CN116148806A - Method and device for determining depth of target object and electronic equipment - Google Patents

Abstract

A method, device, and electronic device for determining the depth of a target object, the method comprising: obtaining object information corresponding to the target object, converting the object information into a multidimensional feature vector based on a data preprocessing model, and inputting the multidimensional feature vector into a preset depth model , to obtain the length information corresponding to the target object. Through the above method, the object information of the target object is input into the data preprocessing model to obtain the multi-dimensional feature vector corresponding to the object information, and the multi-dimensional feature vector is input into the trained preset depth model to determine the depth information of the target object. Since both the data preprocessing model and the preset depth model are pre-trained models, the accuracy of the depth information of the target object determined based on the data preprocessing model and the preset depth model can be ensured.

Description

A method, device and electronic equipment for determining the depth of a target object

technical field

The present application relates to the technical field of intelligent driving, and in particular to a method, device and electronic equipment for determining the depth of a target object.

Background technique

In the development process of autonomous driving technology, the software architecture of autonomous driving can be divided into perception module and decision-making module. The perception module obtains the data outside the vehicle through the sensor in the vehicle system. The module issues instructions based on the data acquired by the sensing module.

When the sensor in the perception module described above is a laser radar, the laser radar can determine the position of the target object in three-dimensional space based on the emission time, return time, emission angle, and return angle of the pulse signal sent, and the laser radar The collected data is regarded as point cloud data, therefore, the processing of the point cloud data collected by lidar is as follows:

When detecting a target object based on the lidar, the lidar will obtain all the first point cloud data in the scene containing the target object, and process the first point cloud data based on the PointNet algorithm to obtain the second point cloud corresponding to the target object Data, the point cloud data is a point cloud collection of scenes or objects composed of multiple points, each point can be represented by the coordinates in the space Cartesian coordinate system, and the minimum value of x needs to be determined from all points x of the target object and the maximum value to obtain the range in the x direction; determine the minimum and maximum values of y from the y of all points of the target object to obtain the range in the y direction; determine the value of z from all points of the target object in z The minimum and maximum values get the range in the z direction.

Based on the 6 values obtained above, the 3D frame of the target object can be determined. Based on the 3D frame, the height information, width information and depth information of the target object can be obtained. Since the laser radar cannot estimate the depth characteristics of the object well, so , the depth information of the target object obtained based on the 3D frame is inaccurate.

Contents of the invention

The present application provides a method, device and electronic equipment for determining the depth of a target object, which are used to improve the accuracy of the depth information of the target object.

In a first aspect, the present application provides a method for determining the depth of a target object, the method comprising:

Obtain object information corresponding to the target object, wherein the object information includes: height information, width information, and target type of the target object;

converting the object information into a multidimensional target feature vector based on a data preprocessing model;

Inputting the multi-dimensional target feature vector into a preset depth model to obtain depth information corresponding to the target object.

Through the above method, the object information of the target object is processed through the data preprocessing model to obtain the multi-dimensional target feature vector corresponding to the target object, and then the multi-dimensional target feature vector is processed through the preset depth model to obtain the depth information of the target object , since the data preprocessing model and the preset depth model are pre-trained models, the accuracy of the obtained depth information of the target object can be ensured.

In a possible design, before obtaining the object information corresponding to the target object, it includes:

Obtaining test object information corresponding to each of the m test objects, wherein the test object information is height information, width information, depth information, and type information of the test object, and m is a positive integer;

performing training based on the m test objects, and obtaining a training model corresponding to the m test objects;

In response to the training model meeting the preset condition, the preset depth models corresponding to the m test objects are determined.

Through the above method, m test objects are trained, and the training models of m test objects are obtained. When the training models meet the preset conditions, the preset depth model is determined, and the training model obtained in the training process is carried out. Screening, improves the accuracy of the preset depth model to predict depth information.

In a possible design, training is performed based on the m test objects, and a training model corresponding to the m test objects is obtained, including:

determining the maximum value of all height information, width information, and depth information corresponding to the m test objects, and category feature vectors corresponding to the m test objects;

Performing normalization processing on the height information and width information corresponding to each test object based on the maximum value, to obtain the two-dimensional feature vectors corresponding to each test object;

Splicing the respective two-dimensional feature vectors of each of the test objects and their corresponding category feature vectors to obtain the respective multi-dimensional feature vectors corresponding to each of the test objects;

Training is performed based on the respective depth information corresponding to each of the test objects and the respective multi-dimensional feature vectors to obtain training models corresponding to the m test objects.

