A target object recognition method, device, equipment and storage medium

Abstract

The present disclosure provides a target object recognition method, device and equipment, and a storage medium, wherein the method comprises: acquiring a target video, and recognizing the motion trajectory of one or more target objects in the target video; extracting a plurality of attribute features of the target object from the motion trajectory of the target object, and constructing a heterogeneous graph between the attribute features of the target object and pre-stored base library features; wherein the attribute features and the base library features are nodes in the heterogeneous graph, and some nodes have an initial connection relationship; updating the initial connection relationship in the heterogeneous graph, determining the matching result of each attribute feature of the target object in the base library features based on the connection relationship between the updated nodes, and recognizing the target object according to the matching result. The technical solution provided by the present disclosure can improve the recognition accuracy of the target object.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and specifically to a method, apparatus, device, and storage medium for identifying target objects. Background Technology

[0002] In some practical applications, it is often necessary to identify target objects in images or videos. For example, target objects can currently be identified using one-stage or two-stage object detection algorithms.

[0003] Due to the variability of real-world application environments and the inconsistent quality of video frames, the accuracy of target object recognition can be low. Therefore, a more accurate method for target object recognition is needed. Summary of the Invention

[0004] In view of this, one or more embodiments of the present disclosure provide a method, apparatus, device and storage medium for identifying target objects, which can improve the accuracy of target object identification.

[0005] This disclosure provides a method for identifying a target object. The method includes: acquiring a target video and identifying the motion trajectories of one or more target objects in the target video; extracting multiple attribute features of the target objects from their motion trajectories and constructing a heterogeneous graph between the attribute features of the target objects and a pre-stored database of features; wherein the attribute features and the database of features serve as nodes in the heterogeneous graph, and some nodes have initial connections; updating the initial connections in the heterogeneous graph, and based on the updated connections between nodes, determining the matching results of each attribute feature of the target object in the database of features, and identifying the target object based on the matching results.

[0006] This disclosure also provides a target object identification device, the device comprising: a trajectory identification unit, configured to acquire a target video and identify the motion trajectory of one or more target objects in the target video; a heterogeneous graph construction unit, configured to extract multiple attribute features of the target objects from the motion trajectories of the target objects and construct a heterogeneous graph between the attribute features of the target objects and a pre-stored base database feature; wherein the attribute features and the base database feature serve as nodes in the heterogeneous graph, and some nodes have initial connection relationships; and an object identification unit, configured to update the initial connection relationships in the heterogeneous graph, and based on the updated connection relationships between nodes, determine the matching results of each attribute feature of the target object in the base database feature, and identify the target object according to the matching results.

[0007] This disclosure also provides an electronic device including a memory and a processor, the memory being used to store a computer program, which, when executed by the processor, implements the above-described method for identifying the target object.

[0008] This disclosure also provides a computer-readable storage medium for storing a computer program that, when executed by a processor, implements the above-described method for identifying a target object.

[0009] This disclosure provides a technical solution through one or more embodiments, which employs a heterogeneous graph construction method to establish a correlation between the attribute features of a target object and pre-stored base database features. The attribute features of the target object are extracted from its motion trajectory and can describe the target object from multiple different aspects. The base database features can serve as standard features of the target object, and the heterogeneous graph can characterize the initial matching relationship between the attribute features and the base database features.

[0010] Subsequently, by updating the initial connection relationships between nodes in the heterogeneous graph, the matching relationship between attribute features and base library features can be accurately obtained. Based on this matching relationship, the target object can be identified.

