An attribute recognition method, device and storage medium

Abstract

This application discloses an attribute recognition method, device, and storage medium. The method includes: detecting whether each target part of a target object contained in a target object image is visible, and obtaining a visibility detection result for each target part; performing attribute recognition on the target object based on the visibility detection results of each target part, and obtaining a target attribute recognition result for the target object. The target attribute recognition result includes target attributes of each target part of the target object, and the target attributes include target attributes other than visibility among the attributes of the target parts. Through the above method, this application can improve the effectiveness of the target attribute recognition result information.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and in particular to an attribute recognition method, device and storage medium. Background Technology

[0002] Currently, attribute recognition of various objects has become an indispensable part of the development of video structured analysis and intelligent surveillance. Among these, the effectiveness of object attribute information is crucial for intelligent surveillance. Current object attribute recognition provides recognition results based on predefined attribute categories. However, in real-world scenarios, various factors can cause acquired object images to be blurry, difficult to distinguish, or obscured, severely impacting the effectiveness of the object attribute recognition results.

[0003] Therefore, improving the effectiveness of object attribute recognition results is of great significance. Summary of the Invention

[0004] The main technical problem addressed by this application is to provide an attribute recognition method, device, and storage medium that can improve the effectiveness of the attribute recognition results of the target object.

[0005] To solve the above-mentioned technical problems, one technical solution adopted in this application is: to provide an attribute recognition method, the method comprising: detecting whether each target part of the target object contained in the target object image is visible, and obtaining the visibility detection result of each target part; performing attribute recognition on the target object based on the visibility detection result of each target part, and obtaining the target attribute recognition result of the target object, wherein the target attribute recognition result includes the target attributes of each target part of the target object, and the target attributes include the target attributes of the target parts other than visibility.

[0006] Specifically, the target object is attribute-identified based on the visibility detection results of each target part to obtain the target attribute identification results of the target object, including: in response to the presence of an invisible detection result in the visibility detection results, the target attribute of the target part corresponding to the invisible detection result is determined as a preset attribute, and the invisible detection result indicates that there is an invisible target part in the target object image.

[0007] The process of detecting whether each target part of the target object contained in the target object image is visible, and obtaining the visibility detection result of each target part, includes: processing the target object image using at least one processing model to obtain the part processing result of each processing model, wherein the part processing result includes information about each target part of the target object; and obtaining the visibility detection result of each target part based on the part processing result of each processing model.

[0008] Specifically, based on the part processing results of each processing model, the part visibility detection results of each target part are obtained, including: using the part processing results of each processing model to obtain the visibility confidence of each processing model for each target part; for each target part, the visibility confidence of each processing model for the target part is combined to determine whether the target part is visible.

[0009] The determination of whether a target part is visible, based on the visibility confidence of each processing model regarding the target part, includes: if the number of processing models that meet the preset confidence conditions reaches a first threshold, the target part is determined to be visible; if the number of processing models that meet the preset confidence conditions does not reach a second threshold, the target part is determined to be invisible; if the first threshold is greater than the second threshold, the target part is determined to be invisible; if the number of processing models that meet the preset confidence conditions reaches the second threshold but does not reach the first threshold, and the sum of the visibility confidence of each processing model regarding the target part is greater than the sum of the confidence thresholds of each processing model regarding the target part, the target part is determined to be visible; otherwise, the target part is determined to be invisible.

[0010] The processing model includes at least one or more of a key point recognition model, a part recognition model, and a part detection model. The key point recognition model is used to identify key points in each target part of the target object, the part recognition model is used to identify each target part of the target object, and the part detection model is used to detect the position of each target part of the target object.

[0011] Among them, the part processing results corresponding to the key point recognition model include the existence confidence of each key point in each target part of the target object, the part processing results corresponding to the part recognition model include the existence confidence of each target part of the target object, and the part processing results corresponding to the part detection model include the position confidence of each target part of the target object.

