Detection method, device and roadside unit

Abstract

The embodiment of the present application provides a detection method, device and roadside unit. Among them, the first roadside unit determines the first image acquired, extracts the first image feature corresponding to the first object from the first image, and sends the first image feature to the second roadside unit, and the second roadside unit determines whether the first object and the second object are the same object based on the first image feature and the second image feature of the second object extracted from the second image. The technical solution provided by the embodiment of the present application realizes efficient object recognition.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of Internet technology, and in particular to a detection method, device and roadside unit. Background Art

[0002] A roadside unit (RSU) is a device deployed on both sides of the road in an intelligent transportation system to identify objects such as vehicles traveling on the road for tracking purposes. Therefore, there is a need to track the same object between different roadside units.

[0003] At present, roadside units are usually connected to one or more cameras to collect images within the road section covered by the roadside unit. Object tracking can be achieved by identifying whether the same object exists in the corresponding images of two roadside units. Therefore, how to efficiently realize object recognition has become a technical problem that technical personnel in this field need to solve. Summary of the invention

[0004] The embodiments of the present application provide a detection method, device and road side unit (English: Road Side Unit, abbreviated as: RSU) for realizing efficient object recognition.

[0005] In a first aspect, an embodiment of the present application provides a detection method, which is applied to a first roadside unit RSU, including:

[0006] Determine a first image acquired;

[0007] extracting a first image feature corresponding to a first object from the first image;

[0008] sending the first image feature to a second RSU;

[0009] The first image feature is combined with a second image feature of a second object extracted from a second image by the second RSU to determine whether the first object and the second object are the same object.

[0010] In a second aspect, an embodiment of the present application provides a detection method, which is applied to a second RSU, and the method includes:

[0011] Receiving a first image feature of a first object sent by a first RSU; wherein the first image feature is extracted from the first image;

[0012] Determine the acquired second image;

[0013] extracting first image features corresponding to the second object from the second image;

[0014] Based on the first image feature and the second image feature, it is determined whether the first object and the second object are the same object.

[0015] In a third aspect, an embodiment of the present application provides a detection device, comprising:

[0016] A first image acquisition module, used to determine a first image acquired;

[0017] A first feature extraction module, used to extract a first image feature corresponding to a first object from the first image;

[0018] A first feature sending module, configured to send the first image feature to a second RSU;

[0019] The first image feature is combined with a second image feature of a second object extracted from a second image by the second RSU to determine whether the first object and the second object are the same object.

[0020] In a fourth aspect, an embodiment of the present application provides a detection device, comprising:

[0021] A first feature receiving module is used to receive a first image feature of a first object sent by a first RSU; wherein the first image feature is extracted from the first image;

[0022] A second image acquisition module, used for image determination and acquisition of a second image;

[0023] A second feature extraction module, used to extract first image features corresponding to a second object from the second image;

[0024] The first object recognition module is used to determine whether the first object and the second object are the same object based on the first image feature and the second image feature.

[0025] In a fifth aspect, an embodiment of the present application provides a roadside unit, including a storage component and a processing component;

[0026] Wherein, the storage component stores one or more computer instructions; when the one or more computer instructions are called and executed by the processing component, the processing component executes the detection method as described in any one of claims 1 to 18.

[0027] In a sixth aspect, an embodiment of the present application provides a roadside unit, including a storage component and a processing component;

[0028] Wherein, the storage component stores one or more computer instructions; when the one or more computer instructions are called and executed by the processing component, the processing component executes the detection method as claimed in claim 19.

[0029] In an embodiment of the present application, the first RSU identifies the first object in the first image acquired and extracts the first image features corresponding to the first object, and sends the first image features to the second RSU. The second RSU extracts the second image features corresponding to the second object from the second image acquired, and based on the first image features and the second image features transmitted by the first RSU, it can be identified whether the first object and the second object are the same object. The first RSU and the second RSU are any two adjacent RSUs. Object recognition is performed locally by the RSU, and only image features are transmitted between the two RSUs, thereby reducing the amount of data transmission, thereby achieving efficient and accurate object recognition. Efficient and accurate object tracking is achieved.

[0030] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 A flow chart of an embodiment of a detection method provided by the present application is shown;

[0033] Figure 2 A schematic diagram of the model structure of the feature coding model of an embodiment of the present application in a practical application is shown;

[0034] Figure 3 A flow chart showing another embodiment of a detection method provided by the present application is shown;

[0035] Figure 4 A schematic diagram showing the deployment of some roadside units in a road section in an actual application of an embodiment of the present application is shown;

[0036] Figure 5 A schematic structural diagram of an embodiment of a detection device provided by the present application is shown;

[0037] Figure 6 A schematic structural diagram of an embodiment of a roadside unit provided by the present application is shown;

[0038] Figure 7 A schematic structural diagram of another embodiment of a detection device provided by the present application is shown;

[0039] Figure 8 A schematic structural diagram of another embodiment of a roadside unit provided in the present application is shown. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0041] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0042] The technical solution of the embodiment of the present application is mainly used in intelligent transportation scenarios. Relying on wireless communication technology and Internet technology, intelligent transportation technology is gradually emerging. In order to realize intelligent transportation management, road side units (English: Road Side Unit, referred to as: RSU) are usually deployed in different sections on both sides of the road. RSU is deployed on both sides of the road and can be connected to one or more acquisition units, such as cameras, so that RSU can obtain the image collected by the acquisition unit, and one acquisition unit can be connected to at least one RSU, that is, the acquisition range can cover the coverage range of at least one RSU.

[0043] In order to track objects such as vehicles traveling on the road, the inventors thought that a feasible way is for the RSU to upload the collected images to the cloud, that is, the server, and the cloud will perform object recognition and then inform the RSU of the recognition results. This requires the cloud to perform a large amount of calculation and storage. If the cloud processing or storage capacity is low, it will affect the efficiency of object recognition.

[0044] After further research, the inventor found that in order to reduce the processing pressure on the cloud, the RSU can perform local processing. Since object tracking needs to be achieved between different RSUs, one RSU can pass the image it has collected to the next RSU adjacent to it, and the next RSU will perform object recognition. However, the inventor also found that although object recognition by the RSU can reduce the processing pressure on the cloud, the transmission of images or videos between RSUs will take up a lot of bandwidth and may even be impossible to complete in a short period of time. Based on this problem, the inventor further thought about it and proposed the technical solution of the embodiment of the present application. In the embodiment of the present application, object recognition is performed by the RSU without the need for cloud recognition, and images are not transmitted directly between RSUs, but image features extracted from the images are transmitted, thereby reducing the amount of data transmission, reducing transmission pressure and reducing hardware requirements. Therefore, efficient and accurate object recognition can be achieved, thereby improving the efficiency and accuracy of object tracking.