[0011] As can be seen, one or more embodiments of this disclosure can analyze the motion trajectory of the target object when identifying it, thereby avoiding the problem of inaccurate identification results caused by poor quality of a single video frame. Simultaneously, multiple attribute features can comprehensively describe the target object from multiple perspectives, and heterogeneous graphs can be used to accurately match multiple attribute features with pre-stored base library features, thereby improving the accuracy of target object identification. Attached Figure Description

[0012] The features and advantages of the embodiments of this disclosure will be more clearly understood by referring to the accompanying drawings, which are illustrative and should not be construed as limiting the present disclosure in any way. In the drawings:

[0013] Figure 1 A schematic diagram of the steps of a target object identification method in one embodiment of this disclosure is shown;

[0014] Figure 2 A schematic diagram of a heterogeneous graph is shown in one embodiment of this disclosure;

[0015] Figure 3 A schematic diagram of an updated heterogeneous graph is shown in one embodiment of this disclosure;

[0016] Figure 4 A flowchart illustrating the identification of a target object is shown in one embodiment of this disclosure;

[0017] Figure 5 A schematic diagram of the functional modules of a target object identification device in one embodiment of this disclosure is shown;

[0018] Figure 6 A schematic diagram of the structure of an electronic device according to one embodiment of the present disclosure is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0020] In scenarios requiring target object identification, factors such as the target object's high speed or similar clothing can lead to low identification accuracy. For example, at a ski resort, skiers typically move at high speeds, resulting in blurry figures in some captured images. Identifying skiers based on blurry images leads to low accuracy. Furthermore, skiers usually change into ski suits upon entering the resort, and since these suits are often similar in style, identifying skiers solely based on suit style will also result in low accuracy.

[0021] Therefore, this disclosure provides a method for identifying target objects. This method can generate multiple attribute features of different types related to the target object. These attribute features can comprehensively describe the target object from different perspectives, providing a data foundation for improving identification accuracy. Subsequently, heterogeneous graphs can be used to accurately match the attribute features with the base database features, thereby improving the identification accuracy of the target object.

[0022] Please see Figure 1 The target object identification method provided in one embodiment of this disclosure may include the following steps.

[0023] S1: Acquire the target video and identify the motion trajectory of one or more target objects in the target video.

[0024] In this embodiment, the target object may be the user to be identified. It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, the user should be informed of the type, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and the user's authorization should be obtained.

[0025] For example, before implementing the technical solutions of the various embodiments of this disclosure, a prompt message can be sent to the user to clearly inform the user that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware such as the electronic device, application program, server, or storage medium that performs the operations of the technical solutions of this disclosure, based on the prompt message.

[0026] As an optional but not limited implementation, the prompt message can be sent to the user in the form of a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control for the user to choose "agree" or "disagree".

[0027] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0028] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0029] It is understood that in the specific implementation of this application, data such as object identifiers and video data are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0030] In this embodiment, cameras deployed in offline scenarios can capture target videos containing the target object as it moves within those scenarios. Subsequent processing of the target video to identify the target object is performed only after authorization from the target object has been obtained. In practical applications, object detection algorithms can be used to analyze the target video frame by frame. For any video frame in the target video, after processing with an object detection algorithm, the location information of one or more bounding boxes corresponding to that video frame can be output. Each bounding box can represent a human body in the video frame; by using the location information of the bounding boxes, the human body present in the video frame can be identified and located.

[0031] Furthermore, after object detection, each bounding box representing a human body will have its own object identifier, which can be, for example, a globally unique number. After performing object detection processing on different video frames of the target video, the position information of the bounding boxes and their object identifiers can be output for each video frame. Bounding boxes with the same object identifier can represent the same human body. Thus, after frame-by-frame detection of the target video, by identifying the bounding boxes with the same object identifier, the motion trajectory of the same human body can be tracked in different video frames.

[0032] In this embodiment, the target object can be any human body to be identified in the target video. In the target video, the target object may only appear in a portion of the video frames and be identified by the object detection algorithm; these frames can then be used as target video frames representing the motion trajectory of the target object. Thus, by performing frame-by-frame object detection on the target video, the motion trajectory of the target object can be determined within the target video.

[0033] It should be noted that different motion trajectories can be obtained for different target objects using the above method. Different motion trajectories can correspond to different target video frames in the target video. For example, for target object A, the motion trajectory can be represented by target video frames from second 5 to second 30 in the target video; while for target object B, the motion trajectory can be represented by target video frames from second 1 to second 20 and from second 40 to second 50 in the target video. In other words, the target video frames representing the motion trajectory can also be different for different target objects.

