Object recognition method and device, electronic equipment and storage medium

Abstract

The invention relates to an object recognition method and device, electronic equipment and a storage medium. The method comprises the steps that a to-be-recognized driving scene graph is input into a preset object recognition model for multi-level feature extraction; inputting the first-level driving scene features into a first recognition prediction layer in a preset object recognition model for first category prediction processing to obtain a first prediction object category; inputting the second-level driving scene feature and the first prediction object category into a second identification prediction layer in a preset object identification model for second category prediction processing to obtain a second prediction object category; and inputting the third-level driving scene feature and the second prediction object category into a third identification prediction layer in a preset object identification model for third category prediction processing, thereby determining a target object category corresponding to the to-be-identified driving scene graph. According to the embodiment of the invention, similar object categories in the to-be-recognized driving scene graph can be effectively distinguished, the classification precision and reliability are improved, and automatic driving decision errors are reduced.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer vision, and particularly relates to an object recognition method and device, electronic equipment and storage medium. BACKGROUND

[0002] With the development of automatic driving technology, the classification demand for objects in a vehicle driving scene is increasingly urgent, and how to improve the accuracy and reliability of object classification in the vehicle driving scene has become a key to ensuring the safe operation of the automatic driving system.

[0003] In related technologies, a single-class classifier is often used to directly identify the object classes in a vehicle driving scene image. However, in related technologies, the number of object classes to be identified in the vehicle driving scene is large, and similar object classes cannot be effectively distinguished, which leads to the inability to adapt to the complexity of the driving scene, low object recognition accuracy and reliability, and further leads to decision-making errors and other problems. SUMMARY

[0004] The present disclosure provides an object recognition method, device, electronic equipment and storage medium to at least solve the problem that similar object classes cannot be effectively distinguished in related technologies, and the object recognition accuracy and reliability are low, which further leads to decision-making errors and other problems. The technical solutions of the present disclosure are as follows: According to a first aspect of an embodiment of the present disclosure, an object recognition method is provided, comprising: obtaining a driving scene image to be identified; inputting the driving scene image to be identified into a feature extraction layer in a preset object recognition model for multi-level feature extraction to obtain a first-level driving scene feature, a second-level driving scene feature and a third-level driving scene feature corresponding to the driving scene image to be identified; inputting the first-level driving scene feature into a first recognition prediction layer in the preset object recognition model, performing first-class prediction processing based on at least one preset object class to obtain a first predicted object class; inputting the second-level driving scene feature and the first predicted object class into a second recognition prediction layer in the preset object recognition model, performing second-class prediction processing based on at least one first sub-object class to obtain a second predicted object class; the at least one first sub-object class is at least one object class under each preset object class; inputting the third-level driving scene feature and the second predicted object class into a third recognition prediction layer in the preset object recognition model, performing third-class prediction processing based on at least one second sub-object class to obtain a third predicted object class; the at least one second sub-object class is at least one object class under each first sub-object class; Based on the third predicted object category, the target object category corresponding to the driving scene map to be identified is determined.

[0005] In an optional embodiment, the first recognition prediction layer includes a self-attention layer and a first category output layer. The step of inputting the first-level driving scene features into the first recognition prediction layer of the preset object recognition model, and performing first category prediction processing based on at least one preset object category to obtain the first predicted object category includes: The first-level driving scene features are input into the self-attention layer for self-attention calculation to obtain the first-level driving scene fusion features. The first-level driving scene fusion feature is input into the first category output layer for first category prediction processing to obtain first object category indication data corresponding to the at least one preset object category; the first object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each preset object category. The first predicted object category is obtained based on the first preset object category threshold and the first object category indication data.

[0006] In an optional embodiment, the second recognition prediction layer includes a first cross-attention layer and a second category output layer. The step of inputting the second-level driving scene features and the first predicted object category into the second recognition prediction layer of the preset object recognition model, and performing second category prediction processing based on at least one first sub-object category to obtain the second predicted object category includes: The second-level driving scene features and the first predicted object category are input into the first cross-attention layer for cross-attention calculation to obtain the second-level driving scene fusion features. The second-level driving scene fusion features are input into the second category output layer for second category prediction processing to obtain second object category indication data corresponding to the at least one first sub-object category; the second object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each first sub-object category. The second predicted object category is obtained based on the second preset object category threshold and the second object category indication data.

[0007] In an optional embodiment, the third recognition prediction layer includes a second cross-attention layer and a third category output layer. The step of inputting the third-level driving scene features and the second predicted object category into the third recognition prediction layer of the preset object recognition model, and performing third category prediction processing based on at least one second sub-object category to obtain the third predicted object category includes: The third-level driving scene features and the second predicted object category are input into the second cross-correlation layer for cross-attention calculation to obtain the third-level driving scene fusion features. The third-level driving scene fusion features are input into the third-category output layer for third-category prediction processing to obtain third-object category indication data corresponding to the at least one second sub-object category; the third-object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each second sub-object category. The third predicted object category is obtained based on the third preset object category threshold and the third object category indication data.

[0008] In an optional embodiment, the method further includes: Obtain the current sample driving scene image and the preset prediction object category corresponding to the current sample driving scene image; The current sample driving scene image is input into the object recognition model to be trained to perform object recognition and obtain the predicted object category of the sample. The object recognition loss is determined based on the predicted object category of the sample and the preset predicted object category; Based on the object recognition loss, the object recognition model to be trained is trained to obtain the preset object recognition model.

[0009] In an optional embodiment, the preset predicted object category includes a first preset predicted object category, a second preset predicted object category, and a third preset predicted object category; the sample predicted object category includes a first sample predicted object category, a second sample predicted object category, and a third sample predicted object category; and determining the object recognition loss based on the sample predicted object category and the preset predicted object category includes: The first object recognition loss is determined based on the first sample predicted object category and the first preset predicted object category; The second object recognition loss is determined based on the second sample predicted object category and the second preset predicted object category; The third object recognition loss is determined based on the predicted object category of the third sample and the predicted object category of the third preset sample. The object recognition loss is obtained by weighting the first object recognition loss, the second object recognition loss, and the third object recognition loss.