[0012] Specifically, the visibility confidence of each processing model with respect to each target part is obtained by utilizing the part processing results of each processing model. This includes: if the processing model includes a keypoint recognition model, then for each target part, the visibility confidence of the keypoint recognition model with respect to the target part is obtained by utilizing the presence confidence of each keypoint in the target part; if the processing model includes a part recognition model, then for each target part, the visibility confidence of the part recognition model with respect to the target part is obtained by utilizing the presence confidence of the target part; if the processing model includes a part detection model, then for each target part, the visibility confidence of the part recognition model with respect to the target part is obtained by utilizing the position confidence of the target part.

[0013] Among them, the preset attribute is an unknown attribute; and / or, the target object is a pedestrian.

[0014] Specifically, the target object is attribute-identified based on the visibility detection results of each target part to obtain the target attribute identification results. This includes: using an attribute identification model to identify the target object's attributes to obtain preliminary attribute identification results, where the preliminary attribute identification results include the initial attributes of each target part of the target object; for each target part, if the target part is not visible, the preset attribute is used as the target attribute of the target part; if the target part is visible, the initial attribute of the target part is used as the target attribute of the target part.

[0015] To solve the above-mentioned technical problems, another technical solution adopted in this application is: to provide an attribute identification device, including a memory and a processor coupled to each other, wherein the memory stores program instructions; and the processor is used to execute the program instructions stored in the memory to implement the above method.

[0016] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide a computer-readable storage medium for storing program instructions that can be executed to implement the above-mentioned method.

[0017] The above scheme first detects whether each target part of the target object in the target object image is visible, obtaining the visibility detection result of each target part. Then, based on the visibility detection results of each target part, it performs attribute recognition on the target object, obtaining the target attribute recognition result of the target object. The target attribute recognition result includes the target attributes of each target part of the target object, and the target attributes include the target attributes of the target part excluding visibility. Therefore, this application can more accurately identify the attributes of each target part by combining the visibility of each target part, reducing the impact of the visibility of the target part on the target object attributes (e.g., the blurry and difficult-to-distinguish attribute information of invisible parts or the information interference brought by invisible attribute information to the target object attribute results), and improving the effectiveness of the target object attribute recognition result information. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an embodiment of the attribute recognition method provided in this application;

[0019] Figure 2 yes Figure 1 The flowchart of step S11 shown is a schematic diagram of one embodiment.

[0020] Figure 3 yes Figure 2 The flowchart of step S22 shown is a schematic diagram of one embodiment;

[0021] Figure 4 This is a flowchart illustrating an embodiment of the attribute recognition method provided in this application;

[0022] Figure 5 yes Figure 1 The flowchart of step S12 shown is a schematic diagram of one embodiment;

[0023] Figure 6 This is a schematic diagram of the framework of an embodiment of the attribute recognition device provided in this application;

[0024] Figure 7 This is a schematic diagram of the framework of the computer-readable storage medium provided in this application. Detailed Implementation

[0025] To make the purpose, technical solution and effects of this application clearer and more explicit, the following describes this application in further detail with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the terms "existence" and "visibility" used herein have the same meaning. For example, "visibility of a part" and "existence of a part" have the same meaning. Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0027] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the attribute recognition method provided in this application. It should be noted that if substantially the same result is obtained, this embodiment does not necessarily reflect that outcome. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown, this embodiment includes:

[0028] S11: Detect whether each target part of the target object contained in the target object image is visible, and obtain the visibility detection result of each target part.

[0029] The method in this embodiment is used to perform attribute recognition on the target object based on the visibility detection results of each target part in the target object image after obtaining the visibility detection results of each target part, so as to obtain the target attribute recognition result of the target object.

[0030] The target object image described herein may be, but is not limited to, a video frame from a captured target object video or an image of the target object captured by an image acquisition device. The target object image contains the target object. The target object described herein may be, but is not limited to, plants, animals, pedestrians, etc. It is understood that the target object itself has at least one part. For example, in some embodiments, the target object is a table, which includes a tabletop and table legs. In other embodiments, the target object is a pedestrian, which includes multiple parts such as the head, head and shoulders, upper body, lower body, and feet. A target part refers to at least one part of the target object; that is, the target part used to detect whether it is visible can be a part of all parts of the target object, or it can be all parts of the target object. The specific parts of the target object included in the target part can be determined according to the actual detection and application needs, and are not specifically limited here.