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0046] Figure 1 This is a flow chart of an embodiment of a detection method provided in an embodiment of the present application. The technical solution of this embodiment can be executed by a first RSU, and the first RSU can refer to any RSU. The method may include the following steps:

[0047] 101: Determine a first image acquired through acquisition.

[0048] Optionally, the first image may be any image acquired by at least one acquisition unit connected to the first RSU. The acquisition unit may be a capture unit for capturing video data, and thus, determining the acquired first image may include:

[0049] Any video frame acquired by at least one acquisition unit connected to the first RSU is used as a first image.

[0050] Optionally, in actual applications, one acquisition unit may be connected to multiple RSUs, and the gap between video frames continuously acquired by one acquisition unit may not be particularly large. Therefore, a video frame may be acquired at a predetermined interval from the video acquired by each acquisition unit connected to the first RSU as the first image.

[0051] 102: Extracting a first image feature corresponding to a first object from the first image.

[0052] Among them, it may be that object detection is first performed on the first image to determine the first object, and then the first image feature corresponding to the first object is extracted. Therefore, the first image feature corresponding to the first object extracted from the first image may include:

[0053] detecting an object position of a first object in the first image;

[0054] extracting a first image region where an object position of the first object is located from the first image;

[0055] A first image feature of the first image region is extracted.

[0056] Of course, it is also possible to directly identify the first object from the first image and extract the first image features corresponding to the first object.

[0057] A plurality of objects may be identified in the first image, and the first object may refer to any one of the objects.

[0058] In intelligent traffic scenarios, objects may refer to vehicles or pedestrians traveling on the road, etc.

[0059] Among them, in the embodiments of the present application, there are multiple ways to implement image feature extraction, for example, the traditional Harris operator can be used to extract Harris features, of course, other image features, such as edge features, color features, motion features, etc., or features obtained by weighted processing of multiple features. In addition, in order to further improve the accuracy, a neural network model can be used for extraction, so the image features can be neural network features, which will be introduced in detail in the following embodiments.

[0060] In addition, the first object may be identified first, and a corresponding extraction strategy may be determined according to the identification result, so that the first image features corresponding to the first object may be extracted from the first image according to the extraction strategy. The extraction strategy may include, for example, the type of features to be extracted and / or the number of features, etc.

[0061] Therefore, the extracting the first image feature corresponding to the first object from the first image includes:

[0062] detecting a first object in the first image;

[0063] Identify the first object to obtain an identification result;

[0064] Determining an extraction strategy corresponding to the recognition result;

[0065] According to the extraction strategy, a first image feature corresponding to the first object is extracted.

[0066] Among them, detecting the first object in the first image can be implemented by using an object detection model to determine the object position of the first object in the first image and then identify it. The specific implementation will be described in the following embodiments.

[0067] Among them, identifying the first object and obtaining the identification result package may be identifying the color, category or brand of the first object and obtaining the identification result; wherein different colors, different categories or different brands correspond to different extraction strategies, and thus the corresponding extraction strategy can be determined based on the identification result.

[0068] Since the extraction strategy may include the type of extracted features, etc., therefore, optionally, extracting the first image feature corresponding to the first object according to the extraction strategy may include:

[0069] Extracting, according to the extraction strategy, image features of at least one feature type corresponding to the first object;

[0070] The image features of the at least one feature type are fused to obtain a first image feature corresponding to the first object.

[0071] The fusion of image features of at least one feature type may be, for example, weighted averaging.

[0072] For example, the color of the first object can be recognized. For the first object of the target color (such as the color that appears more frequently on the road), multiple types of features can be extracted. For example, there are more gray or silver vehicles on the road. In order to effectively distinguish different vehicles that are gray at the same time, it is necessary to collect more image features of feature types. The image features of the multiple feature types can include neural network features, edge features, motion features, etc., and then the image features of the multiple feature types are weighted averaged to obtain the first image features. For the first object of non-target color, only neural network features can be extracted. For another example, when the first object is a vehicle, the brand of the first object can be identified. For vehicles of a specific brand (such as a brand that appears more frequently on the road), multiple types of features can be extracted to achieve effective distinction of the first object, etc. For another example, the model of the first object can be identified. For small cars, since they appear more frequently on the road, multiple types of features can be extracted, etc., so as to effectively distinguish them from other objects.

[0073] 103: Send the first image feature to the second RSU.

[0074] The first image feature is combined with a second image feature of a second object extracted from a second image by the second RSU to determine whether the first object and the second object are the same object.

[0075] The second RSU can determine whether the first object and the second object are the same by comparing the first image features and the second image features, and wirelessly upload the images to the cloud for identification. The identification can be achieved locally, and if the first object and the second object are the same, object tracking can be achieved.

[0076] The second RSU is adjacent to the first RSU, which may refer to the next RSU adjacent to the first RSU.

[0077] In this embodiment, the first image is encoded to obtain the first image feature, and the first image feature is directly transmitted instead of the image itself, which can reduce the data transmission amount, thereby reducing the bandwidth pressure, improving the data transmission efficiency, and ensuring the data processing efficiency. In addition, the object recognition is performed locally by the RSU without uploading to the cloud, and the data processing efficiency can also be guaranteed. Therefore, this embodiment can realize object recognition efficiently and accurately, and improve the efficiency of object tracking.

[0078] The previous RSU adjacent to the first RSU also performs the operation performed by the first RSU and transmits the image feature to the first RSU to achieve continuous object tracking. Therefore, in some embodiments, a third image feature of a third object sent by a third RSU is received;

[0079] Based on the first image feature and the third image feature, it is determined whether the first object and the third object are the same object.

[0080] The third RSU is the previous RSU adjacent to the first RSU.

[0081] In some embodiments, determining whether the first object is the same as the third object based on the first image feature and the third image feature comprises:

[0082] Calculating feature similarity between the first image feature and the third image feature;

[0083] Based on the feature similarity, it is determined whether the first object is identical to the third object.

[0084] Among them, since image features are usually represented by vectors, calculating the feature similarity between the first image feature and the third image feature can be achieved by calculating the vector distance between the first image feature and the third image feature, such as Euclidean distance, cosine distance, etc.

[0085] Based on the feature similarity, if the feature similarity is greater than a first threshold, it can be determined that the first object is the same as the second object, otherwise it is considered that the first object is different from the third object.

[0086] In order to further improve the accuracy of object recognition, in some embodiments, the method may further include:

[0087] identifying a first distance and / or a first angle between the first object and a capture unit corresponding to the first image;

[0088] Receiving a third distance and / or a third angle between a third object and its corresponding collection unit sent by the third RSU;

[0089] Calculating a distance similarity between the first distance and the third distance, and / or calculating an angle similarity between the first angle and the third angle;

[0090] The determining, based on the feature similarity, whether the first object is identical to the third object comprises:

[0091] Based on the distance similarity and / or the angle similarity and the feature similarity, it is determined whether the first object is identical to the third object.