[0034] S3: Extract multiple attribute features of the target object from the motion trajectory of the target object, and construct a heterogeneous graph between the attribute features of the target object and the pre-stored base database features; wherein the attribute features and the base database features serve as nodes in the heterogeneous graph, and some nodes have initial connection relationships.

[0035] In this embodiment, after identifying the target video frame representing the motion trajectory of the target object from the target video, more refined feature recognition can be performed on the target object in the target video frame. Taking one target object as an example, different types of feature sequences related to the target object can be extracted from the target object's motion trajectory from different angles. For example, in a skiing scene, the above-mentioned different types of feature sequences can include at least one of the following: head image feature sequence, human body feature sequence, and clothing icon feature sequence. The number of feature sequences can be flexibly determined according to the requirements of object recognition accuracy and data processing efficiency in the actual application scenario.

[0036] For example, in a specific application example, the target video frames can be processed frame by frame, and the head image features, body features, and clothing icon features of the target object can be extracted from each target video frame. By summarizing the different types of features extracted from each target video frame, three different types of feature sequences can be obtained (namely, head image feature sequence, body feature sequence, and clothing icon feature sequence).

[0037] In one implementation, taking a feature sequence of any target type as an example, type features related to the target object and conforming to that target type can be identified from each target video frame. In practical applications, a feature extraction network can be used to extract the aforementioned type features from the target video frames. Specifically, the feature extraction network can be flexibly selected from various different network models according to the needs of the actual application scenario. These various network models may include, for example, the VGG model, the Inception model, and the ResNet model. Before inputting the target video frames into the feature extraction network, the target video frames can be cropped to different degrees based on the different target types.

[0038] For example, when the target type is an avatar, the head region of the target object can be cropped from the target video frame, and the cropped region image can be input into the feature extraction network to ensure that the features output by the feature extraction network are avatar features. Similarly, if the target type is a human body, the bounding box of the target object in the target video frame can be directly cropped, and the cropped region image can be input into the feature extraction network.

[0039] In this embodiment, after identifying the corresponding type features from each target video frame, the feature sequence composed of the identified type features can be used as the feature sequence of the target type. For different target types, their respective feature sequences can be generated in a similar manner. Therefore, the feature sequence can include multiple type features of the same type. For example, the avatar feature sequence can include the avatar features of the target object identified from each target video frame. Similarly, the clothing icon feature sequence can include the clothing icon features of the target object identified from each target video frame.

[0040] In one implementation, when extracting features from a target video frame and generating different types of type features, quality information for these type features can also be generated simultaneously. This quality information characterizes the quality of the extracted type features. In practical applications, the target object in the target video frame may be occluded or moving too quickly, resulting in less than ideal type features extracted from the target video frame. The quality information can then quantitatively evaluate the quality of each type feature. Different models can be used to generate the quality information for different type features.

[0041] For example, for avatar features, the MegaFace model can be used to output quality information for the avatar features. For human body features, a pose estimation algorithm can be used to identify the pose information corresponding to the human body features. By comparing the pose information with a standard pose, quality information corresponding to the human body features can be generated. The closer the identified pose information is to the standard pose, the better the quality information representation of the human body feature. For clothing icon features, quality information can be represented by parameters such as image contrast and image brightness. In this way, corresponding quality information can be generated for different types of features. In practical applications, by combining the feature extraction model and the quality information generation model, the quality information of each type of feature in the feature sequence can also be generated simultaneously when generating different types of feature sequences.

[0042] In this embodiment, when a target object enters an offline scene, image information conforming to preset rules can be collected with the target object's authorization. For example, when a tourist enters a ski resort, a clear frontal photo can be taken under the guidance of staff. The aforementioned preset rules can be limited to information such as the target object's clothing and posture when collecting image information. For example, when taking a frontal photo of a tourist, the tourist needs to change into ski clothing, wear corresponding ski accessories, and then be photographed in a specified posture.