[0031] It should be noted that, due to various factors, the target object image may be an image in which all target parts of the target object are clearly visible, or it may be an image in which at least one target part of the target object is blurred or occluded. For example, in some application scenarios, the target object is a pedestrian, but in the acquired image of the target pedestrian, at least one target part is blurred or occluded. Here, the at least one part can be any part or a combination of any part of the target pedestrian's head, shoulders, upper body, lower body and feet.

[0032] In this embodiment, after obtaining the target object image, it is necessary to detect whether each target part of the target object contained in the target object image is visible, so as to obtain the visibility detection result of each target part. In some embodiments, for each target part, the visibility result of the target part includes two results: visible and invisible. That is, the visible target parts and invisible target parts can be determined by detection. In other embodiments, for any target part, its visibility result also includes the visible part and the invisible part of the target part. That is, the visible part and the invisible part of any target part can be obtained by detection. For example, if the target object is a pedestrian and the target part is the face, the detection shows that the forehead and eyes of the target face are visible, while the nose and mouth are obscured by objects and are not visible.

[0033] S12: Based on the visibility detection results of each target part, perform attribute recognition on the target object to obtain the target attribute recognition result of the target object. The target attribute recognition result includes the target attributes of each target part of the target object. The target attributes include the target attributes of the target parts other than visibility.

[0034] The target attribute recognition result obtained based on the target part visibility detection result includes the target attributes of each target part of the target object. These target attributes include all target attributes of the target part except for visibility. That is, in this embodiment, the target attributes include not only the visibility of the target part but also other attributes used to describe each target part of the target object, such as attributes describing the color, shape, or direction of the target part. The specific target attribute information included, excluding visibility, can be determined according to the actual application scenario. For example, in some embodiments, the target object is a pedestrian, and the target part of the pedestrian includes the head. The target attribute recognition result includes the visibility of the pedestrian's head, as well as attributes related to the head, such as the attributes of objects worn on the head, movement attributes, and human physiological attributes. Similarly, in other embodiments, the target object is a table, and the target parts of the table include the tabletop and table legs. The target attribute recognition result includes the visibility of each target part of the table, as well as attribute information related to the tabletop and table legs, such as the color and shape of the tabletop, and the number and color of the table legs.

[0035] In this embodiment, the visibility of each target part of the target object in the target object image is first detected to obtain the visibility detection result of each target part. Then, based on the visibility detection results of each target part, attribute recognition is performed on the target object to obtain the target attribute recognition result of the target object. The target attribute recognition result includes the target attributes of each target part of the target object, and the target attributes include the target attributes of the target part other than visibility. Therefore, the solution of this embodiment can more accurately identify the attributes of each target part by combining the visibility of each target part, reduce the influence of the visibility of the target part on the target object attributes (e.g., the blurry and difficult-to-distinguish attribute information of invisible parts or the information interference brought by invisible attribute information to the target object attribute results), and improve the effectiveness of the target object attribute recognition result information.

[0036] In some embodiments, if an invisible detection result exists in the part detection results, the target attribute of the invisible target part corresponding to the invisible detection result is a preset attribute. Here, the invisible detection result indicates the existence of an invisible target part in the target object image, and the preset attribute is used to characterize the attribute information of the invisible part. That is, when the preset attribute exists in the target attribute recognition result, it means that the part corresponding to the preset attribute is invisible, and the attribute information of the corresponding part is unknown. Specific preset attributes can be set according to the actual application scenario. For example, it can include any one or any combination of specific text, blank spaces, numbers, letters, and graphics such as "unknown," without specific limitations here. It is understood that compared to the prior art that provides the attribute recognition result of the target object according to predefined object attribute information, this embodiment determines the target attribute of the target part corresponding to the invisible detection result as a preset attribute, which can effectively avoid the information interference caused by the blurry and difficult-to-distinguish attribute information of the invisible part or the invisible attribute information to the overall target object attribute result, thereby further improving the effectiveness of the target object attribute recognition result information.