[0092] Wherein, based on the distance similarity and / or the angle similarity, and the feature similarity, determining whether the first object is the same as the third object includes:

[0093] Based on the distance similarity and the feature similarity, determining whether the first object is the same as the third object;

[0094] Based on the angle similarity and the feature similarity, determining whether the first object is identical to the third object;

[0095] or,

[0096] Based on the distance similarity, the angle similarity, and the feature similarity, it is determined whether the first object is identical to the third object.

[0097] The distance similarity can be determined by the distance range where the distance difference is located and the similarity that is preset. The angle similarity can also be determined by the angle range where the angle difference is located and the similarity that is preset.

[0098] Since one acquisition unit may cover multiple RSUs, the images obtained by adjacent RSUs may not be very different. Therefore, the distance or angle between the object and the acquisition unit can be added to participate in object recognition.

[0099] Wherein, a weighted average or weighted sum value between different types of similarities may be calculated as the object similarity, and based on the object similarity, determining whether the first object is the same as the third object, for example, based on the distance similarity, the angle similarity, and the feature similarity, determining whether the first object is the same as the third object may include:

[0100] Calculate the object similarity based on the distance similarity, the angle similarity and the feature similarity; wherein the weighted average value or weighted sum value of the distance similarity, the angle similarity and the feature similarity may be used as the object similarity;

[0101] If the object similarity is greater than a second threshold, it is determined that the first object is the same as the third object; otherwise, it is determined that the first object is different from the third object.

[0102] Among them, the first distance and the first angle of the first object and its corresponding acquisition unit can be obtained through image analysis, or through radar measurement deployed at the location of the acquisition unit, etc. The distance and angle measurement method is the same as the prior art and will not be repeated here.

[0103] In addition, the third RSU may send characteristic data of multiple objects. Therefore, in some embodiments,

[0104] The receiving a third distance and / or a third angle between the third object and the collection unit thereof sent by the third RSU may include:

[0105] Receiving distances and / or angles between a plurality of objects and their collection units respectively sent by the third RSU;

[0106] Searching for a third object whose distance difference from the first distance is less than a preset distance, and / or whose angle difference from the first angle is less than a preset angle;

[0107] A third distance and / or a third angle between the third object and its corresponding acquisition unit is determined.

[0108] In order to further improve the accuracy of object recognition, in some embodiments, the method may further include:

[0109] identifying a first object attribute of the first object;

[0110] receiving a third object attribute of a third object sent by the third RSU;

[0111] Calculating the attribute similarity between the first object attribute and the third object attribute;

[0112] The determining, based on the feature similarity, whether the first object is identical to the third object comprises:

[0113] Based on the attribute similarity and the feature similarity, it is determined whether the first object is the same as the third object.

[0114] The object attribute may include at least one attribute factor, and the at least one attribute factor may include one or more of object height, object width, object length, and the like.

[0115] Therefore, calculating the attribute similarity between the first object attribute and the third object attribute may include:

[0116] Calculating the sub-similarity of the same type of attribute factors in the first object attribute and the third object attribute;

[0117] The object attribute similarity is calculated based on the sub-similarity corresponding to at least one attribute factor, and the weighted average value or weighted sum value of the sub-similarity corresponding to at least one attribute factor may be used as the object attribute similarity.

[0118] Among them, based on the attribute similarity and the feature similarity, determining whether the first object is the same as the third object can be based on the attribute similarity and the feature similarity, calculating the object similarity, and then determining whether the first object is the same as the third object based on the object similarity. For example, if the object similarity is greater than a certain threshold, it can be determined that the first object is the same as the third object, otherwise it is considered that the first object is different from the third object.

[0119] In order to further improve the accuracy of object recognition, in some embodiments, the method may further include:

[0120] Identify a first distance and a first angle between the first object and a capture unit corresponding to the first image;

[0121] receiving a third distance and a third angle between a third object and a corresponding collection unit sent by the third RSU;

[0122] Calculating the distance similarity between the first distance and the third distance, and calculating the angle similarity between the first angle and the third angle;

[0123] identifying a first object attribute of the first object;

[0124] receiving a third object attribute of a third object sent by the third RSU;

[0125] Calculating the attribute similarity between the first object attribute and the third object attribute;

[0126] The determining, based on the feature similarity, whether the first object is identical to the third object comprises:

[0127] Based on the distance similarity, the angle similarity, the attribute similarity, and the feature similarity, it is determined whether the first object is identical to the third object.

[0128] Optionally, a weighted average or weighted sum of the distance similarity, the angle similarity, the attribute similarity and the feature similarity may be calculated as the object similarity; based on the object similarity, it is determined whether the first object and the third object are the same.

[0129] From the above description, it can be seen that angle, distance, object attributes, etc. can be added to participate in object recognition.

[0130] Therefore, in some embodiments, the method may further include:

[0131] identifying a first distance and / or a first angle between the first object and a capture unit of the first image;

[0132] The first distance and / or the first angle are sent to a second RSU, so that the second RSU determines the distance similarity and / or angle similarity between the first object and the second object based on the second distance and / or the first angle between the second object and the acquisition unit of the second image, and determines whether the first object is the same as the second object based on the distance similarity and / or the angle similarity and the feature similarity between the first object and the second object.

[0133] In some embodiments, the method may further include:

[0134] identifying a first object attribute of the first object;

[0135] The first object attribute of the first object is sent to the second RSU, so that the second RSU determines the attribute similarity between the first object and the first object based on the second object attribute of the second object and the first object attribute, and determines whether the first object is the same as the second object based on the attribute similarity and the feature similarity between the first object and the second object.

[0136] In order to facilitate object tracking, an object identifier may be set for identifying the object. Therefore, in some embodiments, the method may further include:

[0137] determining an object identifier of the first object;

[0138] The object identifier of the first object is sent to the second RSU, so that when the second RSU determines that the second object is the same as the first object, the object identifier of the first object is set as the object identifier of the second object.

[0139] The object identifier may be an object number that uniquely identifies the first object, etc.

[0140] In some embodiments, the method may further include:

[0141] receiving an object identifier of the third object sent by the third RSU;

[0142] If the first object is identical to the third object, the object ID of the third object is set to the object ID of the first object.

