Feature extraction method and device, vehicle and medium

Abstract

The invention provides a feature extraction method and device, a vehicle and a medium, and relates to the technical field of automatic driving, an image collected by a camera of the vehicle is obtained, then the image is input into a feature extraction model to obtain a target feature, and for the feature extraction model, sample images for training the model comprise a positive sample image and a negative sample image, the sample image carries the positioning information so as to distinguish the positive sample image from the negative sample image, and the sample image does not need to be additionally labeled through the feature extraction model which carries the positioning information and is obtained in an unsupervised training mode, so that the labeling cost is avoided, the training cost of the subsequent feature extraction model is correspondingly reduced, and the training efficiency is improved. Therefore, the cost of feature extraction is reduced.

Description

Technical Field

[0001] The present disclosure relates to the field of autonomous driving technology, and in particular to a feature extraction method, device, vehicle and medium. Background Art

[0002] As the diversity and number of sensors equipped on self-driving cars become more and more complex, different sensors collect information from different perspectives. It is necessary to represent the features of different perspectives with a unified perspective to facilitate subsequent operations such as driving planning. For example, the features of different perspectives can be used in the bird's eye view (BEV) space. At present, the above features are usually output by a model. However, when training the model, the cost of data collection and annotation is high, which increases the training cost of the model, and thus also makes the cost of acquiring the above features high. Summary of the invention

[0003] In order to overcome the problems existing in the related art, the present disclosure provides a feature extraction method, device, vehicle and medium.

[0004] According to a first aspect of an embodiment of the present disclosure, a feature extraction method is provided, the method comprising: acquiring an image captured by a camera of a vehicle; inputting the image into a feature extraction model to obtain target features, wherein the feature extraction model is obtained by unsupervised training of sample images, the sample images carry positioning information, the sample images include positive sample images and negative sample images, and the positioning information is used to distinguish the positive sample images from the negative sample images.

[0005] Optionally, before inputting the image into the feature extraction model to obtain the target feature, it also includes: obtaining a sample image; training the initial model with the sample image to obtain a trained initial model; determining a loss function corresponding to the trained initial model; obtaining a new sample image, and iteratively training the trained initial model according to the new sample image and the loss function until the preset training conditions are met to obtain the feature extraction model.

[0006] Optionally, acquiring sample images includes: acquiring multiple images to be selected, wherein each of the multiple images to be selected carries positioning information, and the positioning information includes a shooting position and a shooting area; determining the positive sample image from the multiple images to be selected, and determining the negative sample image from the multiple images to be selected, wherein the distance between the shooting position corresponding to any of the positive sample images and the shooting positions corresponding to the remaining positive samples is less than a preset distance, and the shooting areas corresponding to each of at least two of the negative sample images are different.

[0007] Optionally, the number of the images is multiple, the number of the cameras is multiple, each of the multiple images is captured by each camera, and the shooting angle of view of the multiple cameras is at least 270°.

[0008] Optionally, the method further includes: performing feature fusion processing on the target features to obtain target features after feature fusion.

[0009] Optionally, the method further includes: converting the format of the target feature to obtain a converted target feature; matching the converted target feature with a preset feature in a preset map to obtain a matching result, wherein the format corresponding to the converted target feature is the same as the format corresponding to the preset feature; and repositioning based on the matching result to obtain the location information of the vehicle.

[0010] Optionally, the target feature includes a bird's-eye view bev feature.

[0011] According to a second aspect of an embodiment of the present disclosure, a feature extraction device is provided, the device comprising: an acquisition module, for acquiring an image acquired by a camera of a vehicle; a feature extraction module, for inputting the image into a feature extraction model to obtain target features, wherein the feature extraction model is obtained by unsupervised training of sample images, the sample images carry positioning information, the sample images include positive sample images and negative sample images, and the positioning information is used to distinguish the positive sample images from the negative sample images.

[0012] According to a third aspect of an embodiment of the present disclosure, a vehicle is provided, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the steps of the feature extraction method described in the first aspect when executing the instructions.

[0013] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the feature extraction method provided in the first aspect of the present disclosure are implemented.