[0043] In this embodiment, after acquiring image information of the target object that conforms to preset rules, the feature extraction method described in the preceding steps can be used to identify various preset type features related to the target object from the acquired image information. For example, the headshot features, body features, and clothing icon features of the target object can be extracted from the acquired image information. Since the image information is acquired under relatively ideal conditions, the various preset type features identified from the image information can serve as the base database features of the target object, which can accurately represent the target object. Simultaneously, after acquiring the base database features of the target object, the standard information of the target object can be bound to these base database features. Thus, for the same target object, multiple base database features extracted based on the target object's image information can be bound to the target object's standard information. The standard information can represent the correct object identifier of the target object, which can be a user identifier entered after confirmation by the target object during image information acquisition.

[0044] Using the methods described above, image information conforming to preset rules can be obtained for any playable object active in an offline scenario. Subsequently, various preset type features related to the playable object can be identified from the image information, and these identified type features are used as the playable object's base feature library. Simultaneously, the playable object's base feature library is bound to the playable object's standard information.

[0045] In this embodiment, after identifying multiple sets of feature sequences of different types from the target object in the target video frame, a representative feature can be determined from each feature sequence. Typically, the representative feature is the highest quality feature among the corresponding type of feature sequences. Subsequently, the object identifier of the target object can be determined based on the representative features in each feature sequence.

[0046] In one implementation, when determining the representative features of each feature sequence, taking any target feature sequence as an example, the quality information of each type of feature in the target feature sequence can be statistically analyzed. This quality information can typically be a numerical value that characterizes the quality. Based on the statistically analyzed quality information, the type feature with the best quality can be used as the representative feature of the target feature sequence. The quality can be determined by the numerical value; a larger value indicates better quality. Of course, if multiple type features with the best quality exist in the same feature sequence, one can be randomly selected from these multiple best-quality type features as the representative feature of that feature sequence.

[0047] In this embodiment, by statistically analyzing the quality information of type features, the type feature with the highest quality can be selected as the representative feature of each feature sequence. For the target object, each representative feature determined from each set of feature sequences can serve as multiple attribute features of the target object. It is evident that multiple different attribute features can correspond to different types. Taking a skiing scene as an example, the multiple attribute features extracted from the target object's movement trajectory can be the target object's avatar features, body features, and clothing icon features.

[0048] In this embodiment, after identifying multiple attribute features of the target object, each attribute feature can be matched with the corresponding type of base library features in sequence, and a heterogeneous graph between the attribute features of the target object and the pre-stored base library features can be constructed based on the matching results.

[0049] Specifically, a heterogeneous graph can include multiple nodes and connections between some of these nodes. The various attribute features of the target object, as well as the pre-stored base database features, can serve as nodes in the heterogeneous graph. Based on the matching relationships between the attribute features and the base database features, initial connections can be established between some nodes. These initial connections are reflected in the heterogeneous graph as edges connecting nodes.

[0050] In this embodiment, the initial connection relationships between some nodes in the heterogeneous graph can be divided into two cases. One case involves establishing a first initial connection relationship between nodes representing multiple attribute features of the same target object. For example... Figure 2 As shown, assuming a target object has three attribute features: avatar feature, body feature, and clothing icon feature, then in a heterogeneous graph, the nodes represented by these three attribute features can all have edges representing the initial connection relationship. Figure 2 In this context, rectangles can represent avatar features, circles can represent human body features, and triangles can represent clothing icon features. It should be noted that, to avoid... Figure 2 The chaos of the lines in Figure 2 Only a portion of the initial connections are shown in the diagram.

[0051] In a heterogeneous graph, in addition to the first initial connectivity relationship described above, for any target attribute feature among the various attribute features of the target object, target base features belonging to the same type as the target attribute feature can be determined from the pre-stored base feature library, and a second initial connectivity relationship can be established between the node represented by the target attribute feature and the nodes represented by each of the target base feature libraries. For example... Figure 2As shown, the pre-stored base features can be from multiple target objects, and these base features can be further divided into base features for avatar types, body types, and clothing icon types. Thus, for attribute features and base features of the same type, edges representing the second initial connection relationship can be established between their respective nodes.

[0052] from Figure 2 As can be seen, there can be multiple target objects. Within each target object, a first initial connection can be established between nodes representing multiple attribute features. For the attribute features of the target object, a second initial connection can be established between attribute features of the same type and base library features.