[0143] In practical applications, the object may specifically refer to a vehicle. Therefore, in some embodiments, the method may further include:

[0144] identifying a license plate number of the first vehicle from the first image;

[0145] If the recognition is successful, the ticket number of the first vehicle is sent to the second RSU so that the second RSU can recognize the license plate number of the second vehicle from the second image. If the recognition is successful, it is determined whether the first vehicle is the same as the second vehicle based on the license plate number of the second vehicle and the license plate number of the second vehicle. If the recognition fails, it is determined whether the first vehicle is the same as the second vehicle based on the first image features and the second image features.

[0146] In some embodiments, the object is a vehicle, and the method further comprises:

[0147] Receiving the license plate number of the third vehicle sent by the third RSU;

[0148] identifying a license plate number of the first vehicle from the first image;

[0149] The determining, based on the first image feature and the third image feature, whether the first object is the same as the third object comprises:

[0150] If the license plate recognition is successful, determining whether the first vehicle is the same as the second vehicle based on the license plate number of the third vehicle and the license plate number of the first license plate;

[0151] If the license plate recognition fails, it is determined whether the first vehicle is the same as the third vehicle based on the first image feature and the third image feature.

[0152] That is, if the license plate number can be identified, the license plate number can be used as a vehicle identifier and vehicle tracking can be achieved.

[0153] However, since license plate number recognition has high requirements on images and is a fine recognition, it may not be successfully recognized. Therefore, if the license plate number cannot be successfully recognized, vehicle recognition is performed based on the first image feature and the second image feature.

[0154] Among them, identifying the license plate number of the first vehicle from the first image may be:

[0155] The license plate number of the first vehicle is recognized from the first image using a license plate recognition model.

[0156] The vehicle recognition model can be obtained by training with vehicle sample images labeled with license plate numbers, with the vehicle sample images serving as model inputs and the license plate numbers serving as model labels, and model training being performed.

[0157] The vehicle identification model is a machine learning model, such as a neural network model.

[0158] It can be seen from the above description that, as an optional manner, extracting the first image feature corresponding to the first object from the first image may include:

[0159] detecting an object position of a first object in the first image;

[0160] extracting a first image region where an object position of the first object is located from the first image;

[0161] A first image feature of the first image region is extracted.

[0162] In some embodiments, detecting an object position of a first object in the first image comprises:

[0163] detecting an object position of a first object in the first image using an object detection model;

[0164] The object detection model is obtained by training based on sample images with the locations of the objects marked. The sample images can be used as model inputs, and the locations of the objects are used as training labels to perform model training.

[0165] The object detection model can be implemented using a neural network model, such as RNN, CNN, etc.

[0166] In some embodiments, extracting the first image feature of the first image area may include:

[0167] A first image feature is extracted from the first image region using a feature encoding model.

[0168] In order to improve the accuracy of image features so that the image features can accurately express the object, the feature encoding model can be obtained based on triple sample image training; the triple sample image includes an origin (Anchor) sample image, a positive sample image and a negative sample image.

[0169] The feature encoding model can be unsupervised or semi-supervised, an end-to-end model, and can be implemented using a neural network model.

[0170] The origin sample image and the positive sample image are images of the same object, while the negative sample image and the origin sample image are not the same object. During the training process, the similarity between the origin sample image and the same type of sample image is less than the similarity between the origin image and the different type of sample image as a constraint for training.

[0171] The use of triplet sample images makes the feature encoding model obtained through training more accurate, thereby improving the accuracy of feature extraction and further improving the accuracy of object recognition.

[0172] In some embodiments, the feature encoding model can be pre-trained in the following manner:

[0173] Get triplet sample image;

[0174] Determining image features of the triplet sample image;

[0175] Based on the image features of the triplet sample images, a feature encoding model is trained using a triplet loss function.

[0176] Among them, the triplet loss function makes images of the same object closer to each other, while images of different objects are farther away from each other. The triplet loss function minimizes the distance between the origin sample image and the sorted sample image, which have the same object, and maximizes the distance between the origin sample image and the heterogeneous sample image, thereby achieving training.

[0177] When performing model training, the image features of the triplet sample images can be pre-set, and the feature encoding model can be obtained by training the triplet loss function, which can be used to extract image features.

[0178] In order to further reduce the amount of data, image features can be represented by embedding vectors.

[0179] As an alternative, Figure 2As shown in , the feature encoding model can be composed of an input layer 201, a feature extraction layer 202, a normalization layer 203, an embedding layer 204, and an output layer 205. The feature extraction layer 202 can be composed of a convolutional layer, a pooling layer, and / or a fully connected layer, and the input layer and the feature extraction layer can be implemented using a CNN architecture.

[0180] During training, the triple sample image is input through the input layer, the features are extracted by the feature extraction layer, and then normalized by the normalization layer. Then, the embedding vector of the triple sample image is obtained by the embedding layer. During the training process, the embedding vector of the triple sample image is a known value, which is used to train the model parameters. After that, the embedding vector of the triple sample image is constrained by the triple loss function of the output layer to obtain the model parameters of the feature encoding model. After a large number of triple sample image training, the feature encoding model can be obtained.

[0181] Therefore, by inputting the first image region into the feature encoding model, an embedding vector corresponding to the first image region can be obtained, and the embedding vector serves as the first image feature.

[0182] In addition, in order to further simplify the operation, the object detection model and the feature encoding model can be synthesized in a network structure to form a feature extraction model, so that in the calculation process, there is no need to first extract the area where the object is located and then perform feature extraction. The image features can be directly obtained by inputting the image into the feature extraction model. Therefore, in some embodiments, extracting the first image features corresponding to the first object from the first image includes:

[0183] A feature extraction model is used to extract first image features corresponding to the first object from the first image; wherein the feature extraction model is obtained by training based on triple sample images that mark the locations of the objects.

[0184] The above text takes the first RSU as the execution subject, and mainly describes the present application solution from the perspective of feature extraction. The following text takes the second RSU as the execution subject, and describes the present application technical solution from the perspective of object comparison. It can be understood that the object comparison operation performed by the second RSU will also be performed by the first RSU, see the interactive operation with the third RSU described above. That is, each RSU will perform object recognition on the image it obtains to extract image features, and compare it with the image features sent by the previous RSU adjacent to it, and send the extracted image features to the next RSU adjacent to it, so that the next RSU can continue to compare.

[0185] like Figure 3As shown, it is a flow chart of another embodiment of a detection method provided by an embodiment of the present application. The technical solution of this embodiment is executed by a second RSU, and the second RSU is the next RSU adjacent to the first RSU. The method may include the following steps:

[0186] 301: Receive a first image feature of a first object sent by a first RSU.

[0187] The first image feature is extracted from the first image. The specific extraction method can be found in the above description and will not be repeated here.

[0188] 302: Determine the acquired second image.

[0189] Optionally, any video frame acquired by at least one acquisition unit connected to the second RSU may be used as the second image.

