cannot cover the walking posture of the target person when wearing sneakers.

[0104] In the embodiment of the present application, specifically, the category reference gait feature is the mean of the gait features of the target category, the target reference gait feature is the mean of the category reference gait features, and the first gait feature is the mean of the gait features of the same angle and the same clothing category.

[0105] The specific process of calculating the target reference gait features of the target person corresponds to various preset classification strategies.

[0106] The target classification strategy is: first classify the clothing, then classify the character features under each clothing by angle. The specific process of calculating the target reference gait features of the target character is:

[0107] a. Average the gait features of the same angle category under each clothing category;

[0108] b. Average the average features of different angle categories under each clothing category;

[0109] c. The average features of different clothing categories are averaged again to obtain the target reference gait features.

[0110] The target classification strategy is: first classify the angles, then classify the clothing based on the character features at each angle. The specific process of calculating the target reference gait features of the target character is as follows:

[0111] a. Average the gait characteristics of the same clothing category under each angle category;

[0112] b. Average the average features of different clothing categories under each angle category;

[0113] c. The average features of different angle categories are averaged again to obtain the target reference gait features.

[0114] Based on this, in some embodiments, when classification is performed based on clothing features and angle features, the target reference gait features of the target person are calculated based on the category reference gait features of each target category, including:

[0115] First, the category reference gait features of the same angle category or clothing category are fused to obtain the fused category reference gait features;

[0116] Then, the target reference gait features of the target person are calculated based on the fused category reference gait features.

[0117] The target classification strategy is: when only clothing is classified, the specific process of calculating the target reference gait features of the target person is:

[0118] a. Average the gait characteristics under each clothing category;

[0119] b. The average features of different clothing categories are averaged again to obtain the target reference gait features.

[0120] The target classification strategy is: when only angle classification is performed, the specific process of calculating the target reference gait features of the target person is as follows:

[0121] a. Average the gait features under each angle category;

[0122] b. The average features of different angle categories are averaged again to obtain the target reference gait features.

[0123] In step S104, the target gait feature is retrieved from a pre-established gait feature library based on the target reference gait feature of the target person to obtain a gait retrieval result; the similarity between the target gait feature and the target reference gait feature meets a preset similarity condition.

[0124] In the embodiment of this application, please refer to Figure 4 , according to the target reference gait feature of the target person, retrieving the target gait feature in a pre-established gait feature library; comprising the following steps S401-S402:

[0125] S401, respectively calculating the similarity between the gait features that meet the screening conditions in the gait feature library and the target reference gait features of the target person;

[0126] S402: Filter out target gait features whose similarity meets a preset similarity condition again from the gait features that meet the screening condition.

[0127] At least some of the gait features in the gait feature library are gait features selected from all gait features in the gait feature library based on a selection operation. For example, if the target person is from City A, the gait features of City A are selected from the gait feature library. If the target person is found to be male, the gait features of males are selected from the gait feature library.

[0128] In some embodiments, the similarity between each gait feature in the gait feature library and the target reference gait feature may also be directly calculated. Here, the screening condition is all gait features.

[0129] In contrast, calculating the similarity between the gait features that meet the screening conditions in the gait feature library and the target reference gait features of the target person can reduce the amount of calculation for a single retrieval.

[0130] In the embodiment of the present application, a single gait feature in the gait feature library is used represents, and d is the dimension of gait features.

[0131] In the embodiment of the present application, cosine similarity is used for measurement, and the similarity calculation can be expressed by the following formula:

[0132]

[0133] Among them, d is the dimension of gait feature, p i The i-th dimension of p is used to represent the target reference gait feature; g i represents the i-th dimension of a single gait feature g in the gait feature library, and s represents the similarity between the gait feature in the gait feature library and the target reference gait feature.

[0134] In the embodiment of the present application, the preset similarity condition is a preset similarity threshold.