[0014] The present disclosure provides a feature extraction method, device, vehicle and medium, which obtain images captured by a camera associated with an automatic driving system of a vehicle, and then input the images into a feature extraction model to obtain target features. For the feature extraction model, sample images for training the model include positive sample images and negative sample images, and the sample images carry positioning information to distinguish between positive sample images and negative sample images. The feature extraction model obtained by carrying positioning information and adopting an unsupervised training method does not need to additionally annotate the sample images, so there is no annotation cost, which correspondingly reduces the training cost of the subsequent feature extraction model, thereby reducing the cost of feature extraction.

[0015] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0017] Figure 1 is a flow chart of a feature extraction method according to an exemplary embodiment;

[0018] Figure 2 is a flow chart of a feature extraction method according to another exemplary embodiment;

[0019] Figure 3 is a flow chart of a feature extraction method according to another exemplary embodiment;

[0020] Figure 4 is a block diagram of a feature extraction device according to an exemplary embodiment;

[0021] Figure 5 is a block diagram of a vehicle showing a feature extraction method according to an exemplary embodiment;

[0022] Figure 6 It is a block diagram of a device for a feature extraction method according to an exemplary embodiment. DETAILED DESCRIPTION

[0023] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0024] It should be noted that all actions of acquiring signals, information or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0025] As the diversity and number of sensors equipped in self-driving cars become more and more complex, multiple sensors collect information from different perspectives, and the data provided by different sensors may have different characteristics and representation methods, which is not convenient for directly providing vision to the driver. Therefore, it is necessary to represent the features of different perspectives with a unified perspective to facilitate subsequent operations such as driving planning. For example, features from different perspectives can be used in the bev space. At present, the above features are usually output by a model. However, when training the model, the cost of data collection and annotation is high, which increases the training cost of the model, and thus also makes the cost of obtaining the above features high.

[0026] To solve the above problems, please refer to Figure 1 , Figure 1 is a flow chart of a feature extraction method according to an exemplary embodiment. The feature extraction method can be applied to Figure 4 The feature extraction device 300 shown, Figure 5 Vehicle 600 and computer readable storage medium. In this embodiment, the application to a vehicle is taken as an example. Figure 1 The process shown is described in detail, and the feature extraction method can specifically include the following steps:

[0027] Step S110: Acquire an image captured by a camera of the vehicle.

[0028] The vehicle is equipped with a camera associated with the autonomous driving system. When the vehicle is driving through the autonomous driving system, the camera on the vehicle collects images along the way.

[0029] Optionally, the number of cameras provided on the vehicle may be one.

[0030] Optionally, the number of cameras installed on the vehicle can be multiple, and multiple cameras can simultaneously capture images to obtain multiple images in different directions. By simultaneously capturing images with multiple cameras, a wider field of view can be obtained through multiple images in different directions. For example, the shooting angle of view of multiple cameras is at least 270°, which can improve driving safety. In addition, multiple cameras can be redundant with each other. When any camera fails, other cameras can still capture images and continue to provide a field of view for the vehicle, further ensuring driving safety.

[0031] For example, the number of cameras may be 4, which are respectively installed at the left front, right front, left rear and right rear of the vehicle, so as to collect perspectives from various directions and obtain a wide field of view. It should be noted that the number of cameras is not limited to the above 4, and may be a smaller number, for example, the number of cameras may be 3. It may also be a larger number, for example, the number of cameras may be 6, 8, etc.

[0032] Step S120: input the image into a feature extraction model to obtain target features, wherein the feature extraction model is obtained by unsupervised training of sample images, the sample images carry positioning information, the sample images include positive sample images and negative sample images, and the positioning information is used to distinguish the positive sample images from the negative sample images.

[0033] In one embodiment, the feature extraction model is stored locally in the vehicle, the vehicle obtains the feature extraction model from the local storage location, and then inputs the image into the feature extraction model to obtain the target feature output by the feature extraction model. This embodiment stores the feature extraction model locally in the vehicle, which facilitates the vehicle to quickly obtain the feature extraction model, so as to quickly extract the target feature.