[0053] In this embodiment, in order to determine the object identifier of the target object based on the heterogeneous graph, an initial weight value can be set for each initial connection. For the first and second initial connections described above, the initial weight values ​​can be set based on different rules.

[0054] Specifically, the first initial connection relationship represents the connection relationship between the attribute features of the same target object. Therefore, a preset fixed value can be set as the initial weight value of the first initial connection relationship. In practical applications, this preset fixed value can be 1, representing the highest correlation between the various attribute features of the target object.

[0055] The initial weight values ​​for the second initial connection relationship can be set based on the similarity between the attribute feature and the base database features. Specifically, taking any target attribute feature among the various attribute features of the target object as an example, target base database features belonging to the same type as the target attribute feature can be determined from the pre-stored base database features. For example, assuming the type of the target attribute feature is avatar type, then during feature matching, only the target attribute feature will be matched with base database features representing avatar features, and not with base database features representing human body features.

[0056] In practical applications, the matching process between attribute features and base database features can be performed by calculating similarity. Specifically, the similarity between the target attribute feature and each target base database feature of the same type can be calculated. Both the target attribute feature and the target base database feature can be represented by vectors, so the similarity between them can be represented by the Euclidean distance between the two vectors. The closer the Euclidean distance, the more similar the two features are.

[0057] In this embodiment, by means of the above method, for target attribute features and target base database features that have a second initial connection relationship, the similarity between the target attribute features and the target base database features can be calculated, and the similarity can be set as the initial weight value of the second initial connection relationship.

[0058] like Figure 2 As shown, the numerical values ​​marked on the edges between nodes represent the initial weight values ​​mentioned above. It can be seen that the initial weight value of the edges between attribute features of the same target object can be set to 1; while the initial weight value of the edges between attribute features and base database features of the same type can be set to the similarity between the attribute feature and the base database feature.

[0059] Using the above method, a heterogeneous graph between attribute features and base database features can be established. In the heterogeneous graph, attribute features and base database features can be used as nodes, and some nodes can have initial connection relationships, which are also bound to initial weight values.

[0060] S5: Update the initial connection relationship in the heterogeneous graph, and based on the updated connection relationship between nodes, determine the matching result of each attribute feature of the target object in the base database features, and determine the object identifier of the target object according to the matching result.

[0061] In this embodiment, the initial connections between nodes in the heterogeneous graph, and the initial weights bound to those connections, can initially reflect the degree of correlation between different features. However, the initial connections and initial weights determined in this way may not be consistent with the actual degree of correlation between features. For example, in... Figure 2 In the initial model, the avatar feature of the first target object has initial connections with various base library features of the same type, and after similarity calculation, three matching results with similarity scores of 0.8, 0.7, and 0.2 are obtained. However, in practical applications, this avatar feature usually corresponds to only one base library feature, and not multiple different base library features simultaneously. Furthermore, this avatar feature often does not correspond to multiple base library features with high similarity scores.

[0062] Therefore, in order to accurately represent the degree of association between nodes through edge weights, the initial weights of the initial connections in the heterogeneous graph can be updated. Based on the updated weights, the initial connections in the heterogeneous graph can be redefined, ensuring that for any given attribute feature, there is at most one underlying feature with a connection in the updated heterogeneous graph. In other words, based on the updated weights, some of the original initial connections in the heterogeneous graph can be removed, and the resulting heterogeneous graph with these removed connections is the updated heterogeneous graph.

[0063] To achieve the above goals, in practical applications, GNN (Graph Neural Networks) algorithms, or variants of GNN algorithms such as GraphSAGE (SAmple and aggreGatE), can be used to update heterogeneous graphs. Specifically, the features (attribute features or base library features) of each node in the heterogeneous graph, the initial connections between nodes, and the initial weights bound to each initial connection can be used as input data for the algorithm. When processing the input data, the GNN algorithm or its variants can re-aggregate the features of each node, thereby redetermining the degree of association between nodes. After processing by the GNN algorithm or its variants, the re-aggregated heterogeneous graph data can be output. Specifically, in the re-aggregated heterogeneous graph, the initial weights can be updated according to the aggregation results.