[0135] Target gait features whose similarity meets a preset similarity condition are screened out again from the gait features that meet the screening conditions. Specifically, target gait features whose similarity is greater than a preset similarity threshold are screened out again from the gait features that meet the screening conditions.

[0136] Exemplarily, target reference gait features with s>0.7 are screened out.

[0137] The target gait features are retrieved from a pre-established gait feature library to obtain a gait retrieval result. In an embodiment of the present application, the target reference gait features are sorted in descending order of similarity to obtain sorted target reference gait features; the sorted target reference gait features are used as the gait retrieval result.

[0138] When determining the identity of the target person based on the gait retrieval results, the identity of the target person can be determined preferentially based on the target gait features with high similarity.

[0139] Based on the same inventive concept, the embodiment of the present application also provides a multi-sequence gait retrieval device corresponding to the multi-sequence gait retrieval method. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the above-mentioned multi-sequence gait retrieval method in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0140] Please refer to Figure 5 , Figure 5 The structure diagram of the multi-sequence gait retrieval device according to an embodiment of the present application is shown. Specifically, the multi-sequence gait retrieval device includes:

[0141] The acquisition module 501 is used to acquire multiple character features of the target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle in the character image;

[0142] A classification module 502 classifies the plurality of character features into a plurality of target categories based on clothing features and / or angle features; wherein the clothing features and / or angle features of the character features in each target category meet a preset classification condition corresponding to the target category;

[0143] A calculation module 503 calculates a target reference gait feature of the target person based on at least part of the gait feature of each target category of the target person;

[0144] The retrieval module 504 retrieves the target gait feature from a pre-established gait feature library based on the target reference gait feature of the target person to obtain a gait retrieval result; the similarity between the target gait feature and the target reference gait feature meets a preset similarity condition.

[0145] The multi-sequence gait retrieval device proposed in the embodiment of the present application can obtain multiple character features of a target person; wherein, the character features include clothing features, angle features and gait features; the angle features represent the shooting angle in the character image; based on the clothing features and / or angle features, the multiple character features are divided into multiple target categories; wherein, the clothing features and / or angle features in the character features in each target category meet the preset classification conditions corresponding to the target category; based on at least part of the gait features of each target category of the target person, the target reference gait features of the target person are calculated; according to the target reference gait features of the target person, the target gait features of the target person are retrieved from a pre-established gait feature library to obtain a gait retrieval result; the similarity between the target gait features of the target person and the target reference gait features meets the preset similarity conditions; in this way, the gait features under different clothing and different angles are fused to obtain a more comprehensive reference gait feature, which can more accurately represent the walking posture of the actual person, thereby more accurately retrieving the gait features and performing identity recognition.

[0146] In some embodiments, in the multi-sequence gait retrieval device, the classification module, when classifying the multiple person features into multiple target categories based on clothing features and / or angle features, is specifically configured to:

[0147] Determining at least one target classification strategy from a plurality of preset classification strategies based on attributes of a plurality of character features; wherein the attributes of the plurality of character features include one of the following: shooting locations of the plurality of character images, shooting times of the plurality of character images; the character images are images from which the character features are extracted;

[0148] According to the target classification strategy, the multiple character features are divided into multiple categories.

[0149] In some embodiments, in the multi-sequence gait retrieval device, the classification module, when classifying the multiple human features into multiple target categories according to the target classification strategy, specifically performs one of the following operations:

[0150] Classifying clothing features among the plurality of character features, determining clothing categories according to the classification results, and classifying the plurality of character features into a plurality of target categories according to the clothing categories;

[0151] Alternatively, the angle features among the plurality of character features are classified, angle categories are determined according to the classification results, and the plurality of character features are classified into a plurality of target categories according to the angle categories;

[0152] Alternatively, the angle features among the plurality of character features are first classified, and angle categories are determined based on the classification results. The plurality of character features are then classified into a plurality of first categories based on the angle categories. The clothing features of each first category are then classified, and clothing categories are determined based on the classification results. The character features of each first category are then classified into at least one target category based on the clothing categories.