[0010] In an optional embodiment, the target object category includes at least one object category corresponding to each object, and the method further includes: If the preset object category corresponding to the current object is not included in the first predicted object category, the object category corresponding to the current object is corrected; the current object is any one of the at least one objects.

[0011] According to a second aspect of the present disclosure, an object recognition device is provided, comprising: The module for acquiring the driving scene image to be identified is used to acquire the driving scene image to be identified. The multi-level feature extraction module is used to input the driving scene image to be identified into the feature extraction layer of the preset object recognition model for multi-level feature extraction, so as to obtain the first-level driving scene features, the second-level driving scene features and the third-level driving scene features corresponding to the driving scene image to be identified; The first category prediction processing module is used to input the first-level driving scene features into the first recognition prediction layer in the preset object recognition model, and perform first category prediction processing based on at least one preset object category to obtain the first predicted object category. The second category prediction processing module is used to input the second-level driving scene features and the first predicted object category into the second recognition prediction layer of the preset object recognition model, and perform second category prediction processing based on at least one first sub-object category to obtain the second predicted object category; the at least one first sub-object category is at least one object category under each preset object category. The third category prediction processing module is used to input the third-level driving scene features and the second predicted object category into the third recognition prediction layer of the preset object recognition model, and perform third category prediction processing based on at least one second sub-object category to obtain the third predicted object category; the at least one second sub-object category is at least one object category under each first sub-object category. The target object category determination module is used to determine the target object category corresponding to the driving scene map to be identified based on the third predicted object category.

[0012] In an optional embodiment, the first recognition prediction layer includes a self-attention layer and a first category output layer, and the first category prediction processing module includes: The self-attention calculation unit is used to input the first-level driving scene features into the self-attention layer for self-attention calculation to obtain the first-level driving scene fusion features. The first category prediction processing unit is used to input the first-level driving scene fusion features into the first category output layer for first category prediction processing to obtain first object category indication data corresponding to the at least one preset object category; the first object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each preset object category. The first predicted object category determination unit is used to obtain the first predicted object category based on the first preset object category threshold and the first object category indication data.

[0013] In an optional embodiment, the second recognition prediction layer includes a first cross-attention layer and a second category output layer, and the second category prediction processing module includes: The first cross-attention calculation unit is used to input the second-level driving scene features and the first predicted object category into the first cross-attention layer to perform cross-attention calculation and obtain the second-level driving scene fusion features. The second category prediction processing unit is used to input the second-level driving scene fusion features into the second category output layer for second category prediction processing to obtain second object category indication data corresponding to the at least one first sub-object category; the second object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each first sub-object category. The second predicted object category determination unit is used to obtain the second predicted object category based on the second preset object category threshold and the second object category indication data.

[0014] In an optional embodiment, the third recognition prediction layer includes a second cross-attention layer and a third category output layer, and the third category prediction processing module includes: The second cross-attention calculation unit is used to input the third-level driving scene features and the second predicted object category into the second cross-correlation layer to perform cross-attention calculation and obtain the third-level driving scene fusion features. The third category prediction processing unit is used to input the third-level driving scene fusion features into the third category output layer for third category prediction processing to obtain third object category indication data corresponding to the at least one second sub-object category; the third object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each second sub-object category. The third predicted object category determination unit is used to obtain the third predicted object category based on the third preset object category threshold and the third object category indication data.

[0015] In an optional embodiment, the apparatus further includes: The preset prediction object category acquisition module is used to acquire the current sample driving scene image and the preset prediction object category corresponding to the current sample driving scene image; The object recognition module is used to input the current sample driving scene image into the object recognition model to be trained for object recognition and obtain the predicted object category of the sample. The object recognition loss determination module is used to determine the object recognition loss based on the object category predicted by the sample and the preset predicted object category. The model training module is used to train the object recognition model to be trained based on the object recognition loss, so as to obtain the preset object recognition model.

[0016] In an optional embodiment, the preset predicted object category includes a first preset predicted object category, a second preset predicted object category, and a third preset predicted object category; the sample predicted object category includes a first sample predicted object category, a second sample predicted object category, and a third sample predicted object category; and the object recognition loss determination module includes: The first object recognition loss determination unit is used to determine the first object recognition loss based on the first sample predicted object category and the first preset predicted object category. The second object recognition loss determination unit is used to determine the second object recognition loss based on the object category predicted by the second sample and the second preset object category. The third object recognition loss determination unit is used to determine the third object recognition loss based on the predicted object category of the third sample and the third preset predicted object category. The object recognition loss determination unit is used to perform weighted processing on the first object recognition loss, the second object recognition loss and the third object recognition loss to obtain the object recognition loss.

[0017] In an optional embodiment, the target object category includes at least one object category corresponding to each object, and the apparatus further includes: The object category correction module is used to correct the object category corresponding to the current object when the preset object category corresponding to the current object is not included in the first predicted object category; the current object is any one of the at least one objects.

[0018] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method as described in any one of the first aspects above.

[0019] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided such that, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the method described in any of the first aspects of the present disclosure.