Through the above method, the multi-dimensional feature vector of the target object is obtained by splicing the two-dimensional feature vector and the category feature vector of the target object, and model training is performed based on the multi-dimensional feature vector and depth information of each test object to ensure the accuracy of the training model .

In a possible design, responding to the training model meeting preset conditions, including:

Obtaining the total number of training times corresponding to the m test objects, and determining the loss value of the training model corresponding to each training of the m test objects, wherein the loss value represents the predicted depth value of the model and the actual depth value of the training sample difference between;

When the total number of training times reaches a preset number of training times, responding to the training model meeting a preset condition; and

When the loss value is less than a preset loss threshold, it is responded that the training model meets a preset condition.

Through the above method, the training model after each training of the m test objects is tested to ensure that the obtained training model meets the preset conditions, thereby improving the accuracy of the training model.

In a possible design, the object information corresponding to the target object is obtained, including:

Determine the three-dimensional detection frame of the target object, and obtain the target type of the target object based on a preset classification algorithm;

In response to the target type being in a preset category set, read out the height information and width information of the target object based on the three-dimensional detection frame, wherein the preset category set includes a plurality of categories corresponding to each preset object ;

The target type, the height information, and the width information are used as object information corresponding to the target object.

Through the above method, the width information and height information of the target object are obtained through the three-dimensional detection frame, and the target type of the target object is determined to be consistent with the category of the preset object in the preset category set, thereby ensuring the obtained object information of the target object accuracy.

In a possible design, the target type of the target object is obtained based on a preset classification algorithm, including:

extracting a plurality of target features corresponding to the target object;

Matching the multiple target features with a preset category feature set to determine multiple similarity values corresponding to the multiple target feature sets, wherein the preset category feature set includes each preset object corresponding to feature set;

The preset category of the preset object corresponding to the maximum similarity is determined from the multiple similarity values, and the preset category is used as the target type of the target object.

Through the above method, the preset category of the preset object corresponding to the maximum similarity is determined, and the preset category is used as the target type of the target object, thereby improving the accuracy of determining the target type of the target object.

In a possible design, the object information is converted into a multidimensional target feature vector based on a data preprocessing model, including:

input the object information corresponding to the target object into the data preprocessing model, and obtain a target two-dimensional feature vector and a target category vector of the target object;

Combining the target two-dimensional feature vector and the target category vector to obtain a multi-dimensional target feature vector corresponding to the target object.

Through the above method, the object information of the target object is converted into a multi-dimensional object feature vector through the data preprocessing model, and the object information of the target object is preprocessed, which is conducive to improving the accuracy of determining the depth information of the target object.

In a second aspect, the present application provides a device for determining the depth of a target object, the device comprising:

An obtaining module, configured to obtain object information corresponding to the target object, wherein the object information includes: height information, width information, and target type of the target object;

A conversion module, configured to convert the object information into a multidimensional target feature vector based on a data preprocessing model;

A depth module, configured to input the multi-dimensional target feature vector into a preset depth model to obtain depth information corresponding to the target object.

In a possible design, the obtaining module is specifically configured to obtain test object information corresponding to each of the m test objects, perform training based on the m test objects, and obtain a training model corresponding to the m test objects, In response to the training model meeting the preset condition, the preset depth models corresponding to the m test objects are determined.

In a possible design, the obtaining module is further configured to determine the maximum value of all height information, width information, and depth information corresponding to the m test objects, and the maximum values of the m test objects respectively. Category feature vectors, based on the maximum value, the height information and width information corresponding to each test object are normalized to obtain the respective two-dimensional feature vectors corresponding to each test object, and the respective two-dimensional feature vectors of each test object are obtained. Dimensional feature vectors and their corresponding category feature vectors are spliced to obtain multi-dimensional feature vectors corresponding to the respective test objects, and training is performed based on the depth information corresponding to each of the test objects and the respective multi-dimensional feature vectors to obtain the described The training model corresponding to m test objects.