[0190] 303: Extracting a second image feature corresponding to a second object from the second image.

[0191] The method for extracting the second image features corresponding to the second object from the second image is the same as the method for extracting the first image features corresponding to the first object from the first image, and will not be described in detail herein.

[0192] The second object may be first identified, and a corresponding extraction strategy may be determined based on the identification result, so that the second image features corresponding to the second object may be extracted from the second image according to the extraction strategy. The extraction strategy may include, for example, the type of features and / or the number of features to be extracted.

[0193] Therefore, optionally, extracting a second image feature corresponding to the second object from the second image may include:

[0194] detecting a second object in the second image;

[0195] Identify the second object to obtain an identification result;

[0196] Determining an extraction strategy corresponding to the recognition result;

[0197] According to the extraction strategy, a second image feature corresponding to the second object is extracted.

[0198] Wherein, extracting the first image feature corresponding to the first object according to the extraction strategy may include:

[0199] Extracting, according to the extraction strategy, image features of at least one feature type corresponding to the first object;

[0200] The image features of the at least one feature type are fused to obtain a first image feature corresponding to the first object.

[0201] Optionally, identifying the first object and obtaining an identification result may include:

[0202] The first object is identified by color, category or brand to obtain an identification result; wherein different colors, different categories or different brands correspond to different extraction strategies.

[0203] 304: Determine whether the first object and the second object are the same object based on the first image feature and the second image feature.

[0204] Optionally, a feature similarity between the first image feature and the second image feature may be calculated;

[0205] Based on the feature similarity, it is determined whether the first object is the same as the second object. For example, if the feature similarity is greater than a first threshold, it is determined that the first object is the same as the second object, otherwise, they are different.

[0206] Among them, the specific way in which the second RSU determines whether the first object and the second object are the same object is the same as the way in which the first RSU determines whether the first object and the third object are the same object. It will not be described in detail here and below. The implementation details can be found in the corresponding embodiments above.

[0207] In some embodiments, the method may further include:

[0208] identifying a second distance and / or a second angle between the second object and a capture unit corresponding to the second image;

[0209] receiving a first distance and / or a second angle between a first object and a corresponding collection unit sent by the first RSU;

[0210] Calculating a distance similarity between the first distance and the second distance, and / or calculating an angle similarity between the first angle and the second angle;

[0211] The determining, based on the feature similarity, whether the first object is identical to the second object comprises:

[0212] Based on the distance similarity and / or the angle similarity and the feature similarity, it is determined whether the first object is identical to the second object.

[0213] In some embodiments, the method may further include:

[0214] identifying a second object attribute of the second object;

[0215] receiving a first object attribute of a first object sent by the first RSU;

[0216] Calculating the attribute similarity between the first object attribute and the second object attribute;

[0217] The determining, based on the feature similarity, whether the first object is identical to the second object comprises:

[0218] Based on the attribute similarity and the feature similarity, it is determined whether the first object is the same as the second object.

[0219] In some embodiments, the method may further include:

[0220] receiving an object identifier of the first object sent by the first RSU;

[0221] If the first object is identical to the second object, the object identifier of the first object is set to the object identifier of the second object.

[0222] In some embodiments, the object is a vehicle, and the method may further include:

[0223] Receiving the license plate number of the first vehicle sent by the first RSU;

[0224] identifying a license plate number of a second vehicle from the second image;

[0225] The determining, based on the first image feature and the third image feature, whether the first object is the same as the second object comprises:

[0226] If the license plate recognition is successful, determining whether the first vehicle is the same as the second vehicle based on the license plate number of the second vehicle and the license plate number of the first vehicle;

[0227] If the license plate recognition fails, it is determined whether the first vehicle is the same as the second vehicle based on the first image feature and the second image feature.

[0228] In addition, the second RSU, like the first RSU, will also continue to transmit the second image feature of the second object acquired by it to the next RSU adjacent to it, thereby realizing continuous tracking of the same object between different RSUs.

[0229] In a practical application, the object described in the embodiment of the present application may refer to a vehicle. In an intelligent transportation scenario, it is mainly used to track the vehicle to implement corresponding processing or provide corresponding services for the vehicle.

[0230] For ease of understanding, Figure 4 The diagram shows the deployment of RSUs on one side of a road section in an actual application, and three RSUs are used as an example. To keep the description consistent with the above, the three RSUs can be represented as a first RSU 401, a second RSU 402, and a third RSU 403. The third RSU 403, the first RSU 401, and the second RSU 402 are arranged in a row on one side of the road section 400. The vehicle travels in the direction from the first RSU to the direction where the second RSU is located.

[0231] Each RSU may be connected to at least one acquisition unit 404. One acquisition unit 404 may also be connected to at least one RSU. Therefore, when the vehicle is moving at a high speed, the image contents of images obtained by two adjacent RSUs may have a small difference or even be the same.

[0232] When the vehicle 405 travels on the road section 400, it will sequentially pass the third RSU 403, the first RSU 401, and the second RSU 402. Since each RSU performs the same operation and interacts with the neighboring RSUs in the same way, the following mainly takes the first RSU 401 and the second RSU 402 as an example for explanation:

[0233] The first RSU may use any video frame acquired by at least one acquisition unit connected thereto as the first image, and of course may also use a video frame acquired from at least one acquisition unit at regular intervals as the first image.

[0234] The first RSU can perform vehicle recognition based on the first image, and any vehicle obtained by recognition can be used as the first vehicle, so that the first image feature corresponding to the first vehicle can be extracted from the first image.

[0235] The first RSU will send the vehicle features of each vehicle to the second RSU, taking the first image features of the first vehicle as an example;

[0236] Similarly, the second RSU will also obtain a video frame from at least one connected acquisition unit as a second image, and vehicle identification can be performed based on the second image. Any identified vehicle can be used as a second vehicle, so that the second image features corresponding to the second vehicle can be extracted from the second image.

[0237] The specific method of image feature extraction can be found in the above description and will not be repeated here.

[0238] The second RSU can determine whether the first vehicle and the second ticket are the same vehicle based on the first image feature and the second image feature, thereby realizing vehicle tracking.

[0239] Since what is transmitted between RSUs is image features rather than images, the amount of data transmitted is reduced, bandwidth usage is reduced, and object recognition efficiency is improved. In addition, local recognition is performed by RSU without uploading to the cloud, which reduces the processing pressure on the cloud and ensures object recognition efficiency. Among them, image features can be extracted using a neural network model trained using triple sample images, which can improve the accuracy of image features and the accuracy of object recognition, thereby achieving efficient and accurate vehicle recognition.