[0020] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects: As can be seen from the technical solution provided in the example, this specification obtains a driving scene image to be identified; inputs the driving scene image to be identified into the feature extraction layer of a preset object recognition model for multi-level feature extraction, obtaining first-level driving scene features, second-level driving scene features, and third-level driving scene features corresponding to the driving scene image to be identified; inputs the first-level driving scene features into the first recognition prediction layer of the preset object recognition model, and performs first-category prediction processing based on at least one preset object category to obtain the first predicted object category; inputs the second-level driving scene features and the first predicted object category into the second recognition prediction layer of the preset object recognition model, and performs second-category prediction processing based on at least one object category under each preset object category to obtain the second predicted object category; inputs the third-level driving scene features and the second predicted object category into the third recognition prediction layer of the preset object recognition model, and performs third-category prediction processing based on at least one object category under each first sub-object category to obtain the third predicted object category; based on the third predicted object category, the target object category corresponding to the driving scene image to be identified is determined. This can realize the adjustment of the attention area of ​​the lower level according to the prediction result of the upper-level object category, so that the object recognition is more focused on the area related to the upper level, effectively distinguishing similar object categories and improving the accuracy and reliability of object recognition.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0023] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment; Figure 2 This is a flowchart illustrating an object recognition method according to an exemplary embodiment; Figure 3 This is a flowchart illustrating a pre-trained preset object recognition model according to an exemplary embodiment; Figure 4 This is a schematic diagram illustrating the structure of a preset object recognition model according to an exemplary embodiment; Figure 5 This is a block diagram illustrating an object recognition device according to an exemplary embodiment; Figure 6 This is a block diagram illustrating an electronic device for object recognition according to an exemplary embodiment; Figure 7This is a block diagram illustrating an electronic device for object recognition according to an exemplary embodiment. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0025] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar different contents and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0026] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0027] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment, such as... Figure 1 As shown, the application environment may include terminal 100 and server 200.

[0028] In an optional embodiment, terminal 100 can be used to provide object recognition services. Specifically, terminal 100 can be, but is not limited to, electronic devices such as smartphones, desktop computers, tablets, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices, or software running on the aforementioned electronic devices, such as applications. Optionally, the operating system running on the electronic device can be, but is not limited to, Android, iOS, Linux, Windows, etc.

[0029] In an optional embodiment, server 200 can provide background services to terminal 100 and perform object recognition processing. Optionally, server 200 can pre-train a preset object recognition model. Correspondingly, object recognition processing can be performed in conjunction with the preset object recognition model. Specifically, server 200 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0030] In addition, it should be noted that, Figure 1 The example shown is merely one application environment provided by this disclosure. In practical applications, other application environments may also be included, such as object recognition processing, which can also be implemented on the terminal.

[0031] In the embodiments described in this specification, the terminal 100 and the server 200 can be directly or indirectly connected through wired or wireless communication, and this disclosure does not impose any restrictions.

[0032] Figure 2 This is a flowchart illustrating an object recognition method according to an exemplary embodiment. This object recognition method is used in a terminal or server, and can also be executed interactively by the terminal and server, such as... Figure 2 As shown, it includes the following steps.

[0033] In step S201, the driving scene image to be identified is obtained.

[0034] In one specific embodiment, the driving scene image to be identified can be a driving scene image to be processed for object recognition. Optionally, the driving scene image to be identified can be an environmental image of the vehicle in the driving scene. Accordingly, the objects in the environmental image need to be classified.

[0035] In step S203, the driving scene image to be identified is input into the feature extraction layer of the preset object recognition model for multi-level feature extraction to obtain the first-level driving scene features, the second-level driving scene features and the third-level driving scene features corresponding to the driving scene image to be identified.

[0036] In a specific embodiment, the preset object recognition model can be a model trained by performing object recognition processing on the object recognition model to be trained based on multiple sample driving scene images and multiple preset predicted object categories corresponding to the sample driving scene images. The object recognition model to be trained can be a classification model to be trained. Specifically, the multiple sample driving scene images can be multiple environmental images around the vehicle within a historical time period, and the multiple preset predicted object categories can represent the real object categories corresponding to the multiple preset sample driving scene images.

[0037] In a specific embodiment, the feature extraction layer can be used to extract global driving scene features at different levels corresponding to the driving scene map to be identified. The global scene features can be global features corresponding to the driving scene map to be identified. The feature extraction layer can include a global feature extraction layer, a first-level feature extraction layer, a second-level feature extraction layer, and a third-level feature extraction layer connected to the global feature extraction layer. The first-level feature extraction layer can be used to output first-level driving scene features, the second-level feature extraction layer can be used to output second-level driving scene features, and the third-level feature extraction layer can be used to output third-level driving scene features. The granularity of the first-level driving scene features is larger than that of the second-level driving scene features, and the granularity of the second-level driving scene features is larger than that of the third-level driving scene features. Optionally, the first-level feature extraction layer, the second-level feature extraction layer, and the third-level feature extraction layer can be convolutional kernels of different sizes. Specifically, the size of the convolutional kernel corresponding to the first-level driving scene features can be 7×7, the size of the convolutional kernel corresponding to the second-level driving scene features can be 5×5, and the size of the convolutional kernel corresponding to the third-level driving scene features can be 3×3.

[0038] In step S205, the first-level driving scene features are input into the first recognition prediction layer of the preset object recognition model, and a first category prediction process is performed based on at least one preset object category to obtain the first predicted object category.

[0039] In a specific embodiment, the first recognition prediction layer can be used to perform a first category prediction. At least one preset object category can be a preset object category corresponding to the first category prediction processing. The first predicted object category can be a predicted object category corresponding to the first category prediction processing. Specifically, at least one preset object category can be a first-level object category corresponding to a pre-set driving scene map to be recognized. Specifically, at least one preset object category can include traffic signs, road participants, traffic lights, etc.

[0040] In an optional embodiment, the first recognition prediction layer may include a self-attention layer and a first category output layer. The first recognition prediction layer, which inputs first-level driving scene features into the preset object recognition model, performs first category prediction processing based on at least one preset object category to obtain the first predicted object category, may include: The first-level driving scene features are input into the self-attention layer for self-attention calculation to obtain the first-level driving scene fusion features; The first-level driving scene fusion features are input into the first category output layer for first category prediction processing to obtain first object category indication data corresponding to at least one preset object category. Based on the first preset object category threshold and the first object category indication data, the first predicted object category is obtained.