In a possible design, the obtaining module is also used to obtain the total number of training times corresponding to the m test objects, and determine the loss value of the training model corresponding to each training of the m test objects, when When the total number of training times reaches a preset number of training times, the response is that the training model meets a preset condition, and when the loss value is smaller than a preset loss threshold, it is response that the training model meets a preset condition.

In a possible design, the obtaining module is further configured to determine the three-dimensional detection frame of the target object, obtain the target type of the target object based on a preset classification algorithm, and respond to the target type being within a preset In category concentration, the height information and width information of the target object are read out based on the three-dimensional detection frame, and the target type, the height information, and the width information are used as object information corresponding to the target object.

In a possible design, the obtaining module is further configured to extract multiple target features corresponding to the target object, match the multiple target features with a preset category feature set, and determine the multiple Multiple similarity values corresponding to target feature sets, determine the preset category of the preset object corresponding to the maximum similarity from the multiple similarity values, and use the preset category as the target of the target object type.

In a possible design, the conversion module is specifically configured to input the object information corresponding to the target object into the data preprocessing model, and obtain the target two-dimensional feature vector and target category vector of the target object , combining the target two-dimensional feature vector and the target category vector to obtain a multi-dimensional target feature vector corresponding to the target object.

In a third aspect, the present application provides an electronic device, including:

memory for storing computer programs;

The processor is configured to implement the steps of the above-mentioned method for determining the depth of a target object when executing the computer program stored in the memory.

In a fourth aspect, a computer-readable storage medium stores a computer program in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned method for determining the depth of a target object are implemented.

For the various aspects and possible technical effects of the above-mentioned first to fourth aspects, please refer to the above description of the technical effects that can be achieved by the first aspect or various possible solutions in the first aspect, and will not be repeated here.

Description of drawings

FIG. 1 is a flow chart of the steps of a method for determining the depth of a target object provided by the present application;

FIG. 2 is a schematic diagram of processing object information of a target object based on a data preprocessing model provided by the present application;

FIG. 3 is a schematic diagram of obtaining depth information of a target object based on a preset depth model provided by the present application;

FIG. 4 is a schematic structural diagram of a device for determining the depth of a target object provided by the present application;

FIG. 5 is a schematic structural diagram of an electronic device provided by the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the application clearer, the application will be further described in detail below in conjunction with the accompanying drawings. The specific operation methods in the method embodiments can also be applied to the device embodiments or system embodiments. It should be noted that in the description of the present application, "plurality" is understood as "at least two". "And/or" describes the association relationship of associated objects, indicating that there may be three types of relationships, for example, A and/or B may indicate: A exists alone, A and B exist simultaneously, and B exists independently. The connection between A and B can mean: A and B are directly connected and A and B are connected through C. In addition, in the description of the present application, words such as "first" and "second" are only used for the purpose of distinguishing descriptions, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

In the previous technology, the specific method of obtaining the depth information of the target object based on the laser radar is to obtain all the first point cloud data in the scene containing the target object based on the laser radar, and process the first point cloud data based on the PointNet algorithm, Obtain the second point cloud data corresponding to the target object. Each point in the second point cloud data can be represented by the coordinates in the space Cartesian coordinate system. It is necessary to determine the minimum value of x from all points x of the target object and Maximum value, get the range in the x direction; determine the minimum and maximum values of y from all points of the target object in y, get the range in the y direction; determine the minimum value of z from all points of the target object in z value and the maximum value to get the range in the z direction.

Based on the 6 values obtained above, the 3D frame of the target object can be determined. Based on the 3D frame, the height information, width information and depth information of the target object can be obtained. Since the laser radar cannot estimate the depth characteristics of the object well, so , the depth information of the target object obtained based on the 3D frame is inaccurate.

In order to solve the problems described above, the present application provides a method for determining the depth of a target object, so as to accurately determine the depth information of the target object. Wherein, the method and the device described in the embodiment of the present application are based on the same technical concept. Since the principles of the problems solved by the method and the device are similar, the embodiments of the device and the method can be referred to each other, and the repetition will not be repeated.

The embodiments of the present application will be described in detail below in conjunction with the accompanying drawings.

Referring to Figure 1, the present application provides a method for determining the depth of a target object, which can accurately determine the depth information of the target object, and the implementation process of the method is as follows:

Step S1: Obtain object information corresponding to the target object.