[0064] In this embodiment, after updating the initial weight values ​​in the heterogeneous graph, the connection relationships between nodes in the heterogeneous graph can be further modified based on the updated weight values ​​to ensure that for the same attribute feature, there is at most one base library feature with a connection relationship.

[0065] In practical applications, a weighted bipartite graph matching algorithm can be used to update the connections between nodes in a heterogeneous graph. After processing by the weighted bipartite graph matching algorithm, a set of edges is obtained, where the two vertices of each edge can be located in the set of attribute features and the set of base features, respectively, and no two edges have the same vertex. Thus, after processing by the weighted bipartite graph matching algorithm, some of the original initial connections in the heterogeneous graph can be canceled. For example, in... Figure 2 In the original process, the avatar features of the first target object had initial connections with all the base database features representing avatar types. However, after processing by the weighted bipartite graph matching algorithm, only the connection with the largest weight value was retained for this avatar feature (e.g., ...). Figure 3 As shown in the figure), and for the base database features connected to this avatar feature, there are no other connected avatar features of target objects. It should be noted that, in order to avoid... Figure 3 The chaos of the lines in Figure 3 Only a portion of the connections are shown in the image.

[0066] The above method can be used to update the initial weight values ​​of the initial connections in the heterogeneous graph, and based on the updated weight values, some initial connections in the heterogeneous graph can be canceled, so that for any attribute feature, there is at most one base library feature with a connection in the updated heterogeneous graph.

[0067] In this embodiment, after updating the heterogeneous graph, the matching results of each attribute feature of the target object in the base database feature can be determined based on the updated heterogeneous graph. Specifically, for any target attribute feature among the attribute features of the target object, a target base database feature that has a connection relationship with the target attribute feature can be identified in the updated heterogeneous graph, and this target base database feature is used as the base database feature that matches the target attribute feature.

[0068] Of course, in some scenarios, the target attribute features may not have underlying features with connectivity relationships in the updated heterogeneous graph. For example, in... Figure 3 In the second target object's human features, there are no corresponding features in the database with a connection relationship. This indicates that the recognition results based on the target video cannot associate the attribute features with the pre-stored database features. The reason may be that the target object did not have corresponding database features stored beforehand. In other words, the tourist in the offline scene may not have initially entered database features, or the tourist may not have granted permission to match database features. Therefore, based on the captured target video, no matching database features can be found. Alternatively, the attribute features identified from the target video may not be accurate enough. For example, if the target object's image in the target video is too blurry, the identified attribute features will be inaccurate, making it impossible to match with the pre-stored database features.

[0069] As described above, for each attribute feature of a target object, each attribute feature will match at most one base database feature. Therefore, when determining the object identifier of a target object, we can first check whether each attribute feature of the target object has a matching base database feature. If each representative feature has its own matching base database feature, then we can collect the standard information of each matching base database feature, and then determine the object identifier of the target object based on the collected standard information.

[0070] Please see Figure 4 Specifically, if the statistically derived standard information corresponds to the same object identifier, then that same object identifier can be used as the object identifier of the target object. For example, after the three attribute features of the target object are matched with the base database features, matching base database features O, P, and Q can be obtained. The standard information bound to these three base database features O, P, and Q all point to user M, so user M can be used as the object identifier of the target object.

[0071] On the other hand, if the statistically obtained standard information corresponds to different object identifiers, it indicates that the object identifiers corresponding to the various attribute features are inconsistent. In this case, a prompt message indicating that the object identifier cannot be confirmed can be generated. For example, after matching the three attribute features of the target object with the base database features, matching base database features D, E, and F can be obtained. However, the standard information bound to base database features D and E both point to user K, while the standard information bound to base database feature F points to user L. This indicates that there is a contradiction in the object identifier in the matching results, and the object identifier of the target object cannot be determined. Subsequently, the object identifier of the target object can be further confirmed through other video data, but in the current round of data processing, the object identifier of the target object cannot be confirmed.