[0041] In a specific embodiment, the self-attention layer can be used to perform self-attention calculation on the first-level driving scene features. Optionally, the specific structure of the self-attention layer can be a self-attention mechanism. Specifically, the aforementioned first-level driving scene features can be query features, key features, and value features in the self-attention mechanism. Correspondingly, the similarity between the query features and key features can be calculated using dot product, and after Softmax normalization, it is weighted and summed with the value features to obtain the first-level driving scene fusion features.

[0042] In a specific embodiment, the first category output layer can be used to output first object category indication data. This first object category indication data can indicate the probability that the driving scene image to be identified includes objects corresponding to each preset object category. Optionally, the specific structure of the first category output layer can be configured by combining a fully connected layer and a sigmoid function. The fully connected layer can output at least one initial first object category indication data (logits) corresponding to a preset object category. The sigmoid function is used to obtain the first object category indication data corresponding to at least one preset object category. The formula for the sigmoid function is: x represents the initial indication data for the second object category. Specifically, taking at least one preset object category as traffic signs, road participants, traffic lights, etc., and the threshold for the first preset object category as 0.9, the first number output by the first category output layer can be preset to correspond to the preset object category of traffic signs, the second number output by the first category output layer to correspond to the preset object category of road participants, and the third number output by the first category output layer to correspond to the preset object category of traffic lights. The initial indication data for the first object category can be [2.5, -1.2, 3.1], and the indication data for the first object category can be [0.92, 0.23, 0.96]. Correspondingly, the first predicted object category can be [1, 0, 1].

[0043] In the above embodiments, the first set of driving scene feature representation is enhanced by self-attention, probabilistic indication data is generated to achieve refined object category judgment, and threshold filtering is combined to ensure the accuracy of object recognition.

[0044] In step S207, the second-level driving scene features and the first predicted object category are input into the second recognition prediction layer of the preset object recognition model, and the second category prediction processing is performed based on at least one first sub-object category to obtain the second predicted object category.

[0045] In a specific embodiment, the second recognition prediction layer can be used to perform a second category prediction. At least one first sub-object category can be at least one object category under each preset object category. The second predicted object category can be the predicted object category corresponding to the second category prediction processing. Specifically, at least one first sub-object category can be a second-level object category corresponding to a pre-set driving scene map to be recognized. Specifically, at least one first sub-object category can include warning signs, prohibition signs, instruction signs, etc. (traffic signs), pedestrians, motor vehicles, non-motor vehicles, etc. (road participants), motor vehicle traffic lights, pedestrian traffic lights, non-motor vehicle traffic lights, etc. (traffic lights).

[0046] In an optional embodiment, the second recognition prediction layer may include a first cross-attention layer and a second category output layer. The process of inputting the second-level driving scene features and the first predicted object category into the second recognition prediction layer of the preset object recognition model, and performing second category prediction processing based on at least one first sub-object category to obtain the second predicted object category may include: The second-level driving scene features and the first predicted object category are input into the first cross-attention layer for cross-attention calculation to obtain the second-level driving scene fusion features. The second-level driving scene fusion features are input into the second-class output layer for second-class prediction processing to obtain second-object category indication data corresponding to at least one first sub-object category; The second predicted object category is obtained based on the second preset object category threshold and the second object category indication data.

[0047] In a specific embodiment, the first cross-attention layer can be used to perform cross-attention calculation on the second-level driving scene features and the first predicted object category. Optionally, the specific structure of the first cross-attention layer can be a cross-attention mechanism. Specifically, the first predicted object category can be the query feature in the cross-attention mechanism, and the second-level driving scene features can be key features and value features. Correspondingly, the similarity between the query feature and the key feature can be calculated using dot product, and after Softmax normalization, it is weighted and summed with the value feature to obtain the second-level driving scene fusion feature.

[0048] In one specific embodiment, the second category output layer can be used to output second object category indication data. This second object category indication data can indicate the probability that the driving scene map to be identified includes objects corresponding to each first sub-object category. Optionally, the specific structure of the second category output layer can be configured by combining a fully connected layer and a sigmoid function. The fully connected layer can output at least one initial second object category indication data (logits) corresponding to a first sub-object category, and the sigmoid function can obtain the second object category indication data corresponding to at least one first sub-object category. Specifically, at least one first sub-object category is used as a warning. Taking signs, prohibition signs, pedestrians, and motor vehicles as an example, with a second preset object category threshold of 0.9, the first number output by the second category output layer can be preset to correspond to the first sub-object category of warning signs, the second number output by the second category output layer to correspond to the first sub-object category of prohibition signs, the third number output by the second category output layer to correspond to the first sub-object category of pedestrians, and the fourth number output by the second category output layer to correspond to the first sub-object category of motor vehicles. The initial indication data of the second object category can be [2.5, -1.2, 3.1, 0.8], and the second object category indication data can be [0.92, 0.23, 0.96, 0.69]. Correspondingly, the second predicted object category can be [1, 0, 1, 0].

[0049] In the above embodiments, the cross-attention mechanism of guiding the second-level classification through the first-level classification achieves hierarchical focus, ensures the logical consistency of the three-level classification, greatly improves the recognition accuracy of the first sub-object category, and adapts to the decision-making requirements of hierarchical consistency in driving scenarios.

[0050] In step S209, the third-level driving scene features and the second predicted object category are input into the third recognition prediction layer of the preset object recognition model, and the third category prediction processing is performed based on at least one second sub-object category to obtain the third predicted object category.