[0034] In another embodiment, the feature extraction model is stored in a server connected to the vehicle, the vehicle downloads the feature extraction model from the server connected to it, and then inputs the image into the feature extraction model to obtain the target features output by the feature extraction model. In this embodiment, the feature extraction model is stored in a server connected to the vehicle, which can reduce the occupancy of the vehicle's local memory by the feature extraction model.

[0035] Exemplarily, the target feature includes a bird's-eye view BEV feature. In one embodiment, the image captured by the camera is an image in a two-dimensional space. Since the directly captured image can only provide a limited field of view, especially in the side field of view and the rear field of view, there are blind spots, which makes the vehicle unable to fully perceive the entire environment. In addition, due to the perspective effect, the directly captured image makes distant objects appear smaller and nearby objects appear larger, which will cause the shape and size of the object on the two-dimensional image to be different from the actual situation, making object detection and distance measurement more difficult. Therefore, the image is input into the feature extraction model, and the two-dimensional image is spliced ​​and processed into a BEV feature of a three-dimensional BEV space through the feature extraction model. The BEV feature is more realistic than the feature of the two-dimensional space, and the spliced ​​BEV feature has a wider field of view, which can display obstacles in all directions of the vehicle, thereby improving the safety and comfort of the vehicle's automatic driving. Optionally, the feature extraction model can be a RegNet model.

[0036] The feature extraction method provided in this embodiment obtains an image captured by a camera associated with the vehicle's automatic driving system, and then inputs the image into a feature extraction model to obtain target features. For the feature extraction model, sample images for training the model include positive sample images and negative sample images. The sample images carry positioning information to distinguish between positive sample images and negative sample images. The feature extraction model obtained by unsupervised training with positioning information does not need to additionally label the sample images, so there is no labeling cost, which correspondingly reduces the training cost of the subsequent feature extraction model, thereby reducing the cost of feature extraction.

[0037] Optionally, before step S120, a feature extraction model needs to be pre-trained. For example, see Figure 2 , the feature extraction method further includes:

[0038] Step S210: Acquire a sample image.

[0039] In one embodiment, a plurality of images to be selected are obtained, wherein each of the plurality of images to be selected carries positioning information. As a method, each of the plurality of cameras associated with the automatic driving system collects positioning information while collecting the images to be selected. It can be understood that a position sensor is installed corresponding to the position where each camera is located, and the camera and the position sensor are bound. The positioning information collected by the sensor is the positioning information of the image to be selected collected by the camera. Optionally, the position sensor can be a global positioning system (GPS), an ultrasonic sensor, etc. Optionally, the plurality of images to be selected can be collected by the camera of the same vehicle, or by the cameras of different vehicles. Optionally, the positioning information includes a shooting location and a shooting area. It can be understood that the positioning information consists of a shooting area and a shooting location. The shooting area is a larger range than the shooting location. For example, the shooting area can be the area where the province is located, the area where the city is located, the area where the community is located, the area where the parking lot is located, etc. The shooting location is a more specific positioning, such as longitude and latitude. Then, the positive sample image is determined from the multiple images to be selected, and the negative sample image is determined from the multiple images to be selected, wherein the distance between the shooting position corresponding to any of the positive sample images and the shooting position corresponding to the remaining positive samples is less than a preset distance, and the shooting areas corresponding to each of at least two of the negative sample images are different. For example, the preset distance is 5 meters.

[0040] Optionally, the ratio and distribution of positive and negative images will also affect the training results of the model. An appropriate ratio and balance of positive and negative images can help the model learn and recognize better. The number of positive and negative images is selected according to the performance of the model.

[0041] Optionally, the selected positive sample images may be taken by cameras of different vehicles, or by different cameras of the same vehicle. Similarly, the selected negative sample images may be taken by cameras of different vehicles, or by different cameras of the same vehicle.

[0042] Step S220: training the initial model using the sample image to obtain a trained initial model.

[0043] The sample images are input into the initial model for unsupervised training to obtain the trained initial model. During the model training process, the positive sample images are used to help the model learn and identify the target features, thereby improving the accuracy of the model. The negative sample images are used as comparison and reference during the training process. Through the negative sample images, the model can learn the difference between the target features and other features, thereby improving the generalization ability and robustness of the model.