The embodiment of the present application is to obtain accurate depth information of the target object. First, it is necessary to detect the target object based on the laser radar to obtain the point cloud data corresponding to the target object. Although the accuracy of the depth information of the target object obtained by the laser radar is low, but , based on the lidar, the width information and height information of the target object can be obtained.

In order to obtain accurate width information and height information of the target object, it is necessary to perform clustering processing on the obtained point cloud data. In the embodiment of the present application, the clustering processing method can be European clustering, and PointNet can be used to classify the target object. Algorithm, after clustering the point cloud data, a 3D frame corresponding to the target object will be generated. After obtaining the 3D frame, the 3D frame needs to be filtered and corrected. Filtering is used to remove the noise data in the 3D frame and other Interference and correction are used to obtain accurate height information and width information based on the point cloud data.

Specifically, the filtering and correction method is usually to detect whether the height and width of the 3D frame are smaller than the preset value. When the height and width are smaller than the preset value, the 3D frame is considered to be caused by noise, and the outlier will be used The detection method is corrected. By performing outlier detection on the point cloud data in the 3D frame, the 3D frame is regenerated after removing the outlier points. Since the outlier point detection is a technology well known to those skilled in the art, so we will not do too much here illustrate.

Based on the processed 3D frame, in order to obtain accurate depth information of the target object, it is also necessary to determine whether the target type of the target object is in the preset category set. When the target type of the target object is in the preset category set, the target object is acquired. The process of obtaining the depth information of the target object; when the target type of the target object is not in the preset category set, the process of obtaining the depth information of the target object will be exited.

Since the embodiment of the present application obtains the depth information of the target object based on the data preprocessing model and the preset depth model, it is necessary to obtain the data preprocessing model and the preset depth model before obtaining the object information of the target object. The specific obtaining process is as follows :

Obtain the test object information corresponding to each of the m test objects, m is a positive integer, the test object information includes: the height information of the test object information, the width information, the depth information, and the type information of the test object, from all the test objects tested Determine the maximum value from the height information, width information, and depth information, and the category feature vectors corresponding to each test object, and then normalize the height information and width information of the test object. The normalization process is as follows:

Divide the height information of each test object by the maximum value, and divide the width information of each test object by the maximum value to obtain the two-dimensional feature vector of each test object, the two-dimensional feature vector includes: the height information of each test object corresponds to The one-dimensional feature vector of each test object, and the one-dimensional feature vector corresponding to the width information of each test object, and then the two-dimensional feature vector and the category feature vector are spliced to obtain the multi-dimensional feature vector corresponding to each test object, so as to obtain the data preprocessing model The process of processing data.

In order to make the depth information obtained based on the preset depth model accurate, it is necessary to perform training based on the corresponding depth information of each test object and its corresponding multi-dimensional feature vector, and input the multi-dimensional feature vector into the fully connected module. The fully connected module in the embodiment of this application The module can include: a fully connected (fully connected layers, FC) layer, a batch normalization (Batch Normalization, BN) layer, and a linear rectification function (Rectified Linear Unit, ReLU) layer, which is used to integrate feature vectors into A value to reduce the impact of the feature position on the classification results. The BN layer is used to prevent overfitting during the test process. The ReLU layer is used to increase the nonlinear relationship between the layers of the neural network, and the multi-dimensional features of each test object The vector input fully connected module is trained multiple times to obtain the test model corresponding to each test object.

Determine the total number of training times for each test object, and determine the loss value of the training model corresponding to each test object during each training. The loss value represents the difference between the model's predicted depth value and the depth value of the training sample. When the total When the number of training times reaches the preset number of training times, it means that the training model obtained from the last training meets the preset conditions; or when the loss value is less than the preset loss value, it means that the training model corresponding to the loss value meets the preset conditions The specific formula of the loss value is as follows:

表示测试物体i的预测深度值，需要进行说明的是，l_i是对测试物体深度进行归一化后的数值。In the above formula, L represents the loss value, N represents the number of training times, l _i represents the actual depth value of the test object i,

represents the predicted depth value of the test object i, and it should be noted that l _i is a value after normalizing the depth of the test object.

When the training model meets the conditions described above, the determined training model is used as the preset depth model.