[0037] Please see Figure 2 , Figure 2 yes Figure 1 The flowchart shown is a schematic diagram of one embodiment of step S11. It should be noted that if substantially the same result is achieved, this embodiment does not necessarily follow the same pattern. Figure 2 For the sake of clarity, in this embodiment, step S11 further includes steps S21 and S22:

[0038] S21: Process the target object image using at least one processing model to obtain the part processing results of each processing model, wherein the part processing results include information about each target part of the target object.

[0039] In this embodiment, after acquiring the target object image, at least one processing model is used to process the target object image to obtain the part processing results of each processing model. The part processing results include information about each target part of the target object. For example, if the target object includes two target parts, part A and part B, then the part processing results of each processing model include the processing result information of each processing model regarding target part A and target part B of the target object. In this embodiment, the information about each target part refers to information related to each target part of the target object, which may include, but is not limited to, the location information and visibility information of each target part. The specific information about each target part of the target object obtained by the specific processing model can be determined according to actual needs and is not specifically limited here.

[0040] In this embodiment, "at least one processing model" means that the number of processing models can be one or more. Each processing model is used to detect the visibility of different target parts in an image of the target object, that is, to detect the visible and invisible parts of each target part. Therefore, the number and selection of processing models can be determined based on the actual detection performance of each processing model. It is understood that if using only one processing model can effectively detect the visibility of target parts, then the number of at least one processing model is one. If three recognition models need to be used for fusion judgment to effectively detect the visibility of target parts, then the number of at least one processing model is three. It should be noted that this is only an example and does not limit the number of at least one processing model.

[0041] S22: Based on the part processing results of each processing model, the part visibility detection results are obtained.

[0042] In this embodiment, after obtaining the part processing results of each processing model in step S21, the part visibility detection results are obtained based on the part processing results of each processing model. The part visibility detection results include the visibility detection results of each target part of the target object; that is, the part visibility detection results include the visible and invisible results corresponding to each target part of the target object.

[0043] Specifically, please refer to Figure 3 , Figure 3 yes Figure 2 The flowchart shown is a schematic diagram of one embodiment of step S22. It should be noted that if substantially the same result is achieved, this embodiment does not necessarily follow the same pattern. Figure 3 For the sake of clarity, in this embodiment, step S22 further includes steps S31 and S32:

[0044] S31: Using the part processing results of each processing model, obtain the visibility confidence of each processing model for each target part.

[0045] The method in this embodiment is used to determine whether a target part is visible by combining the visibility confidence of each processing model with respect to each target part after obtaining the visibility confidence of each processing model with respect to the target part.

[0046] In one embodiment, such as Figure 4As shown, at least one processing model includes a keypoint recognition model, which is used to identify keypoints in each target part of the target object. Each target part of the target object corresponds to multiple keypoints. The specific number of keypoints corresponding to each target part of the target object can be determined according to the actual application scenario and the specific target object, and is not specifically limited here. In this embodiment, the keypoints in each target part may include, but are not limited to, information such as the location of each keypoint in each target part and the confidence level of the visibility of each keypoint. In one embodiment, the keypoint recognition model is used to identify the confidence level of the visibility of each keypoint in each target part of the target object. In another embodiment, the keypoint recognition model is used to identify the location and visibility confidence level of each keypoint in each part of the target object. That is, the processing result obtained by processing the target object image using at least one processing model includes the visibility confidence level of each keypoint in each target part of the target object. For each keypoint, the result of whether the keypoint is visible can be obtained based on the corresponding confidence level of its visibility. In some embodiments, if the confidence level of a key point being visible is greater than a preset threshold, then the key point is determined to be visible. In other embodiments, if the confidence level of a key point being visible is greater than the surrounding values, then the key point is determined to be visible.

[0047] Specifically, in one embodiment, the key points in each part of the target object obtained by the key point recognition model are heat maps corresponding to each key point. Each key point corresponds to a heat map, which contains the location information of the corresponding key point in the target object image and the visibility confidence information of the key point. The visibility or existence of each key point can be determined based on the visibility confidence information of the key point in the heat map. In this embodiment, after obtaining the existence confidence of each key point in each part, the visibility confidence of the key point recognition model about the target part can be obtained by using the existence confidence of each key point in the target part.