[0051] In a specific embodiment, the third recognition prediction layer can be used to perform third category prediction. At least one second sub-object category can be at least one object category under each first sub-object category. The third predicted object category can be the predicted object category corresponding to the second category prediction processing. Specifically, at least one second sub-object category can be a third-level object category corresponding to a pre-set driving scenario map to be recognized. Specifically, at least one second sub-object category can include construction warnings, intersection warnings, etc. (warning signs), no parking, no U-turn, etc. (prohibition signs), straight-ahead instructions, left-turn instructions, right-turn instructions, etc. (instruction signs), ordinary pedestrians, children, the elderly, etc. (pedestrians), cars, buses, police cars, etc. (motor vehicles), bicycles, tricycles, etc. (non-motor vehicles), red lights, green lights, yellow lights, etc. (motor vehicle traffic lights), pedestrian red lights, pedestrian green lights, etc. (pedestrian traffic lights), non-motor vehicle red lights, non-motor vehicle yellow lights, non-motor vehicle green lights, etc. (non-motor vehicle traffic lights).

[0052] In an optional embodiment, the third recognition prediction layer may include a second cross-attention layer and a third category output layer. The second recognition prediction layer, which inputs the second-level driving scene features and the first predicted object category into the preset object recognition model, performs second category prediction processing based on at least one first sub-object category to obtain the second predicted object category, may include: The third-level driving scene features and the second predicted object category are input into the second cross-correlation layer for cross-attention calculation to obtain the third-level driving scene fusion features. The third-level driving scene fusion features are input into the third-category output layer for third-category prediction processing to obtain third-category indication data corresponding to at least one second sub-object category. Based on the third preset object category threshold and the third object category indicator data, the third predicted object category is obtained.

[0053] In a specific embodiment, the second cross-attention layer can be used to perform cross-attention calculation on the third-level driving scene features and the second predicted object category. Optionally, the specific structure of the second cross-attention layer can be a cross-attention mechanism. Specifically, the second predicted object category can be the query feature in the cross-attention mechanism, and the third-level driving scene features can be key features and value features. Correspondingly, the similarity between the query feature and the key feature can be calculated using dot product, and after Softmax normalization, it is weighted and summed with the value feature to obtain the third-level driving scene fusion feature.

[0054] In one specific embodiment, the third category output layer can be used to output third object category indication data. This third object category indication data can indicate the probability that the driving scene map to be identified includes objects corresponding to each second sub-object category. Optionally, the specific structure of the third category output layer can be configured by combining a fully connected layer and a sigmoid function. The fully connected layer can output initial third object category indication data (logits) corresponding to at least one second sub-object category. The sigmoid function is then used to obtain the third object category indication data corresponding to at least one second sub-object category. Specifically, at least one second sub-object category can be selected as construction warning, no parking, ordinary pedestrian, bus, or pedestrian green light. Taking a third preset object category threshold of 0.9 as an example, the first number output by the third category output layer can be preset to correspond to the second sub-object category of construction warning, the second number output by the third category output layer to correspond to the second sub-object category of no parking, the third number output by the third category output layer to correspond to the second sub-object category of ordinary pedestrian, the fourth number output by the third category output layer to correspond to the second sub-object category of bus, and the fifth number output by the third category output layer to correspond to the second sub-object category of pedestrian green light. The initial indication data of the third object category can be [2.5, -1.2, 3.1, 0.8, 2.0], and the indication data of the third object category can be [0.92, 0.23, 0.96, 0.69, 0.88]. Correspondingly, the third predicted object category can be [1, 0, 1, 0, 0].

[0055] In the above embodiments, the cross-attention mechanism is used to determine the feature recognition range of the third-level driving scene, so that the recognition results are accurate to the smallest granularity required for driving decision-making, thus ensuring the safety and reliability of driving decision-making.

[0056] In step S2011, the target object category corresponding to the driving scene map to be identified is determined based on the third predicted object category.

[0057] In one specific embodiment, the target object category can be the object category corresponding to the identified driving scene map.

[0058] In an optional embodiment, the target object category may include at least one object category corresponding to each object, and the method may further include: If the preset object category corresponding to the current object is not included in the first predicted object category, the object category corresponding to the current object is corrected.

[0059] In a specific embodiment, the current object can be any of at least one object. Taking at least one object as a police car, bus, ordinary pedestrian, pedestrian green light, and no U-turn as an example, the first predicted object category can be road participants and traffic lights. Correspondingly, the preset object category corresponding to the current object includes road participants, traffic lights, and traffic signs. The preset object category corresponding to the current object is not included in the first predicted object category. Specifically, the object category corresponding to the current object is modified to police car, bus, ordinary pedestrian, and pedestrian green light.

[0060] In the above embodiments, by verifying the hierarchical relationship from superior to subordinate, logically contradictory object recognition results are filtered out, ensuring that the final output object category is both accurate and conforms to the semantic logic of the driving scenario, thereby improving the reliability of the model output and the safety of decision-making.

[0061] In an optional embodiment, the above method may further include: a step of pre-training a preset object recognition model, specifically, as follows: Figure 3 As shown, the following steps may be included: In step S301, the current sample driving scene map and the preset prediction object category corresponding to the current sample driving scene map are obtained; In a specific embodiment, the current sample driving scene map can be the training data of the current training cycle. Specifically, the current sample driving scene map can be randomly determined from multiple sample driving scene maps, or a portion of the sample driving scene maps that have not participated in model training can be randomly selected from multiple sample driving scene maps as the current sample driving scene map; the sample predicted object category can be used to indicate the object category predicted by the current object to be trained.

[0062] In step S303, the current sample driving scene image is input into the object recognition model to be trained for object recognition to obtain the predicted object category of the sample; In a specific embodiment, determining the object recognition loss based on the predicted object category of the sample and the preset predicted object category may include combining a preset loss function to determine the object recognition loss between the predicted object category of the sample and the preset predicted object category; specifically, the object recognition loss can characterize the object recognition performance of the current object recognition model to be trained. Specifically, the preset loss function can be set to... ,in, The corresponding cross-entropy can be processed for the first category prediction. The corresponding cross-entropy can be processed for the second category prediction. The corresponding cross-entropy can be processed for third-category prediction. The weights corresponding to the cross-entropy can be predicted for the first category. The weights corresponding to the cross-entropy can be processed for the second category prediction. The weights of the cross-entropy can be used to predict the third category.