[0240] Figure 5 A schematic diagram of a structure of an embodiment of a detection device provided in an embodiment of the present application, the device may include:

[0241] A first image acquisition module 501 is used to determine a first image acquired;

[0242] A first feature extraction module 502, configured to extract a first image feature corresponding to a first object from the first image;

[0243] A first feature sending module 503, configured to send the first image feature to a second RSU;

[0244] The first image feature is combined with a second image feature of a second object extracted from a second image by the second RSU to determine whether the first object and the second object are the same object.

[0245] In some embodiments, the first image acquisition module is specifically configured to use any video frame acquired by at least one acquisition unit connected to the first RSU as the first image.

[0246] In some embodiments, the apparatus may further include:

[0247] A second feature receiving module, configured to receive a third image feature of a third object sent by a third RSU;

[0248] Based on the first image feature and the third image feature, it is determined whether the first object and the third object are the same object.

[0249] The second object recognition module is used to determine whether the first object and the third object are the same object based on the first image feature and the third image feature.

[0250] In some embodiments, the second object recognition module is specifically configured to calculate a feature similarity between the first image feature and the third image feature; and based on the feature similarity, determine whether the first object is the same as the third object.

[0251] In some embodiments, the apparatus may further include:

[0252] A first parameter identification module, used to identify a first distance and / or a first angle between the first object and a collection unit corresponding to the first image;

[0253] A first parameter receiving module, configured to receive a third distance and / or a third angle between a third object and its corresponding collection unit sent by the third RSU;

[0254] a first calculation module, configured to calculate a distance similarity between the first distance and the third distance, and / or calculate an angle similarity between the first angle and the third angle;

[0255] The second object recognition module determines whether the first object is identical to the third object based on the feature similarity. Specifically, the second object recognition module determines whether the first object is identical to the third object based on the distance similarity and / or the angle similarity and the feature similarity.

[0256] In some embodiments, the first parameter receiving module is specifically used to receive the distances and / or angles between multiple objects sent by the third RSU and their collection units respectively; search for a third object whose distance difference with the first distance is less than a preset distance, and / or whose angle difference with the first angle is less than a preset angle; and determine a third distance and / or a third angle between the third object and its corresponding collection unit.

[0257] In some embodiments, the apparatus may further include:

[0258] A first attribute recognition module, used to recognize a first object attribute of the first object;

[0259] A first attribute receiving module, configured to receive a third object attribute of a third object sent by the third RSU;

[0260] A second calculation module, used for calculating the attribute similarity between the first object attribute and the third object attribute;

[0261] The second object recognition module determines whether the first object is identical to the third object based on the feature similarity, and may specifically determine whether the first object is identical to the third object based on the attribute similarity and the feature similarity.

[0262] In some embodiments, the apparatus may further include:

[0263] A first parameter identification module, used to identify a first distance and / or a first angle between the first object and a collection unit corresponding to the first image;

[0264] a first parameter sending module, configured to send the first distance and / or the first angle to a second RSU, so that the second RSU determines the distance similarity and / or angle similarity between the first object and the second object based on the second distance and / or the first angle between the second object and the acquisition unit of the second image, and determines whether the first object is the same as the second object based on the distance similarity and / or the angle similarity and the feature similarity between the first object and the second object.

[0265] In some embodiments, the apparatus may further include:

[0266] A first attribute recognition module, used to recognize a first object attribute of the first object;

[0267] The second attribute sending module is used to send the first object attribute of the first object to the second RSU, so that the second RSU determines the attribute similarity between the first object and the first object based on the second object attribute of the second object and the first object attribute, and determines whether the first object is the same as the second object based on the attribute similarity and the feature similarity between the first object and the second object.

[0268] In some embodiments, the apparatus may further include:

[0269] an identification determination module, configured to determine an object identification of the first object;

[0270] The identification sending module sends the object identification of the first object to the second RSU, so that when the second RSU determines that the second object is the same as the first object, the object identification of the first object is set as the object identification of the second object.

[0271] In some embodiments, the object is a vehicle, and the apparatus further comprises:

[0272] a license plate recognition module, configured to recognize the license plate number of the first vehicle from the first image;

[0273] A license plate sending module is used to send the ticket number of the first vehicle to the second RSU if the recognition is successful, so that the second RSU can recognize the license plate number of the second vehicle from the second image; if the recognition is successful, determine whether the first vehicle is the same as the second vehicle based on the license plate number of the second vehicle and the license plate number of the second ticket; if the recognition fails, determine whether the first vehicle is the same as the second vehicle based on the first image features and the second image features.

[0274] In some embodiments, the apparatus may further include:

[0275] an identification receiving module, configured to receive an object identification of the third object sent by the third RSU;

[0276] An identification setting module is used to set the object identification of the third object as the object identification of the first object if the first object is the same as the third object.

[0277] In some embodiments, the object is a vehicle, and the apparatus may further include:

[0278] A license plate receiving module, used for receiving the license plate number of the third vehicle sent by the third RSU;

[0279] a license plate recognition module, configured to recognize a license plate number of a first vehicle from the first image;

[0280] The second object recognition module is specifically used to determine whether the first vehicle is the same as the third vehicle based on the license plate number of the third vehicle and the license plate number of the first vehicle if license plate recognition is successful; and to determine whether the first vehicle is the same as the third vehicle based on the first image feature and the third image feature if license plate recognition fails.

[0281] In some embodiments, the first feature extraction module is specifically used to detect the object position of the first object in the first image; extract the first image area where the object position of the first object is located from the first image; and extract the first image feature of the first image area.

[0282] In some embodiments, the first feature extraction module detects the object position of the first object in the first image, including: using an object detection model to detect the object position of the first object in the first image; wherein the object detection model is trained based on sample images that mark the location of the object.

[0283] In some embodiments, the first feature extraction module extracts the first image feature of the first image area, including: using a feature encoding model to extract the first image feature from the first image area; wherein the feature encoding model is obtained based on triplet sample image training; the triplet sample image includes an origin sample image, a similar sample image, and a heterogeneous sample image.

[0284] In some embodiments, the apparatus may further include:

[0285] The model training module is used to obtain a triplet sample image; determine the image features of the triplet sample image; and train a feature encoding model using a triplet loss function based on the image features of the triplet sample image.

[0286] In some embodiments, the first feature extraction module is specifically used to extract first image features corresponding to the first object from the first image using a feature extraction model; wherein the feature extraction model is trained based on triplet sample images of labeled object locations.

[0287] In some embodiments, the first feature extraction module can be specifically used to detect a first object in the first image; identify the first object to obtain an identification result; determine an extraction strategy corresponding to the identification result; and extract a first image feature corresponding to the first object according to the extraction strategy.