After the data preprocessing model and the preset depth model are determined, the three-dimensional detection frame of the target object is obtained, and the target type of the target object is obtained based on the preset classification algorithm. The preset classification algorithm can be the PointNet algorithm, which is determined based on the preset classification algorithm After finding the target type of the target object, it is necessary to judge whether the target type of the target object is in the preset category set. The specific judgment process is as follows:

The server extracts multiple target features of the target object, and matches the multiple target features with the preset category feature set. The preset category feature set contains the feature set corresponding to each preset object, and determines the feature and the similarity value of each feature set in the preset category feature set, and then determine the maximum similarity value from the obtained multiple similarity values, and obtain the preset category of the preset object corresponding to the maximum similarity value, and The preset category serves as the target type of the target object.

When it is determined that the target type of the target object is in the preset category based on the above description, the width information, height information, and object information of the target object are determined based on the three-dimensional detection frame, and the obtained width information, height information, and target type are obtained. Object information as the target object.

Based on the above description, the object information of the target object is obtained. Since the data preprocessing model and the preset depth model are determined before obtaining the object information, the data preprocessing model and the preset depth model can be obtained based on the object information. Input is beneficial to determine the depth information of the target object.

Step S2: Transform the object information into a multi-dimensional target feature vector based on the data preprocessing model.

Since the above has described the process of obtaining the data preprocessing model and the object information of the target object, the height information, width information and target type of the target object can be extracted from the object information, and then the height information, width information and The target type is input into the data preprocessing model to obtain the target two-dimensional feature vector and target category vector corresponding to the target object, and splice the target two-dimensional feature vector and target category vector to obtain the multi-dimensional target feature vector corresponding to the target object.

It should be noted that the schematic diagram of processing the object information of the target object based on the data preprocessing model is shown in Figure 2. In Figure 2, after inputting the object information into the data preprocessing model, the data preprocessing model will respectively The height information and width information of the target object are normalized to obtain the one-dimensional feature vectors corresponding to the height information and width information respectively, and then the multiple target features of the target object are determined by the embedding method to obtain an n-dimensional feature vector, n is positive Integer, finally, splice the one-dimensional feature vector corresponding to the height information and the width information with the n-dimensional feature vector to obtain the n+2-dimensional feature vector, and use the spliced n+2-dimensional feature vector as the data preprocessing model output.

Based on the method described above, the multi-dimensional target feature vector corresponding to the target object is obtained based on the data preprocessing model, so that the multi-dimensional target feature vector can represent the object information of the target object, which is beneficial to obtain the depth information of the target object.

Step S3: Input the multi-dimensional target feature vector into a preset depth model to obtain depth information corresponding to the target object.

The multi-dimensional target feature vector and the preset depth model have been determined above. In order to obtain the depth information of the target object, the multi-dimensional target feature vector needs to be input into the preset depth model. The schematic diagram of obtaining the depth information of the target object based on the preset depth model is shown in the figure 3, in Figure 3, the multi-dimensional target feature vector is input into the preset depth model, the depth information of the target object is predicted through the fully connected model in the preset depth model, and finally, the preset depth information is output , so as to obtain the depth information corresponding to the target object.

Based on the method described above, the height information and width information of the target object are determined through the lidar, and then the depth information of the target object is predicted through the trained data preprocessing model and the preset depth model, avoiding the For objects with inaccurate depth information of the target object, since the data preprocessing model and the preset depth model are models that have been trained many times, the accuracy of the depth information of the target object obtained based on the preset depth model can be ensured.

Based on the same inventive concept, an electronic device is also provided in an embodiment of the present application, which can realize the function of the aforementioned device for determining the depth of a target object. Referring to FIG. 4, the electronic device includes:

An obtaining module 401, configured to obtain object information corresponding to a target object, wherein the object information includes: height information, width information, and target type of the target object;

A conversion module 402, configured to convert the object information into a multidimensional target feature vector based on a data preprocessing model;

The depth module 403 is configured to input the multi-dimensional target feature vector into a preset depth model to obtain depth information corresponding to the target object.

In a possible design, the obtaining module 401 is specifically configured to obtain the test object information corresponding to each of the m test objects, perform training based on the m test objects, and obtain the training model corresponding to the m test objects , in response to the training model meeting the preset condition, determine the preset depth models corresponding to the m test objects.