[0072] Furthermore, in one implementation, considering that not every attribute feature has a matching base database feature, if at least one attribute feature among the target object's attribute features does not have a matching base database feature, a prompt message indicating that the object identifier cannot be confirmed can be directly generated. In this way, by initially detecting the matching results, the situation where the object identifier cannot be identified can be directly determined, avoiding the subsequent statistical process of standard information and reducing the complexity of data processing.

[0073] This disclosure provides a technical solution through one or more embodiments, which employs a heterogeneous graph construction method to establish a correlation between the attribute features of a target object and pre-stored base database features. The attribute features of the target object are extracted from its motion trajectory and can describe the target object from multiple different aspects. The base database features can serve as standard features of the target object, and the heterogeneous graph can characterize the initial matching relationship between the attribute features and the base database features.

[0074] Subsequently, by updating the initial connection relationships between nodes in the heterogeneous graph, the matching relationship between attribute features and base library features can be accurately obtained. Based on this matching relationship, the target object can be identified.

[0075] As can be seen, one or more embodiments of this disclosure can analyze the motion trajectory of the target object when confirming its object identification, thereby avoiding the problem of inaccurate identification results caused by poor quality of a single video frame. Simultaneously, multiple attribute features can comprehensively describe the target object from multiple perspectives, and heterogeneous graphs can accurately match multiple attribute features with pre-stored base database features, thereby improving the accuracy of target object identification.

[0076] Please see Figure 5 One embodiment of this disclosure also provides a target object identification device, the device comprising:

[0077] The trajectory recognition unit 100 is used to acquire the target video and identify the motion trajectory of one or more target objects in the target video.

[0078] The heterogeneous graph construction unit 200 is used to extract multiple attribute features of the target object from the motion trajectory of the target object, and construct a heterogeneous graph between the attribute features of the target object and the pre-stored base database features; wherein the attribute features and the base database features serve as nodes in the heterogeneous graph, and some nodes have initial connection relationships.

[0079] The object recognition unit 300 is used to update the initial connection relationship in the heterogeneous graph, and based on the updated connection relationship between nodes, determine the matching result of each attribute feature of the target object in the base database features, and identify the target object according to the matching result.

[0080] The specific processing logic of each functional module can be found in the description of the aforementioned method implementation method, and will not be repeated here.

[0081] The various units described in the above embodiments can be implemented by a computer chip or by a product with a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0082] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0083] Please see Figure 6 One embodiment of this disclosure also provides an electronic device, which includes a memory and a processor. The memory is used to store a computer program, which, when executed by the processor, implements the above-described method for identifying the target object.

[0084] This disclosure also provides a computer-readable storage medium for storing a computer program that, when executed by a processor, implements the above-described method for identifying a target object.

[0085] The processor can be a central processing unit (CPU). It can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof.

[0086] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods in the above-described embodiments.

[0087] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0088] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0089] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, embodiments of apparatus, devices, and storage media are basically similar to method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0090] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0091] Although embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method of identifying a target object, characterized by, The method comprises: acquiring a target video, and identifying respective motion trajectories of one or more target objects in the target video; extracting a plurality of attribute features of the target objects from the motion trajectories of the target objects, and constructing a heterogeneous graph between the attribute features of the target objects and pre-stored base features; wherein the attribute features and the base features serve as nodes in the heterogeneous graph, and some nodes have an initial connection relationship therebetween; updating the initial connection relationship in the heterogeneous graph, and determining matching results of the respective attribute features of the target objects in the base features based on the connection relationship between the updated nodes, and identifying the target objects according to the matching results; wherein the construction of the heterogeneous graph between the attribute features of the target objects and the pre-stored base features comprises: establishing a first initial connection relationship between nodes represented by the respective attribute features of the target objects; for any target attribute feature among the respective attribute features of the target objects, determining target base features belonging to the same type as the target attribute feature among the pre-stored base features, and establishing a second initial connection relationship between the node represented by the target attribute feature and nodes represented by the respective target base features; the updating of the initial connection relationship in the heterogeneous graph comprises: updating initial weight values of the respective initial connection relationships in the heterogeneous graph, and canceling some initial connection relationships in the heterogeneous graph based on the updated weight values, and taking the heterogeneous graph with some initial connection relationships canceled as an updated heterogeneous graph.