[0044] Step S230: Determine the loss function corresponding to the trained initial model.

[0045] Both positive and negative images are used to calculate the loss function so that the model can be optimized and adjusted. For example, the positive and negative images are input into the output features obtained by the initial model, and the loss function is constructed based on the difference between the positive image and the output features.

[0046] It should be noted that the construction method of the loss function is not limited to this, and can also be designed according to the characteristics of the specific model and task, aiming to extract the implicit information to the greatest extent and optimize the prediction or generation ability of the model.

[0047] Optionally, the loss function may be a reconstruction loss function, a contrast loss function, a distribution matching loss function, a divergence loss function, or the like.

[0048] Step S240: obtain a new sample image, and iteratively train the trained initial model according to the new sample image and the loss function until a preset training condition is met, thereby obtaining the feature extraction model.

[0049] New sample images are obtained. The method for obtaining the new sample images can be as shown in step S210, which will not be described in detail. Optionally, the number of new sample images can be adjusted according to the trained initial model, so that the number of new sample images can be different from the number of previous sample images.

[0050] The preset training condition may be that the obtained loss function converges, or that the number of training times reaches a preset number.

[0051] Optionally, the method further includes: performing feature fusion processing on the target features to obtain target features after feature fusion.

[0052] This implementation method performs feature fusion on target features, which can integrate information between different features and comprehensively consider information from different perspectives, thereby forming a more comprehensive feature representation and improving the safety of autonomous driving.

[0053] Optionally, the feature extraction method provided by the present disclosure can be applied in an application scenario of vehicle relocation, and the method further includes: converting the format of the target feature to obtain the converted target feature, for example, the format of the target feature is chw format, and the format of the converted target feature is cee format. Matching the converted target feature with the preset feature in the preset map to obtain a matching result, wherein the format corresponding to the converted target feature is the same as the format corresponding to the preset feature, so that the two can be matched. Relocation is performed based on the matching result to obtain the location information of the vehicle.

[0054] Optionally, the feature extraction method provided by the present disclosure is not limited to the automatic driving system of the above-mentioned vehicle, but can also be applied to the automatic driving system of a drone. The target features obtained by the feature extraction method provided by the present disclosure can also be applied to application scenarios such as loop detection and parking perception. The target features can provide a wider field of view to complete target recognition, thereby effectively avoiding obstacles and effectively increasing the comfort and safety of automatic driving.

[0055] Based on the above embodiment, this embodiment provides a feature extraction method. Figure 3 The method includes: each camera on a plurality of cameras associated with the vehicle's automatic driving system collects an image to obtain a plurality of images, the plurality of images (imags) are input into a feature extraction model to obtain bev features output by the feature extraction model, the bev features are then fused through a feature fusion layer to obtain fused bev features, the bev features are formatted through image index features to obtain bev features after format conversion, and relocalization, loop detection and other operations are performed based on the bev features after format conversion.

[0056] The feature extraction method provided by the present disclosure is compatible with a multi-camera model of multiple cameras and feature extraction models. Since multiple cameras are installed on the vehicle, when some of the multiple cameras have abnormalities, the other cameras can still collect images normally. When training the model, only images with positioning information are needed, and there are no additional annotation requirements, which is low cost. In addition, optimizing the perception algorithm based on an iterative method can steadily improve the perception performance, focusing on images of different difficulties at different stages, so that negative sample images can be fully utilized to improve the performance of the model.

[0057] Based on the same inventive concept, the present disclosure also provides a feature extraction device, Figure 4is a block diagram of a feature extraction device according to an exemplary embodiment. Figure 4 , the feature extraction device 300 includes: a collection module 310 and a feature extraction module 320;

[0058] The acquisition module 310 is used to acquire images acquired by a camera of the vehicle;

[0059] The feature extraction module 320 is used to input the image into a feature extraction model to obtain target features, wherein the feature extraction model is obtained by unsupervised training of sample images, the sample images carry positioning information, the sample images include positive sample images and negative sample images, and the positioning information is used to distinguish the positive sample images from the negative sample images.