[0048] Specifically, taking part A of the target object as an example, the visibility confidence of the keypoint recognition model regarding part A is obtained by using the existence confidence of each keypoint in part A. The visibility confidence of part A is calculated using the following formula:

[0049]

[0050] Where conf(A) represents the visibility confidence of part A, n represents the number of keypoints, and exist represents the visibility of keypoint X. i Therefore, exist is 1 if the key point X i If invisible, then exist is 0, conf(x i ) represents the key point X iVisible confidence level.

[0051] Understandably, after obtaining the visibility confidence of each keypoint in each part using the keypoint recognition model, the visibility of each keypoint is first determined based on its visibility confidence. After determining the visibility of each keypoint, the average confidence of all keypoints corresponding to part A is calculated using the formula described above to obtain the visibility confidence of part A. The visibility confidence of the keypoint recognition model for each part of the target object is obtained in the same way.

[0052] In one embodiment, such as Figure 4 As shown, at least one processing model includes a part recognition model, which is used to identify each target part of the target object. Specifically, the part recognition model is used to identify the existence confidence of each target part of the target object. That is, the part processing result obtained by processing the target object image using at least one processing model includes the existence confidence of each target part of the target object. After obtaining the visibility confidence of each target part of the target object using the part recognition model, for each target part, the visibility confidence of the part recognition model with respect to the target part is obtained using the existence confidence of each target part. In this embodiment, the visibility confidence of each target part of the target object can be directly obtained using the part recognition model. In some embodiments, the obtained visibility confidence of the part recognition model with respect to the target part only includes the visibility confidence information of each target part. In other embodiments, the obtained visibility confidence of the part recognition model with respect to the part includes the visibility confidence and the invisibility confidence of each target part, wherein the sum of the visibility confidence and the invisibility confidence is 1.

[0053] In some embodiments, such as Figure 4 As shown, at least one processing model includes a part detection model. The part detection model is used to detect the positions of each target part of the target object and obtain the confidence score of each detected target part position. The detected target part positions may include, but are not limited to, the coordinates of the center point of the target part, or the coordinates of the upper left and lower right corners of the detection box corresponding to the target part. The position confidence score is the confidence score that the position corresponds to the existence of the target part. In other words, in this embodiment, the part detection model can be used to identify the positions of each target part of the target object and the corresponding position confidence scores. That is, by processing the target object image using at least one processing model, the obtained part processing result includes the positions of each target part of the target object and the corresponding position confidence scores. It can be understood that in this embodiment, after obtaining the position confidence scores of each target part of the target object using the part detection model, the visibility confidence score of the part detection model regarding the part can be obtained using the position confidence scores of each target part.

[0054] It should be noted that, in some embodiments, at least one processing model includes one or more of the key point recognition model, part recognition model, and part detection model mentioned above. The determination of at least one processing model can be based on the actual detection effect of each model on the visibility of the target part, and no specific limitation is made here.

[0055] S32: For each target part, determine whether the target part is visible by combining the visibility confidence of each processing model.

[0056] In this embodiment, after obtaining the visibility confidence of each processing model for each target part by utilizing the part processing results of each processing model, for each target part, it is necessary to combine the visibility confidence of each processing model for the target part to determine whether the target part is visible.

[0057] In determining whether a target region is visible by integrating the visibility confidence scores of various processing models, the following criteria need to be considered:

[0058] The first discrimination criterion is: if the number of processing models satisfying the pre-set reliability condition reaches a first threshold, then the target part is determined to be visible. The pre-set reliability condition is that the visibility confidence of the processing model regarding the target part is greater than the confidence threshold for the processing model regarding the target part. The second discrimination criterion is: if the number of processing models satisfying the pre-set reliability condition does not reach a second threshold, then the target part is determined to be invisible. The first threshold is greater than the second threshold. The third discrimination criterion is: if the number of processing models satisfying the pre-set reliability condition reaches the second threshold but does not reach the first threshold, and the sum of the visibility confidence of all processing models regarding the target part is greater than the sum of the confidence thresholds for all processing models regarding the target part, then the target part is determined to be visible; otherwise, the target part is determined to be invisible. The specific values ​​of the first and second thresholds in the above discrimination criteria can be determined based on the number of processing models and the discrimination effect of part visibility.