[0063] In step S305, the object recognition loss is determined based on the sample predicted object category and the preset predicted object category; In a specific embodiment, training the object recognition model to be trained based on object recognition loss to obtain a preset object recognition model may include: combining the object recognition loss to update the model parameters of the object recognition model to be trained, and based on the updated object recognition model to be trained, repeatedly determining the current sample driving scene map and the preset predicted object category corresponding to the current sample driving scene map from multiple sample driving scene maps, and updating the model parameters of the object recognition model to be trained in a loop iteration operation until a preset convergence condition is met, and the object recognition model to be trained corresponding to the preset convergence condition is the preset object recognition model.

[0064] In step S307, the object recognition model to be trained is trained based on the object recognition loss to obtain the preset object recognition model.

[0065] In a specific embodiment, the preset convergence conditions can be set according to the actual application, such as the number of iterations of the loop operation reaching a preset number, the change detection loss being less than a specified threshold, etc., which can be set according to the training speed and model accuracy requirements.

[0066] In a specific embodiment, such as Figure 4 As shown, Figure 4 This is a schematic diagram of the structure of a preset object recognition model provided in an embodiment of this application. Specifically, in conjunction with... Figure 4As can be seen, the driving scene image to be identified is input into a shared feature extraction layer, which includes a global feature extraction layer E, and a first-level feature extraction layer (Conv 1), a second-level feature extraction layer (Conv 2), and a third-level feature extraction layer (Conv 3) connected to the global feature extraction layer, to obtain driving scene features. These driving scene features are then input into three convolutional kernels of different sizes. Specifically, the granularity of the first-level driving scene feature M1 output by convolutional kernel 1 (Conv 1) is larger than that of the second-level driving scene feature M2 output by convolutional kernel 2 (Conv 2), and the granularity of the second-level driving scene feature M2 output by convolutional kernel 2 (Conv 2) is larger than that of convolutional kernel 3 (Conv 3). 3) The output of the third-level driving scene feature M3: Further, given the three-level driving scene features with different granularities, each level of driving scene feature is used as the input to three sequentially cascaded recognition and prediction layers (HGM1, HGM2, and HGM3). Each recognition and prediction layer contains a sequentially connected self-attention layer (or cross-attention layer) and a category output layer. After combining the cross-attention layer with the weighted fusion of each level of driving scene features and the previous level of driving scene features (or the self-attention layer with the weighted fusion of each level of driving scene features), the category output layer is used to predict the category of the driving scene fusion features (M1', M2', M3'). The object category is then determined by combining the preset object category threshold with the object category indication data output by the category output layer. This predicted object category (Output1 and Output2) is then used as an input to the next layer of cross-attention layer. In this way, the driving scene fusion features of each layer of cross-attention layer contain the predicted object category information of the previous layer.

[0067] In the above embodiments, a preset object recognition model is trained by the current sample driving scene map and the preset predicted object category corresponding to the current sample driving scene map, which ensures the learning direction and recognition accuracy of the model and achieves accurate recognition of driving scenes.

[0068] In an optional embodiment, the preset predicted object category may include a first preset predicted object category, a second preset predicted object category, and a third preset predicted object category; the sample predicted object category may include a first sample predicted object category, a second sample predicted object category, and a third sample predicted object category; and determining the object recognition loss based on the sample predicted object category and the preset predicted object category may include: The first object recognition loss is determined based on the first sample predicted object category and the first preset predicted object category; The second object recognition loss is determined based on the predicted object category of the second sample and the second preset predicted object category. The third object recognition loss is determined based on the predicted object category of the third sample and the third preset predicted object category. The object recognition loss is obtained by weighting the first object recognition loss, the second object recognition loss, and the third object recognition loss.

[0069] In a specific embodiment, the first preset predicted object category can represent the real object category corresponding to the pre-set first sample predicted object category, the second preset predicted object category can represent the real object category corresponding to the pre-set second sample predicted object category, and the third preset predicted object category can represent the real object category corresponding to the pre-set third sample predicted object category. The first sample predicted object category can be the predicted object category of the current sample driving scene image under at least one preset object, the second sample predicted object category can be the predicted object category of the current sample driving scene image under at least one first sub-object, and the third sample predicted object category can be the predicted object category of the current sample driving scene image under at least one second sub-object.

[0070] In a specific embodiment, the first object recognition loss can characterize the difference between the first sample predicted object category and the first preset predicted object category, the second object recognition loss can characterize the difference between the second sample predicted object category and the second preset predicted object category, and the third object recognition loss can characterize the difference between the third sample predicted object category and the third preset predicted object category. Optionally, the weights corresponding to the first object recognition loss, the second object recognition loss, and the third object recognition loss can be adjusted according to actual conditions.

[0071] In the above embodiments, the object recognition loss is determined based on the weights of the loss at each level, thereby improving the model training accuracy and the object recognition accuracy of the model. Furthermore, by adjusting the weighting coefficients, it can be adapted to different vehicle scenarios without retraining the entire model, thus reducing the training intensity of the model.