2. The method of claim 1, wherein, The extraction of the plurality of attribute features of the target objects from the motion trajectories of the target objects comprises: identifying a plurality of groups of different types of feature sequences related to the target objects in the motion trajectories of the target objects; respectively determining respective representative features from the respective groups of feature sequences, and taking the respective representative features as the plurality of attribute features of the target objects.

3. The method of claim 2, wherein, The identification of a plurality of groups of different types of feature sequences related to the target objects in the motion trajectories of the target objects comprises: for any target type among the different types, identifying type features related to the target objects and conforming to the target type from the motion trajectories of the target objects; taking a feature sequence composed of the identified respective type features as a feature sequence of the target type.

4. The method of claim 2, wherein, Each of the respective groups of feature sequences comprises a plurality of respective type features, and each of the respective type features has respective quality information; the respective representative features are determined from the respective groups of feature sequences as follows: for any target feature sequence among the plurality of groups of different types of feature sequences, statistics the quality information of each type feature in the target feature sequence; based on the statistical quality information, taking a type feature with optimal quality as a representative feature of the target feature sequence.

5. The method of claim 1, wherein, The method further comprises: setting a pre-set fixed value as an initial weight value of the first initial connection relationship; For the target attribute feature and the target base library feature existing the second initial connection relationship, a similarity between the target attribute feature and the target base library feature is calculated, and the similarity is set as an initial weight value of the second initial connection relationship.

6. The method of claim 1, wherein, The matching result of each attribute feature of the target object in the base library features includes: For any target attribute feature of each attribute feature of the target object, a target base library feature existing a connection relationship with the target attribute feature is identified in the updated heterogeneous graph, and the target base library feature is taken as a base library feature matched with the target attribute feature.

7. The method of claim 1, wherein, According to the matching result, the target object is identified. If each attribute feature of the target object has a respective matched base library feature, respective standard information of the matched base library features is counted. According to each counted standard information, an object identifier of the target object is determined.

8. The method of claim 7, wherein, The method further includes: If at least one attribute feature of each attribute feature of the target object does not have a matched base library feature, a prompt message representing that the object identifier cannot be confirmed is generated.

9. The method of claim 7, wherein, According to each counted standard information, an object identifier of the target object is determined, including: If each counted standard information corresponds to a same object identifier, the same object identifier is taken as the object identifier of the target object. If each counted standard information corresponds to different object identifiers, a prompt message representing that the object identifier cannot be confirmed is generated.

10. An apparatus for identifying a target object, characterized by comprising: The device includes: A trajectory identification unit configured to acquire a target video and identify respective motion trajectories of one or more target objects in the target video; A heterogeneous graph construction unit configured to extract a plurality of attribute features of the target objects from the motion trajectories of the target objects, and construct a heterogeneous graph between the attribute features of the target objects and pre-stored base library features; wherein the attribute features and the base library features are taken as nodes in the heterogeneous graph, and some nodes have initial connection relationships; An object identification unit configured to update the initial connection relationships in the heterogeneous graph, determine a matching result of each attribute feature of the target objects in the base library features based on the connection relationships between the updated nodes, and identify the target object according to the matching result. The construction of the heterogeneous graph between the attribute features of the target objects and the pre-stored base library features includes: A first initial connection relationship is established between nodes represented by each attribute feature of the target objects; For any target attribute feature of each attribute feature of the target object, a target base library feature belonging to a same type as the target attribute feature is determined in the pre-stored base library features, and a second initial connection relationship is established between a node represented by the target attribute feature and nodes represented by each target base library feature; The updating of the initial connection relationships in the heterogeneous graph includes: The initial weight values of each of the initial connection relationships in the heterogeneous graph are updated, and based on the updated weight values, some of the initial connection relationships in the heterogeneous graph are cancelled, and the heterogeneous graph in which some of the initial connection relationships are cancelled is taken as an updated heterogeneous graph.

11. An electronic device, comprising: The electronic device includes a memory and a processor, the memory is used to store a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium is used to store a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1 to 9.

Images