[0153] Alternatively, the clothing features among the multiple character features are first classified, the clothing category is determined based on the classification results, the multiple character features are divided into multiple first categories based on the clothing category, and then the angle features of each first category are classified, the angle category is determined based on the classification results, and the character features of each first category are divided into at least one target category based on the angle category.

[0154] In some embodiments, in the multi-sequence gait retrieval device, the classification module, when classifying clothing features among a plurality of character features and determining clothing categories based on the classification results, is specifically configured to:

[0155] Performing cluster analysis on the clothing features among the plurality of character features according to a preset cluster analysis algorithm to obtain at least one cluster;

[0156] Each cluster is considered as a clothing category.

[0157] In some embodiments, in the multi-sequence gait retrieval device, the classification module, when classifying the angle features among the multiple character features, is specifically configured to:

[0158] Inputting the plurality of character features into a trained viewpoint classification model;

[0159] The angle features in the plurality of character features are processed by the perspective classification model to obtain a perspective label for each angle feature; wherein, angle features with the same perspective label have the same angle category.

[0160] In some embodiments, in the multi-sequence gait retrieval device, the calculation module, when calculating the target reference gait features of the target person based on at least part of the gait features of each target category of the target person, is specifically configured to:

[0161] Calculating a class reference gait feature for each target class based on at least a portion of the gait features of the target person;

[0162] Based on the class reference gait features of each target class, a target reference gait feature of the target person is calculated.

[0163] In some embodiments, in the multi-sequence gait retrieval device, the retrieval module, when retrieving the target gait feature from a pre-established gait feature library based on the target reference gait feature of the target person, is specifically configured to:

[0164] Calculating the similarity between the gait features that meet the screening conditions in the gait feature library and the target reference gait features of the target person respectively;

[0165] Target gait features whose similarity meets the preset similarity conditions are again screened out from the gait features that meet the screening conditions.

[0166] Based on the same inventive concept, an electronic device corresponding to the multi-sequence gait retrieval method is also provided in the embodiment of the present application. Since the principle of solving the problem by the electronic device in the embodiment of the present application is similar to the above-mentioned multi-sequence gait retrieval method in the embodiment of the present application, the implementation of the electronic device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0167] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of the electronic device described in an embodiment of the present application is shown. Specifically, the electronic device 600 includes: a processor 602, a memory 601 and a bus. The memory 601 stores machine-readable instructions executable by the processor 602. When the electronic device 600 is running, the processor 602 communicates with the memory 601 through the bus. When the machine-readable instructions are executed by the processor 602, the steps of the multi-sequence gait retrieval method are performed.

[0168] Based on the same inventive concept, a computer-readable storage medium corresponding to the multi-sequence gait retrieval method is also provided in the embodiment of the present application. Since the principle of solving the problem by the computer-readable storage medium in the embodiment of the present application is similar to the above-mentioned multi-sequence gait retrieval method in the embodiment of the present application, the implementation of the computer-readable storage medium can refer to the implementation of the method, and the repeated parts will not be repeated.

[0169] A computer-readable storage medium stores a computer program, which executes the steps of the multi-sequence gait retrieval method when executed by a processor.

[0170] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0171] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0172] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0173] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, platform server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0174] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A multi-sequence gait retrieval method, characterized in that: include: Acquire multiple character features of the target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle of the character image; Based on clothing features and / or angle features, the plurality of character features are divided into a plurality of target categories; wherein the clothing features and / or angle features in the character features of each target category meet the preset classification conditions corresponding to the target category; Calculating a target reference gait feature of the target person based on at least part of the gait feature of each target category of the target person; According to the target reference gait feature of the target person, the target gait feature is retrieved from a pre-established gait feature library to obtain a gait retrieval result; the similarity between the target gait feature and the target reference gait feature meets a preset similarity condition.