[0059] It should be noted that each processing model corresponds to a confidence threshold for each target part. For example, if there are two processing models, namely the first processing model and the second processing model, and the target object has two target parts, A and B, then the confidence thresholds of the processing models for the target parts include: the first confidence threshold of the first processing model for part A, the second confidence threshold of the first processing model for part B, the third confidence threshold of the second processing model for part A, and the fourth confidence threshold of the second processing model for part B. The confidence thresholds for each target part of each processing model can be determined based on its visibility detection performance.

[0060] It should also be noted that for any target part of the target object, such as part A, if the number of processing models that meet the pre-set confidence conditions satisfies the first discrimination condition, then part A is determined to be visible. If the second discrimination condition is met, then part A is determined to be invisible. If neither the first nor the second discrimination condition is met, then it is necessary to further determine whether the third discrimination condition is met. If the third discrimination condition is met, then part A is determined to be visible; otherwise, part A is determined to be invisible. The method for determining the visibility of other target parts of the target object is the same as that for part A, and will not be elaborated on here.

[0061] In one specific embodiment, such as Figure 4 As shown, the number of at least three processing models is three, namely a key point recognition model, a component recognition model, and a component detection model. The corresponding first quantity threshold is 2, and the second quantity threshold is 1. In this embodiment, for the target part A of the target object, if the number of processing models that meet the preset confidence conditions reaches the first quantity threshold 2, that is, at least two of the three processing models have a visibility confidence of at least two processing models about part A that reaches the confidence threshold corresponding to at least two processing models about part A, then part A is determined to be visible. If the number of processing models meeting the pre-set confidence conditions does not reach the second threshold 1 (i.e., the visibility confidence of all three processing models regarding part A does not reach the confidence threshold corresponding to each model for part A), then part A is determined to be invisible. If the number of processing models meeting the pre-set confidence conditions reaches the second threshold 1 but does not reach the first threshold 2, and the sum of the visibility confidence of each processing model regarding part A is greater than the sum of the confidence thresholds of each processing model regarding part A, then part A is determined to be visible; otherwise, part A is determined to be invisible. That is, if only one of the three processing models has a visibility confidence of part A greater than its confidence threshold, then it is necessary to further determine whether the sum of the visibility confidence of the three processing models regarding part A is greater than the sum of the confidence thresholds of each processing model regarding part A. If so, then part A is determined to be visible; otherwise, part A is determined to be invisible. It is understandable that the logic for determining the visibility of other target parts of the target object is the same as for part A, and will not be elaborated further here. It should be noted that the above is only used as an example and does not limit the number of at least one processing model, the first number threshold, or the second number threshold.

[0062] Please see Figure 5 , Figure 5 yes Figure 1 The flowchart shown is a schematic diagram of one embodiment of step S12. It should be noted that in this embodiment, if there are substantially the same result, this embodiment does not necessarily use the same result. Figure 5 The illustrated process sequence is limited. For example... Figure 5As shown, in this embodiment, the target object is attribute-identified based on the visibility detection results of each target part, resulting in the target attribute identification result of the target object, including:

[0063] S51: Use the attribute recognition model to identify the attributes of the target object and obtain preliminary attribute recognition results, wherein the preliminary attribute recognition results include the initial attributes of each target part of the target object.