[0072] As can be seen from the technical solutions provided in the embodiments of this specification above, this specification obtains a driving scene image to be identified; inputs the driving scene image to be identified into the feature extraction layer of a preset object recognition model for multi-level feature extraction, obtaining first-level driving scene features, second-level driving scene features, and third-level driving scene features corresponding to the driving scene image to be identified; inputs the first-level driving scene features into the first recognition prediction layer of the preset object recognition model, performs first-category prediction processing based on at least one preset object category, and obtains the first predicted object category; inputs the second-level driving scene features and the first predicted object category into the second recognition prediction layer of the preset object recognition model, performs second-category prediction processing based on at least one object category under each preset object category, and obtains the second predicted object category; inputs the third-level driving scene features and the second predicted object category into the third recognition prediction layer of the preset object recognition model, performs third-category prediction processing based on at least one object category under each first sub-object category, and obtains the third predicted object category; based on the third predicted object category, determines the target object category corresponding to the driving scene image to be identified. This can realize the adjustment of the attention area of ​​the lower level according to the prediction result of the upper-level object category, so that the object recognition is more focused on the area related to the upper level, effectively distinguishing similar object categories, and improving the accuracy and reliability of object recognition.

[0073] Figure 5 This is a block diagram illustrating an object recognition device according to an exemplary embodiment. (Refer to...) Figure 5 The device includes: The driving scene image acquisition module 510 is used to acquire the driving scene image to be identified. The multi-level feature extraction module 530 is used to input the driving scene image to be identified into the feature extraction layer in the preset object recognition model for multi-level feature extraction, and obtain the first-level driving scene features, the second-level driving scene features and the third-level driving scene features corresponding to the driving scene image to be identified; The first category prediction processing module 550 is used to input the first-level driving scene features into the first recognition prediction layer in the preset object recognition model, and perform first category prediction processing based on at least one preset object category to obtain the first predicted object category. The second category prediction processing module 570 is used to input the second-level driving scene features and the first predicted object category into the second recognition prediction layer in the preset object recognition model, and perform second category prediction processing based on at least one first sub-object category to obtain the second predicted object category; at least one first sub-object category is at least one object category under each preset object category. The third category prediction processing module 590 is used to input the third-level driving scene features and the second predicted object category into the third recognition prediction layer of the preset object recognition model, and perform third category prediction processing based on at least one second sub-object category to obtain the third predicted object category; at least one second sub-object category is at least one object category under each first sub-object category. The target object category determination module 5110 is used to determine the target object category corresponding to the driving scene map to be identified based on the third predicted object category.

[0074] In an optional embodiment, the first recognition prediction layer includes a self-attention layer and a first category output layer, and the first category prediction processing module 550 includes: The self-attention computation unit is used to input the first-level driving scene features into the self-attention layer for self-attention computation to obtain the first-level driving scene fusion features. The first category prediction processing unit is used to input the first-level driving scene fusion features into the first category output layer for first category prediction processing to obtain first object category indication data corresponding to at least one preset object category; the first object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each preset object category. The first predicted object category determination unit is used to obtain the first predicted object category based on the first preset object category threshold and the first object category indication data.

[0075] In an optional embodiment, the second recognition prediction layer includes a first cross-attention layer and a second category output layer, and the second category prediction processing module 570 includes: The first cross-attention calculation unit is used to input the second-level driving scene features and the first predicted object category into the first cross-attention layer to perform cross-attention calculation and obtain the second-level driving scene fusion features. The second category prediction processing unit is used to input the second-level driving scene fusion features into the second category output layer for second category prediction processing to obtain second object category indication data corresponding to at least one first sub-object category; the second object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each first sub-object category. The second predicted object category determination unit is used to obtain the second predicted object category based on the second preset object category threshold and the second object category indication data.

[0076] In an optional embodiment, the third recognition prediction layer includes a second cross-attention layer and a third category output layer, and the third category prediction processing module 590 includes: The second cross-attention calculation unit is used to input the third-level driving scene features and the second predicted object category into the second cross-correlation layer to perform cross-attention calculation and obtain the third-level driving scene fusion features. The third category prediction processing unit is used to input the third-level driving scene fusion features into the third category output layer for third category prediction processing to obtain third object category indication data corresponding to at least one second sub-object category; the third object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each second sub-object category. The third predicted object category determination unit is used to obtain the third predicted object category based on the third preset object category threshold and the third object category indication data.

[0077] In an optional embodiment, the above-described apparatus further includes: The preset prediction object category acquisition module is used to acquire the current sample driving scene image and the preset prediction object category corresponding to the current sample driving scene image; The object recognition module is used to input the current sample driving scene image into the object recognition model to be trained for object recognition and obtain the predicted object category of the sample; The object recognition loss determination module is used to determine the object recognition loss based on the predicted object category of the sample and the preset predicted object category. The model training module is used to train the object recognition model to be trained based on the object recognition loss, so as to obtain the preset object recognition model.

[0078] In an optional embodiment, the preset predicted object categories include a first preset predicted object category, a second preset predicted object category, and a third preset predicted object category; the sample predicted object categories include a first sample predicted object category, a second sample predicted object category, and a third sample predicted object category; and the object recognition loss determination module includes: The first object recognition loss determination unit is used to determine the first object recognition loss based on the first sample predicted object category and the first preset predicted object category. The second object recognition loss determination unit is used to determine the second object recognition loss based on the predicted object category of the second sample and the second preset predicted object category. The third object recognition loss determination unit is used to determine the third object recognition loss based on the predicted object category of the third sample and the third preset predicted object category. The object recognition loss determination unit is used to perform weighted processing on the first object recognition loss, the second object recognition loss and the third object recognition loss to obtain the object recognition loss.

[0079] In an optional embodiment, the target object category includes at least one object category corresponding to each object, and the above-described apparatus further includes: The object category correction module is used to correct the object category of the current object when the preset object category corresponding to the current object is not included in the first predicted object category; the current object is any one of at least one objects.

[0080] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0081] Figure 6 This is a block diagram illustrating an electronic device for object recognition according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an object recognition method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse. Figure 7 This is a block diagram illustrating an electronic device for object recognition according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an object recognition method. Those skilled in the art will understand that Figure 6 or Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements. In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the object recognition method as described in the embodiments of this disclosure.

[0082] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the object recognition method of the present disclosure.