2. The multi-sequence gait retrieval method according to claim 1, characterized in that: The method of classifying the plurality of character features into a plurality of target categories based on clothing features and / or angle features includes: Determining at least one target classification strategy from a plurality of preset classification strategies based on attributes of a plurality of character features; wherein the attributes of the plurality of character features include one of the following: shooting locations of the plurality of character images, shooting times of the plurality of character images; the character images are images from which the character features are extracted; According to the target classification strategy, the multiple character features are divided into multiple categories.

3. The multi-sequence gait retrieval method according to claim 2, characterized in that: According to the target classification strategy, the multiple character features are classified into multiple target categories, at least one of the following: Classifying clothing features among the plurality of character features, determining clothing categories according to the classification results, and classifying the plurality of character features into a plurality of target categories according to the clothing categories; Alternatively, the angle features among the plurality of character features are classified, angle categories are determined according to the classification results, and the plurality of character features are classified into a plurality of target categories according to the angle categories; Alternatively, the angle features among the plurality of character features are first classified, and angle categories are determined based on the classification results. The plurality of character features are then classified into a plurality of first categories based on the angle categories. The clothing features of each first category are then classified, and clothing categories are determined based on the classification results. The character features of each first category are then classified into at least one target category based on the clothing categories. Alternatively, the clothing features among the multiple character features are first classified, the clothing category is determined based on the classification results, the multiple character features are divided into multiple first categories based on the clothing category, and then the angle features of each first category are classified, the angle category is determined based on the classification results, and the character features of each first category are divided into at least one target category based on the angle category.

4. The multi-sequence gait retrieval method according to claim 3, characterized in that: Classify clothing features from multiple character features and determine clothing categories based on the classification results, including: Performing cluster analysis on the clothing features among the plurality of character features according to a preset cluster analysis algorithm to obtain at least one cluster; Each cluster is considered as a clothing category.

5. The multi-sequence gait retrieval method according to claim 3, characterized in that: Classify angle features in multiple character features, including: Inputting the plurality of character features into a trained viewpoint classification model; The angle features in the plurality of character features are processed by the perspective classification model to obtain a perspective label for each angle feature; wherein, angle features with the same perspective label have the same angle category.

6. The multi-sequence gait retrieval method according to claim 1, characterized in that: Calculating target reference gait features of the target person based on at least part of the gait features of each target category of the target person includes: Calculating a class reference gait feature for each target class based on at least a portion of the gait features of the target person; Based on the class reference gait features of each target class, a target reference gait feature of the target person is calculated.

7. The multi-sequence gait retrieval method according to claim 1, characterized in that: According to the target reference gait feature of the target person, the target gait feature is retrieved from a pre-established gait feature library; including: Calculating the similarity between the gait features that meet the screening conditions in the gait feature library and the target reference gait features of the target person respectively; Target gait features whose similarity meets the preset similarity conditions are again screened out from the gait features that meet the screening conditions.

8. A multi-sequence gait retrieval device, characterized in that: include: An acquisition module is used to acquire multiple character features of a target person; wherein the character features include clothing features, angle features, and gait features; the angle features represent the shooting angle in the character image; a classification module, configured to classify the plurality of character features into a plurality of target categories based on clothing features and / or angle features; wherein the clothing features and / or angle features of the character features in each target category satisfy a preset classification condition corresponding to the target category; a calculation module, configured to calculate a target reference gait feature of the target person based on at least a portion of the gait features of each target category of the target person; The retrieval module is used to retrieve the target gait feature from a pre-established gait feature library based on the target reference gait feature of the target person to obtain a gait retrieval result; the similarity between the target gait feature and the target reference gait feature meets the preset similarity condition.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the multi-sequence gait retrieval method according to any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the multi-sequence gait retrieval method according to any one of claims 1 to 7.

Images