In a possible design, the obtaining module 401 is further configured to determine the maximum value of all height information, width information, and depth information corresponding to the m test objects, and the m test objects correspond to Based on the maximum value, the corresponding height information and width information of each test object are normalized to obtain the two-dimensional feature vector corresponding to each test object, and the respective test object The two-dimensional feature vectors and their corresponding category feature vectors are spliced to obtain the multi-dimensional feature vectors corresponding to the respective test objects, and the training is performed based on the depth information corresponding to the respective test objects and the respective multi-dimensional feature vectors to obtain the respective multi-dimensional feature vectors. The training model corresponding to the m test objects.

In a possible design, the obtaining module 401 is also used to obtain the total number of training times corresponding to the m test objects, and determine the loss value of the training model corresponding to each training of the m test objects, When the total number of training times reaches the preset number of training times, the response is that the training model meets the preset condition, and when the loss value is smaller than the preset loss threshold, the response is that the training model meets the preset condition.

In a possible design, the obtaining module 401 is further configured to determine the three-dimensional detection frame of the target object, obtain the target type of the target object based on a preset classification algorithm, and respond to the Assuming that the categories are concentrated, the height information and width information of the target object are read out based on the three-dimensional detection frame, and the target type, the height information, and the width information are used as the object information corresponding to the target object.

In a possible design, the obtaining module 401 is further configured to extract multiple target features corresponding to the target object, match the multiple target features with a preset category feature set, and determine the A plurality of similarity values corresponding to a plurality of target feature sets, determining a preset category of a preset object corresponding to the maximum similarity from the multiple similarity values, and using the preset category as the target object target type.

In a possible design, the conversion module 402 is specifically configured to input the object information corresponding to the target object into the data preprocessing model, and obtain the target two-dimensional feature vector and target category of the target object vector, combining the target two-dimensional feature vector and the target category vector to obtain a multi-dimensional target feature vector corresponding to the target object.

Based on the same inventive concept, the embodiment of the present application also provides an electronic device, which can realize the function of the aforementioned device for determining the depth of a target object. Referring to FIG. 5, the electronic device includes:

At least one processor 501, and a memory 502 connected to at least one processor 501. The embodiment of the present application does not limit the specific connection medium between the processor 501 and the memory 502. In FIG. 5, the connection between the processor 501 and the memory 502 Take the connection through the bus 500 as an example. The bus 500 is represented by a thick line in FIG. 5 , and the connection manners between other components are only for schematic illustration and are not limited thereto. The bus 500 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in FIG. 5 , but it does not mean that there is only one bus or one type of bus. Alternatively, the processor 501 may also be called a controller, and the name is not limited.

In the embodiment of the present application, the memory 502 stores instructions executable by at least one processor 501, and at least one processor 501 executes the instructions stored in the memory 502 to perform a method for determining the depth of a target object discussed above. The processor 501 may implement functions of various modules in the apparatus shown in FIG. 4 .

Wherein, the processor 501 is the control center of the device, and various interfaces and lines can be used to connect various parts of the entire control device, and by running or executing instructions stored in the memory 502 and calling data stored in the memory 502, the Various functions and processing data of the device, so as to monitor the device as a whole.

In a possible design, the processor 501 may include one or more processing units, and the processor 501 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application programs etc., the modem processor mainly handles wireless communication. It can be understood that the foregoing modem processor may not be integrated into the processor 501 . In some embodiments, the processor 501 and the memory 502 can be implemented on the same chip, and in some embodiments, they can also be implemented on independent chips.

The processor 501 may be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. Realize or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general purpose processor may be a microprocessor or any conventional processor or the like. The steps of a method for determining the depth of a target object disclosed in the embodiments of the present application may be directly executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

The memory 502, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs and modules. The memory 502 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Magnetic Memory, Disk, discs and more. Memory 502 is, but is not limited to, any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. The memory 502 in the embodiment of the present application may also be a circuit or any other device capable of implementing a storage function, and is used for storing program instructions and/or data.

By designing and programming the processor 501, the code corresponding to a method for determining the depth of the target object introduced in the foregoing embodiments can be solidified into the chip, so that the chip can execute the implementation of the embodiment shown in Figure 1 during operation. A procedure for determining the depth of a target object. How to design and program the processor 501 is well known to those skilled in the art and will not be repeated here.