[0064] In this process, the attribute recognition model is used to identify the attributes of the target object image. The preliminary attribute recognition results include the initial attributes of each target part of the target object. The initial attributes of each target part represent the attribute information about the target part that is pre-set by the attribute recognition model. For example, in some embodiments, the target object is a pedestrian, and the target parts of the target object include the head, head and shoulders, upper body, lower body, and feet. The pre-set head attribute information includes: head-wearing object attributes, head movement attributes, and human physiological attributes. The pre-set head and shoulder attribute information includes: head-to-shoulder-wearing object attributes. The pre-set upper body attribute information includes: upper body-wearing object and corresponding color attributes, body orientation attributes, and whether the hands are holding an object. The pre-set lower body target attribute information includes: leg-wearing object and corresponding color attributes. The pre-set foot attribute information includes: shoe color attributes and shoe style attributes. It should be noted that in this step, regardless of whether the target parts of the target object are visible or not, the corresponding recognition results will be given according to the pre-set attribute information about the target parts. For example, the head of the target object may be obscured by an object and therefore not visible, but the preliminary attribute recognition results may include the aforementioned pre-set head attribute information: head-wearing object attributes, head movement attributes, and human physiological attributes.

[0065] S52: For each target part, if the target part is not visible, the preset attribute is used as the target attribute of the target part; if the target part is visible, the initial attribute of the target part is used as the target attribute of the part.

[0066] In this embodiment, after using the attribute recognition model to identify the attributes of the target object and obtaining the preliminary attribute recognition result, the target attribute recognition result of the target object is obtained by using the preliminary attribute recognition result and the visibility detection result of the target part.

[0067] Specifically, in this embodiment, for each target part, if the target part is invisible, a preset attribute is used as the target attribute of the invisible target part; if the target part is visible, the initial attribute of the target part is used as the target attribute of the part. That is, if all target parts of the target object are visible, the initial attribute of the target part is the target attribute of the target part; if there are invisible parts among the target parts of the target object, the initial attribute corresponding to the invisible target part is adjusted to the preset attribute, so that in the obtained target attribute recognition result, the target attribute of the visible target part is the corresponding initial attribute, and the target attribute of the invisible target part is the preset attribute.

[0068] In other embodiments, the attribute recognition model can be directly used to process the target object image and the target part visibility detection results to obtain the target attribute recognition results of the target object. That is, the attribute recognition model will output the corresponding recognition results of the target attributes of the visible target parts in the target object image according to the preset attribute information based on the target part visibility detection results, and output the target attributes of the invisible target parts as preset attributes.

[0069] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0070] Please see Figure 6 , Figure 6 This is a schematic diagram of a framework of an embodiment of the attribute recognition device provided in this application. In this embodiment, the attribute recognition device 60 includes a memory 61 and a processor 62.

[0071] Processor 62 can also be referred to as a CPU (Central Processing Unit). Processor 62 may be an integrated circuit chip with signal processing capabilities. Processor 62 can also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. A general-purpose processor can be a microprocessor, or processor 62 can be any conventional processor 62, etc.

[0072] The memory 61 in the attribute recognition device 60 is used to store the program instructions required for the processor 62 to run.

[0073] The processor 62 is used to execute program instructions to implement the method provided in any embodiment of the attribute identification method described above and any non-conflicting combination thereof.

[0074] Please see Figure 7 , Figure 7This is a schematic diagram of the framework of the computer-readable storage medium provided in this application. The computer-readable storage medium 70 of this application embodiment stores program instructions 71, which, when executed, implement the method provided by any embodiment and any non-conflicting combination of the attribute identification method. The program instructions 71 can form a program file and be stored in the aforementioned computer-readable storage medium 70 in the form of a software product, so that a computer device (which may be a personal computer, server, or network device, etc.) can execute all or part of the steps of the methods of various embodiments of this application. The aforementioned computer-readable storage medium 70 includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0075] The above scheme first detects whether each target part of the target object in the target object image is visible, obtaining the visibility detection result of each target part. Then, based on the visibility detection results of each target part, it performs attribute recognition on the target object, obtaining the target attribute recognition result of the target object. The target attribute recognition result includes the target attributes of each target part of the target object, and the target attributes include the target attributes of the target part excluding visibility. Therefore, this application can more accurately identify the attributes of each target part by combining the visibility of each target part, reducing the impact of the visibility of the target part on the target object attributes (e.g., the blurry and difficult-to-distinguish attribute information of invisible parts or the information interference brought by invisible attribute information to the target object attribute results), and improving the effectiveness of the target object attribute recognition result information.