[0288] In some embodiments, the first feature extraction module extracts the first image feature corresponding to the first object according to the extraction strategy, including: extracting image features of at least one feature type corresponding to the first object according to the extraction strategy; and fusing the image features of the at least one feature type to obtain the first image feature corresponding to the first object.

[0289] In some embodiments, the first feature extraction module identifies the first object to obtain an identification result, including: identifying the first object by color, category or brand to obtain an identification result; wherein different colors, different categories or different brands correspond to different extraction strategies.

[0290] Figure 5 The detection device can perform Figure 1 The implementation principle and technical effect of the detection method described in the embodiment are not described in detail. The specific way in which each module and unit performs operations in the detection device in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0291] In one possible design, Figure 5 The detection device of the embodiment shown can be implemented as a roadside unit, such as Figure 6 As shown, the roadside unit may include a storage component 601 and a processing component 602;

[0292] The storage component 601 stores one or more computer instructions. When the one or more computer instructions are called and executed by the processing component, the processing component 602 executes Figure 1 The detection method of the illustrated embodiment.

[0293] The processing component 602 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0294] The storage component 601 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0295] Of course, a computing device may also include other components, such as input / output interfaces, communication components, etc.

[0296] The input / output interface provides an interface between the processing component and the peripheral interface module. The above-mentioned peripheral interface module can be an output device, an input device, an acquisition component, etc.

[0297] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0298] The present application also provides a computer-readable storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The detection method of the illustrated embodiment.

[0299] Figure 7 A schematic diagram of the structure of another embodiment of a detection device provided in an embodiment of the present application, the device may include:

[0300] A first feature receiving module 701 is configured to receive a first image feature of a first object sent by a first RSU; wherein the first image feature is extracted from a first image;

[0301] A second image acquisition module 702, used for image determination and acquisition of a second image;

[0302] A second feature extraction module 703, configured to extract first image features corresponding to a second object from the second image;

[0303] The first object recognition module 704 is configured to determine whether the first object and the second object are the same object based on the first image feature and the second image feature.

[0304] Figure 7 The detection device can perform Figure 3 The implementation principle and technical effect of the detection method described in the embodiment are not described in detail. The specific way in which each module and unit performs operations in the detection device in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0305] In one possible design, Figure 7 The detection device of the embodiment shown can be implemented as a roadside unit, such as Figure 8 As shown, the roadside unit may include a storage component 801 and a processing component 802;

[0306] The storage component 801 stores one or more computer instructions. When the one or more computer instructions are called and executed by the processing component, the processing component 802 executes Figure 1 The detection method of the illustrated embodiment.

[0307] The processing component 802 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0308] The storage component 801 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0309] Of course, a computing device may also include other components, such as input / output interfaces, communication components, etc.

[0310] The input / output interface provides an interface between the processing component and the peripheral interface module. The above-mentioned peripheral interface module can be an output device, an input device, an acquisition component, etc.

[0311] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0312] The present application also provides a computer-readable storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 3 The detection method of the illustrated embodiment.

[0313] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0314] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0315] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0316] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A detection method, characterized in that: Applied in the first roadside unit RSU, including: Determine a first image acquired; extracting a first image feature corresponding to a first object from the first image; Sending the first image feature to a second roadside unit RSU; The first image feature is combined with the second image feature of the second object extracted from the second image by the second roadside unit RSU to determine whether the first object and the second object are the same object, including: identifying a first distance and / or a first angle between the first object and the acquisition unit of the first image; sending the first distance and / or the first angle to the second roadside unit RSU, so that the second roadside unit RSU determines the first distance similarity and / or the first angle similarity between the first object and the second object based on the second distance and / or the second angle between the second object and the acquisition unit of the second image, and determines whether the first object and the second object are the same based on the first distance similarity and / or the first angle similarity and the first feature similarity between the first object and the second object, the first distance similarity being the similarity between the first distance and the second distance, the first angle similarity being the similarity between the first angle and the second angle, and the first feature similarity being the similarity between the first image feature corresponding to the first object and the second image feature corresponding to the second object.

2. The method according to claim 1, characterized in that The determining of the first image acquired by acquisition comprises: Any video frame acquired by at least one acquisition unit connected to the first roadside unit RSU is used as the first image.

3. The method according to claim 1, characterized in that Also includes: receiving a third image feature of a third object sent by a third roadside unit RSU; Based on the first image feature and the third image feature, it is determined whether the first object and the third object are the same object.

4. The method according to claim 3, characterized in that The determining, based on the first image feature and the third image feature, whether the first object and the third object are the same object comprises: Calculating a second feature similarity between the first image feature and the third image feature; Based on the second feature similarity, it is determined whether the first object is identical to the third object.

5. The method according to claim 4, characterized in that Also includes: identifying a first distance and / or a first angle between the first object and a capture unit corresponding to the first image; Receiving a third distance and / or a third angle between a third object and its corresponding collection unit sent by the third roadside unit RSU; Calculating a second distance similarity between the first distance and the third distance, and / or calculating a second angle similarity between the first angle and the third angle; The determining, based on the second feature similarity, whether the first object is identical to the third object comprises: Based on the second distance similarity and / or the second angle similarity, and the second feature similarity, it is determined whether the first object is identical to the third object.

6. The method according to claim 5, characterized in that The receiving a third distance and / or a third angle between a third object and a collection unit thereof sent by the third roadside unit RSU comprises: Receiving distances and / or angles between a plurality of objects and their collection units respectively sent by the third roadside unit RSU; Searching for a third object whose distance difference from the first distance is less than a preset distance, and / or whose angle difference from the first angle is less than a preset angle; A third distance and / or a third angle between the third object and its corresponding acquisition unit is determined.

7. The method according to claim 4, characterized in that Also includes: identifying a first object attribute of the first object; Receiving a third object attribute of a third object sent by the third roadside unit RSU; Calculating the attribute similarity between the first object attribute and the third object attribute; The determining, based on the second feature similarity, whether the first object is identical to the third object comprises: Based on the attribute similarity and the feature similarity, it is determined whether the first object is the same as the third object.

8. The method according to claim 1, characterized in that: Also includes: identifying a first object attribute of the first object; The first object attribute of the first object is sent to a second roadside unit RSU, so that the second roadside unit RSU determines the first attribute similarity between the first object and the second object based on the second object attribute of the second object and the first object attribute, and determines whether the first object is the same as the second object based on the first attribute similarity and the first feature similarity between the first object and the second object.

9. The method according to claim 1, characterized in that: Also includes: determining a first object identifier of the first object; The first object identifier of the first object is sent to the second roadside unit RSU, so that when the second roadside unit RSU determines that the second object is the same as the first object, the first object identifier of the first object is set as the second object identifier of the second object.