[0060] Optionally, the feature extraction device 300 further includes: a sampling module, a first training module, a loss calculation module and a second training module;

[0061] A sampling module, used to obtain a sample image;

[0062] A first training module, used to train the initial model with the sample image to obtain a trained initial model;

[0063] The loss calculation module is used to determine the loss function corresponding to the initial model after training;

[0064] The second training module is used to obtain a new sample image, and iteratively train the trained initial model according to the new sample image and the loss function until a preset training condition is met to obtain the feature extraction model.

[0065] Optionally, the sampling module includes: a first sampling module and a second sampling module;

[0066] A first sampling module is used to obtain a plurality of images to be selected, wherein each of the plurality of images to be selected carries positioning information, and the positioning information includes a shooting position and a shooting area;

[0067] A second sampling module is used to determine the positive sample image from the multiple images to be selected, and to determine the negative sample image from the multiple images to be selected, wherein the distance between the shooting position corresponding to any one of the positive sample images and the shooting position corresponding to the remaining positive samples is less than a preset distance, and the shooting area corresponding to each of at least two of the negative sample images is different.

[0068] Optionally, the number of the images is multiple, the number of the cameras is multiple, each of the multiple images is captured by each camera, and the shooting angle of view of the multiple cameras is at least 270°.

[0069] Optionally, the feature extraction device 300 further includes: a feature fusion module;

[0070] The feature fusion module is used to perform feature fusion processing on the target features to obtain the target features after feature fusion.

[0071] Optionally, the feature extraction device 300 further includes: a conversion module, a matching module and a relocation module;

[0072] A conversion module, used for converting the target feature into a format to obtain the converted target feature;

[0073] A matching module, used for matching the converted target feature with a preset feature in a preset map to obtain a matching result, wherein the format corresponding to the converted target feature is the same as the format corresponding to the preset feature;

[0074] A repositioning module is used to perform repositioning based on the matching result to obtain the position information of the vehicle.

[0075] Optionally, the target feature includes a bird's-eye view bev feature.

[0076] Regarding the feature extraction device 300 in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0077] The present disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, and when the program instructions are executed by a processor, the steps of the feature extraction method provided by the present disclosure are implemented.

[0078] Figure 5 is a block diagram of a vehicle 600 according to an exemplary embodiment. For example, the vehicle 600 may be a hybrid vehicle, a non-hybrid vehicle, an electric vehicle, a fuel cell vehicle, or other types of vehicles. The vehicle 600 may be an autonomous vehicle or a semi-autonomous vehicle.

[0079] Reference Figure 5 , the vehicle 600 may include various subsystems, for example, an infotainment system 610, a perception system 620, a decision control system 630, a drive system 640, and a computing platform 650. The vehicle 600 may also include more or fewer subsystems, and each subsystem may include multiple components. In addition, each subsystem and each component of the vehicle 600 may be interconnected by wire or wireless means.

[0080] In some embodiments, the infotainment system 610 may include a communication system, an entertainment system, and a navigation system, among others.

[0081] The perception system 620 may include several sensors for sensing information about the environment around the vehicle 600. For example, the perception system 620 may include a global positioning system (the global positioning system may be a GPS system, or a Beidou system or other positioning systems), an inertial measurement unit (IMU), a laser radar, a millimeter wave radar, an ultrasonic radar, and a camera.

[0082] The decision control system 630 may include a computing system, a vehicle controller, a steering system, a throttle, and a braking system.

[0083] The drive system 640 may include components that provide powered motion for the vehicle 600. In one embodiment, the drive system 640 may include an engine, an energy source, a transmission system, and wheels. The engine may be one or a combination of an internal combustion engine, an electric motor, and an air compression engine. The engine is capable of converting energy provided by the energy source into mechanical energy.

[0084] Some or all functions of the vehicle 600 are controlled by a computing platform 650. The computing platform 650 may include at least one processor 651 and a memory 652, and the processor 651 may execute instructions 653 stored in the memory 652.

[0085] The processor 651 may be any conventional processor, such as a commercially available CPU. The processor may also include a graphics processor (Graphic Process Unit, GPU), a field programmable gate array (Field Programmable Gate Array, FPGA), a system on chip (System on Chip, SOC), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC) or a combination thereof.