[0076] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0077] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0078] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0079] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0080] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0081] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0083] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. An attribute recognition method, characterized in that, The method includes: The visibility of each target part in the target object image is detected to obtain the visibility detection result of each target part; for each target part, whether the target part is visible is determined by combining the visibility confidence of each processing model about the target part; Based on the visibility detection results of each target part, the target object is attribute-identified to obtain the target attribute identification result of the target object. The target attribute identification result includes the target attributes of each target part of the target object. The target attributes include the target attributes of the target parts other than visibility. The target attributes of invisible target parts are preset attributes. Specifically, determining whether the target region is visible by combining the visibility confidence scores of each processing model for the target region includes: If the number of processing models that meet the preset confidence conditions does not reach the second quantity threshold, or if the number of processing models that meet the preset confidence conditions reaches the second quantity threshold, and the sum of the visibility confidence of each processing model about the target part is not greater than the sum of the confidence thresholds of each processing model about the target part, then the target part is determined to be invisible; the preset confidence conditions are that the visibility confidence of the processing model about the target part is greater than the confidence threshold of the processing model about the target part, the second quantity threshold is less than the first quantity threshold, and the first quantity threshold is less than the total number of processing models.

2. The method according to claim 1, characterized in that, The step of detecting whether each target part of the target object contained in the target object image is visible, and obtaining the visibility detection result of each target part, includes: The target object image is processed using at least one processing model to obtain each... The processing model's part processing results, wherein the part processing results include information about each target part; Using the location processing results of each processing model, the visibility confidence of each processing model with respect to each target location is obtained; For each target location, the visibility confidence of each processing model with respect to the target location is used to determine whether the target location is visible.

3. The method according to claim 1, characterized in that, Determining whether the target region is visible by combining the visibility confidence scores of each processing model with respect to the target region further includes: If the number of processing models that meet the preset confidence conditions reaches the first number threshold, then the target part is determined to be visible, and the first number threshold is greater than the second number threshold. If the number of processing models that meet the preset confidence conditions reaches the second quantity threshold but does not reach the first quantity threshold, and the sum of the visibility confidence of each processing model about the target part is greater than the sum of the confidence thresholds of each processing model about the target part, then the target part is determined to be visible.

4. The method according to claim 2, characterized in that, The at least one processing model includes one or more of a key point recognition model, a part recognition model, and a part detection model, wherein the key point recognition model is used to identify key points in each target part of the target object, the part recognition model is used to identify each target part of the target object, and the part detection model is used to detect the position of each target part of the target object.

5. The method according to claim 4, characterized in that, The part processing result corresponding to the key point recognition model includes the existence confidence of each key point in each target part of the target object; the part processing result corresponding to the part recognition model includes the existence confidence of each target part of the target object; and the part processing result corresponding to the part detection model includes the position confidence of each target part of the target object. The step of obtaining the visibility confidence of each processing model with respect to each target region using the region processing results of each processing model includes: If the processing model includes the key point recognition model, then for each target part, the visibility confidence of the key point recognition model with respect to the target part is obtained by using the existence confidence of each key point in the target part. If the processing model includes the part recognition model, then for each target part, the visibility confidence of the part recognition model with respect to the target part is obtained by using the existence confidence of the target part. If the processing model includes the part detection model, then for each target part, the visibility confidence of the part recognition model with respect to the target part is obtained by using the position confidence of the target part.

6. The method according to claim 1, characterized in that, The preset attribute is an unknown attribute; and / or, the target object is a pedestrian.

7. The method according to claim 1, characterized in that, The step of performing attribute recognition on the target object based on the location visibility detection results of each target location to obtain the target attribute recognition result of the target object further includes: The target object is identified using an attribute recognition model to obtain preliminary attribute recognition results, wherein the preliminary attribute recognition results include the initial attributes of each target part of the target object; For each target part, if the target part is not visible, the preset attribute is used as the target attribute of the target part; if the target part is visible, the initial attribute of the target part is used as the target attribute of the target part.

8. An attribute recognition device, characterized in that, Including interconnected memory and processor, The memory stores program instructions; The processor is used to execute program instructions stored in the memory to implement the method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program instructions that can be executed to implement the method of any one of claims 1-7.

Images