10. The method according to claim 9, characterized in that The object is a vehicle, and the method further comprises: identifying a first license plate number of a first vehicle from the first image; If the recognition is successful, the first license plate number of the first vehicle is sent to the second roadside unit RSU so that the second roadside unit RSU can recognize the second license plate number of the second vehicle from the second image. If the recognition is successful, based on the first license plate number of the first vehicle and the second license plate number of the second vehicle, it is determined whether the first vehicle is the same as the second vehicle; if the recognition fails, based on the first image features and the second image features, it is determined whether the first vehicle is the same as the second vehicle.

11. The method according to claim 3, characterized in that Also includes: receiving a third object identifier of the third object sent by the third roadside unit RSU; If the first object is identical to the third object, the third object identifier of the third object is set as the first object identifier of the first object.

12. The method according to claim 11, characterized in that The object is a vehicle, and the method further comprises: receiving a third license plate number of a third vehicle sent by a third roadside unit RSU; identifying a first license plate number of a first vehicle from the first image; The determining, based on the first image feature and the third image feature, whether the first object is the same as the third object comprises: If the license plate recognition is successful, determining whether the first vehicle is the same as the third vehicle based on the third license plate number of the third vehicle and the first license plate number of the first vehicle; If the license plate recognition fails, it is determined whether the first vehicle is the same as the third vehicle based on the first image feature and the third image feature.

13. The method according to claim 1, characterized in that Extracting a first image feature corresponding to the first object from the first image includes: Detecting an object position of a first object in the first image; extracting a first image region where the object position of the first object is located from the first image; A first image feature of the first image region is extracted.

14. The method according to claim 13, characterized in that Detecting the object position of the first object in the first image comprises: An object detection model is used to detect an object position of a first object in the first image; wherein the object detection model is trained based on sample images in which the positions of the objects are marked.

15. The method according to claim 13, characterized in that The extracting the first image feature of the first image area comprises: A first image feature is extracted from the first image region using a feature coding model; wherein the feature coding model is obtained based on triplet sample image training; the triplet sample image includes an origin sample image, a similar sample image, and a heterogeneous sample image.

16. The method according to claim 15, characterized in that The feature encoding model is pre-trained in the following manner: Get triplet sample image; Determining image features of the triplet sample image; Based on the image features of the triplet sample images, a feature encoding model is trained using a triplet loss function.

17. The method according to claim 1, characterized in that Extracting a first image feature corresponding to the first object from the first image comprises: A feature extraction model is used to extract first image features corresponding to the first object from the first image; wherein the feature extraction model is obtained by training based on triple sample images that mark the locations of the objects.

18. The method according to claim 1, characterized in that Extracting a first image feature corresponding to the first object from the first image comprises: detecting a first object in the first image; Identify the first object to obtain an identification result; Determining an extraction strategy corresponding to the recognition result; According to the extraction strategy, a first image feature corresponding to the first object is extracted.

19. The method according to claim 18, characterized in that The extracting, according to the extraction strategy, a first image feature corresponding to the first object includes: Extracting, according to the extraction strategy, image features of at least one feature type corresponding to the first object; The image features of the at least one feature type are fused to obtain a first image feature corresponding to the first object.

20. The method according to claim 18, characterized in that The identifying the first object and obtaining an identification result includes: The first object is identified by color, category or brand to obtain an identification result; wherein different colors, different categories or different brands correspond to different extraction strategies.

21. A detection method, characterized in that: Applied to a second roadside unit RSU, the method comprises: Receiving a first image feature of a first object sent by a first roadside unit RSU; wherein the first image feature is extracted from the first image; Determine a second image acquired by acquisition; extract first image features corresponding to the second object from the second image; Based on the first image feature and the second image feature, determining whether the first object and the second object are the same object includes: after receiving the first distance and / or the first angle sent by the first roadside unit RSU, determining the first distance similarity and / or the first angle similarity between the first object and the second object based on the second distance and / or the second angle between the second object and the acquisition unit of the second image, and determining whether the first object and the second object are the same based on the first distance similarity and / or the first angle similarity and the first feature similarity between the first object and the second object, the first distance similarity is the similarity between the first distance and the second distance, the first angle similarity is the similarity between the first angle and the second angle, and the first feature similarity is the similarity between the first image feature corresponding to the first object and the second image feature corresponding to the second object.

22. A detection device, characterized in that: include: A first image acquisition module, used to determine a first image acquired; A first feature extraction module, used to extract a first image feature corresponding to a first object from the first image; A first feature sending module, used for sending the first image feature to a second roadside unit RSU; The first image feature is combined with the second image feature of the second object extracted from the second image by the second roadside unit RSU to determine whether the first object and the second object are the same object, including: identifying a first distance and / or a first angle between the first object and the acquisition unit of the first image; sending the first distance and / or the first angle to the second roadside unit RSU, so that the second roadside unit RSU determines the first distance similarity and / or the first angle similarity between the first object and the second object based on the second distance and / or the second angle between the second object and the acquisition unit of the second image, and determines whether the first object and the second object are the same based on the first distance similarity and / or the first angle similarity and the first feature similarity between the first object and the second object, the first distance similarity being the similarity between the first distance and the second distance, the first angle similarity being the similarity between the first angle and the second angle, and the first feature similarity being the similarity between the first image feature corresponding to the first object and the second image feature corresponding to the second object.

23. A detection device, characterized in that: include: A first feature receiving module is used to receive a first image feature of a first object sent by a first roadside unit RSU; wherein the first image feature is extracted from the first image; A second image acquisition module, used for image determination and acquisition of a second image; A second feature extraction module, used to extract first image features corresponding to a second object from the second image; A first object recognition module, used to determine whether the first object and the second object are the same object based on the first image feature and the second image feature, including: after receiving the first distance and / or the first angle sent by the first roadside unit RSU, determining the first distance similarity and / or the first angle similarity between the first object and the second object based on the second distance and / or the second angle between the second object and the acquisition unit of the second image, and determining whether the first object and the second object are the same based on the first distance similarity and / or the first angle similarity and the first feature similarity between the first object and the second object, the first distance similarity is the similarity between the first distance and the second distance, the first angle similarity is the similarity between the first angle and the second angle, and the first feature similarity is the similarity between the first image feature corresponding to the first object and the second image feature corresponding to the second object.

24. A roadside unit, characterized in that: including storage components and processing components; Wherein, the storage component stores one or more computer instructions; when the one or more computer instructions are called and executed by the processing component, the processing component executes the detection method as described in any one of claims 1 to 20.

25. A roadside unit, characterized in that: including storage components and processing components; Wherein, the storage component stores one or more computer instructions; when the one or more computer instructions are called and executed by the processing component, the processing component executes the detection method as claimed in claim 21.

Images