[0086] The memory 652 may be implemented by any type of volatile or nonvolatile 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.

[0087] In addition to the instructions 653 , the memory 652 may also store data, such as road maps, route information, and data such as the location, direction, and speed of the vehicle. The data stored in the memory 652 may be used by the computing platform 650 .

[0088] In the embodiment of the present disclosure, the processor 651 may execute instruction 653 to complete all or part of the steps of the above-mentioned feature extraction method.

[0089] Figure 6 1900 is a block diagram of a device for a feature extraction method according to an exemplary embodiment. For example, the device 1900 may be provided as a server. Figure 6 , the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions executable by the processing component 1922, such as an application. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above-mentioned feature extraction method.

[0090] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output interface 1958. The device 1900 may operate based on an operating system stored in the memory 1932.

[0091] In another exemplary embodiment, a computer program product is further provided. The computer program product includes a computer program executable by a programmable device. The computer program has a code portion for executing the above feature extraction method when executed by the programmable device.

[0092] In summary, the feature extraction method, device, vehicle and medium provided by the present invention obtain images collected by a camera associated with the vehicle's automatic driving system, and then input the images into a feature extraction model to obtain target features. For the feature extraction model, sample images for training the model include positive sample images and negative sample images. The sample images carry positioning information to distinguish between positive sample images and negative sample images. The feature extraction model obtained by carrying positioning information and adopting an unsupervised training method does not need to additionally label the sample images, so there is no labeling cost, which correspondingly reduces the training cost of the subsequent feature extraction model, thereby reducing the cost of feature extraction.

[0093] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the present disclosure. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0094] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A feature extraction method, characterized in that: The method comprises: Obtaining images captured by a camera of a vehicle; The image is input into a feature extraction model to obtain target features, wherein the feature extraction model is obtained by unsupervised training of sample images, the sample images carry positioning information, the sample images include positive sample images and negative sample images, and the positioning information is used to distinguish the positive sample images from the negative sample images.

2. The method according to claim 1, characterized in that Before inputting the image into a feature extraction model to obtain target features, the method further includes: Get a sample image; Training the initial model with the sample image to obtain a trained initial model; Determine the loss function corresponding to the initial model after training; A new sample image is obtained, and the trained initial model is iteratively trained according to the new sample image and the loss function until a preset training condition is met, thereby obtaining the feature extraction model.

3. The method according to claim 2, characterized in that The obtaining of the sample image comprises: Acquire a plurality of images to be selected, wherein each of the plurality of images to be selected carries positioning information, and the positioning information includes a shooting position and a shooting area; The positive sample image is determined from the multiple images to be selected, and the negative sample image is determined from the multiple images to be selected, wherein the distance between the shooting position corresponding to any one of the positive sample images and the shooting position corresponding to the remaining positive samples is less than a preset distance, and the shooting area corresponding to each of at least two of the negative sample images is different.

4. The method according to claim 1, characterized in that: The number of the images is multiple, the number of the cameras is multiple, each of the multiple images is captured by each camera, and the shooting angle of view of the multiple cameras is at least 270°.

5. The method according to claim 1, characterized in that Also includes: Perform feature fusion processing on the target features to obtain target features after feature fusion.

6. The method according to claim 1, characterized in that Also includes: Convert the target feature into a format to obtain a converted target feature; Matching the converted target feature with a preset feature in a preset map to obtain a matching result, wherein the format corresponding to the converted target feature is the same as the format corresponding to the preset feature; Repositioning is performed based on the matching result to obtain the position information of the vehicle.

7. The method according to any one of claims 1 to 6, characterized in that: The target features include bird's-eye view bev features.

8. A feature extraction device, characterized in that: The device comprises: An acquisition module, used to obtain images captured by a camera of a vehicle; A feature extraction module is used to input the image into a feature extraction model to obtain target features, wherein the feature extraction model is obtained by unsupervised training of sample images, the sample images carry positioning information, the sample images include positive sample images and negative sample images, and the positioning information is used to distinguish the positive sample images from the negative sample images.

9. A vehicle, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the steps of any one of the methods of claims 1 to 7 when executing the instructions.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images