Based on the same inventive concept, an embodiment of the present application also provides a storage medium, the storage medium stores computer instructions, and when the computer instructions are run on the computer, the computer executes the method for determining the depth of the target object discussed above.

In some possible implementations, various aspects of the method for determining the depth of a target object provided by the present application can also be implemented in the form of a program product, which includes program code. When the program product is run on the device, the program code uses The control device executes the steps in a method for determining the depth of a target object according to various exemplary embodiments of the present application described above in this specification.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

Obviously, those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. In this way, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims (10)

1. A method of determining a depth of a target object, comprising:

obtaining object information corresponding to a target object, wherein the object information comprises: the height information, the width information and the target type of the target object;

converting the object information into a multi-dimensional target feature vector based on a data preprocessing model;

and inputting the multi-dimensional target feature vector into a preset depth model to obtain depth information corresponding to the target object.

2. The method of claim 1, comprising, prior to obtaining object information corresponding to the target object:

obtaining test object information corresponding to m test objects, wherein the test object information is the height information, the width information and the depth information of the test objects and the type information of the test objects, and m is a positive integer;

training based on the m test objects to obtain training models corresponding to the m test objects;

and determining a preset depth model corresponding to the m test objects in response to the training model meeting preset conditions.

3. The method of claim 2, wherein training based on the m test objects to obtain training models corresponding to the m test objects comprises:

determining the maximum value of all height information, width information and depth information corresponding to the m test objects and category feature vectors corresponding to the m test objects respectively;

normalizing the height information and the width information corresponding to each test object based on the maximum value to obtain two-dimensional feature vectors corresponding to each test object;

splicing the two-dimensional feature vectors of each test object and the corresponding class feature vectors to obtain the multidimensional feature vectors of each test object;

training is carried out based on the depth information corresponding to each test object and the multidimensional feature vectors, and training models corresponding to the m test objects are obtained.

4. The method of claim 2, wherein responding to the training model meeting a preset condition comprises:

obtaining total training times corresponding to the m test objects, and determining a loss value of a training model corresponding to each training of the m test objects, wherein the loss value represents the difference between a model prediction depth value and an actual training sample depth value;

when the total training times reach preset training times, responding to the training model to accord with preset conditions; and

and when the loss value is smaller than a preset loss threshold value, responding to the training model to meet preset conditions.

5. The method of claim 1, wherein obtaining object information corresponding to the target object comprises:

determining a three-dimensional detection frame of the target object, and obtaining a target type of the target object based on a preset classification algorithm;

responding to the target type in a preset category set, and reading out the height information and the width information of the target object based on the three-dimensional detection frame, wherein the preset category set comprises a plurality of categories corresponding to preset objects;

and taking the target type, the height information and the width information as object information corresponding to the target object.

6. The method of claim 5, wherein obtaining the target type of the target object based on a preset classification algorithm comprises:

extracting a plurality of target characteristics corresponding to the target object;

matching the plurality of target features with a preset category feature set to determine a plurality of similarity values corresponding to the plurality of target feature sets, wherein the preset category feature set comprises feature sets corresponding to each preset object;

and determining a preset category of the preset object corresponding to the maximum similarity from the plurality of similarity values, and taking the preset category as the target type of the target object.

7. The method of claim 1, wherein converting the object information into a multi-dimensional target feature vector based on a data preprocessing model comprises:

inputting the object information corresponding to the target object into the data preprocessing model to obtain a target two-dimensional feature vector and a target category vector of the target object;

and combining the target two-dimensional feature vector and the target category vector to obtain a multi-dimensional target feature vector corresponding to the target object.

8. An apparatus for determining the depth of a target object, comprising:

the device comprises an obtaining module, a processing module and a processing module, wherein the obtaining module is used for obtaining object information corresponding to a target object, and the object information comprises: the height information, the width information and the target type of the target object;

the conversion module is used for converting the object information into a multidimensional target feature vector based on a data preprocessing model;

and the depth module is used for inputting the multi-dimensional target feature vector into a preset depth model to obtain depth information corresponding to the target object.

9. An electronic device, comprising:

a memory for storing a computer program;

a processor for carrying out the method steps of any one of claims 1-7 when executing a computer program stored on said memory.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored therein a computer program which, when executed by a processor, implements the method steps of any of claims 1-7.

Images