[0083] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0084] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An object recognition method, characterized in that, include: Obtain the image of the driving scene to be identified; The driving scene image to be identified is input into the feature extraction layer of the preset object recognition model for multi-level feature extraction to obtain the first-level driving scene features, the second-level driving scene features and the third-level driving scene features corresponding to the driving scene image to be identified; The first-level driving scene features are input into the first recognition prediction layer of the preset object recognition model, and a first category prediction process is performed based on at least one preset object category to obtain the first predicted object category. The second-level driving scenario features and the first predicted object category are input into the second recognition prediction layer of the preset object recognition model. The second category prediction processing is performed based on at least one first sub-object category to obtain the second predicted object category. The at least one first sub-object category is at least one object category under each preset object category. The third-level driving scenario features and the second predicted object category are input into the third recognition prediction layer of the preset object recognition model, and the third category prediction processing is performed based on at least one second sub-object category to obtain the third predicted object category. The at least one second sub-object category is at least one object category under each first sub-object category; Based on the third predicted object category, the target object category corresponding to the driving scene map to be identified is determined.

2. The object recognition method according to claim 1, characterized in that, The first recognition and prediction layer includes a self-attention layer and a first category output layer. The first-level driving scene features are input into the first recognition and prediction layer of the preset object recognition model, and a first category prediction process is performed based on at least one preset object category to obtain the first predicted object category, including: The first-level driving scene features are input into the self-attention layer for self-attention calculation to obtain the first-level driving scene fusion features. The first-level driving scene fusion feature is input into the first category output layer for first category prediction processing to obtain first object category indication data corresponding to the at least one preset object category; the first object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each preset object category. The first predicted object category is obtained based on the first preset object category threshold and the first object category indication data.

3. The object recognition method according to claim 1, characterized in that, The second recognition prediction layer includes a first cross-attention layer and a second category output layer. The second-level driving scene features and the first predicted object category are input into the second recognition prediction layer of the preset object recognition model. Second category prediction processing is performed based on at least one first sub-object category to obtain the second predicted object category, including: The second-level driving scene features and the first predicted object category are input into the first cross-attention layer for cross-attention calculation to obtain the second-level driving scene fusion features. The second-level driving scene fusion features are input into the second category output layer for second category prediction processing to obtain second object category indication data corresponding to the at least one first sub-object category; the second object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each first sub-object category. The second predicted object category is obtained based on the second preset object category threshold and the second object category indication data.

4. The object recognition method according to claim 1, characterized in that, The third recognition and prediction layer includes a second cross-attention layer and a third category output layer. The third-level driving scene features and the second predicted object category are input into the third recognition and prediction layer of the preset object recognition model. Third category prediction processing is performed based on at least one second sub-object category to obtain the third predicted object category, which includes: The third-level driving scene features and the second predicted object category are input into the second cross-correlation layer for cross-attention calculation to obtain the third-level driving scene fusion features. The third-level driving scene fusion features are input into the third-category output layer for third-category prediction processing to obtain third-object category indication data corresponding to the at least one second sub-object category; the third-object category indication data is used to indicate the probability that the driving scene map to be identified includes the object corresponding to each second sub-object category. The third predicted object category is obtained based on the third preset object category threshold and the third object category indication data.

5. The object recognition method according to claim 1, characterized in that, The method further includes: Obtain the current sample driving scene image and the preset prediction object category corresponding to the current sample driving scene image; The current sample driving scene image is input into the object recognition model to be trained to perform object recognition and obtain the predicted object category of the sample. The object recognition loss is determined based on the predicted object category of the sample and the preset predicted object category; Based on the object recognition loss, the object recognition model to be trained is trained to obtain the preset object recognition model.

6. The object recognition method according to claim 5, characterized in that, The preset predicted object categories include a first preset predicted object category, a second preset predicted object category, and a third preset predicted object category. The sample predicted object categories include a first sample predicted object category, a second sample predicted object category, and a third sample predicted object category. Determining the object recognition loss based on the sample predicted object categories and the preset predicted object categories includes: The first object recognition loss is determined based on the first sample predicted object category and the first preset predicted object category; The second object recognition loss is determined based on the second sample predicted object category and the second preset predicted object category; The third object recognition loss is determined based on the predicted object category of the third sample and the predicted object category of the third preset sample. The object recognition loss is obtained by weighting the first object recognition loss, the second object recognition loss, and the third object recognition loss.

7. The object recognition method according to claim 1, characterized in that, The target object category includes at least one object category corresponding to each object, and the method further includes: If the preset object category corresponding to the current object is not included in the first predicted object category, the object category corresponding to the current object is corrected; the current object is any one of the at least one objects.

8. An object recognition device, characterized in that, include: The module for acquiring the driving scene image to be identified is used to acquire the driving scene image to be identified. The multi-level feature extraction module is used to input the driving scene image to be identified into the feature extraction layer of the preset object recognition model for multi-level feature extraction, so as to obtain the first-level driving scene features, the second-level driving scene features and the third-level driving scene features corresponding to the driving scene image to be identified; The first category prediction processing module is used to input the first-level driving scene features into the first recognition prediction layer in the preset object recognition model, and perform first category prediction processing based on at least one preset object category to obtain the first predicted object category. The second category prediction processing module is used to input the second-level driving scene features and the first predicted object category into the second recognition prediction layer of the preset object recognition model, and perform second category prediction processing based on at least one first sub-object category to obtain the second predicted object category; the at least one first sub-object category is at least one object category under each preset object category. The third category prediction processing module is used to input the third-level driving scenario features and the second predicted object category into the third recognition prediction layer of the preset object recognition model, and perform third category prediction processing based on at least one second sub-object category to obtain the third predicted object category. The at least one second sub-object category is at least one object category under each first sub-object category; The target object category determination module is used to determine the target object category corresponding to the driving scene map to be identified based on the third predicted object category.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the object recognition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the object recognition method as described in any one of claims 1 to 7.

Images