Intelligent driving relocation method, vehicle and storage medium

Abstract

The embodiment of the invention relates to an intelligent driving repositioning method, a vehicle and a storage medium, and relates to the technical field of intelligent driving, and the method comprises the steps: obtaining a current road condition image, initial positioning information matched with the timestamp of the current road condition image, and a positioning code feature corresponding to the initial positioning information; the positioning code features, the image features of the current road condition image and the image features of the corresponding candidate map images are input into an attention feature matching model together for feature matching, the model is guided to pay attention to an area consistent with the initial positioning information, and deep interaction of the image features and initial positioning information constraints is achieved; the accuracy of feature matching is improved, so that when the current positioning result of the vehicle is determined based on the feature matching result output by the attention feature matching model, the robustness of the repositioning precision can be improved based on the accurate feature matching result, and the accuracy of the repositioning precision can be improved by determining the feature matching between the candidate map image and the current road condition image. The repositioning efficiency can be improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of intelligent driving, and in particular to an intelligent driving repositioning method, a vehicle and a storage medium. BACKGROUND

[0002] In an intelligent driving system, a vehicle needs to rely on a repositioning function to quickly recover an accurate pose in the case of positioning initialization failure, short-time occlusion or long-time failure.

[0003] In related technologies, when visual repositioning is performed, some repositioning schemes perform coarse positioning based on sensor data such as a Global Navigation Satellite System (GNSS), an Inertial Measurement Unit (IMU) and a wheel speed sensor, and narrow down the matching range of repositioning based on a coarse positioning result. However, this way only relies on position constraints for repositioning, and it is difficult to guarantee the repositioning accuracy in a complex environment. SUMMARY

[0004] The present disclosure provides an intelligent driving repositioning method, a vehicle and a storage medium to solve the problem that it is difficult to guarantee the repositioning accuracy in a complex environment in related technologies only relying on position constraints for repositioning.

[0005] In a first aspect, an embodiment of the present disclosure provides an intelligent driving repositioning method, comprising: obtaining a current road image collected by a vehicle, initial positioning information matched with a timestamp of the current road image, and positioning encoding features corresponding to the initial positioning information; determining a candidate map image from a map image library based on the initial positioning information; performing feature extraction on the current road image and the candidate map image respectively to obtain road features corresponding to the current road image and map features corresponding to the candidate map image; inputting the positioning encoding features, the road features and the map features into a pre-trained attention feature matching model, and obtaining a feature matching result of the road features and the map features output by the attention feature matching model; determining a current positioning result of the vehicle based on the feature matching result.

[0006] Optionally, the initial positioning information includes a vehicle position, a vehicle attitude and a pose covariance matrix. The determining of the candidate map image from the map image library based on the initial positioning information comprises: determining an initial search range based on the vehicle position and the pose covariance matrix; adjust the initial search range based on a heading angle in the vehicle pose to obtain a target search range; search the map image library based on the target search range to obtain the candidate map image.

[0007] In this embodiment, by acquiring the vehicle position, the vehicle pose and the pose covariance matrix during coarse positioning, and determining the initial search range based on the vehicle position and the pose covariance matrix, the search range is dynamically constrained by the pose covariance matrix, so as to improve the search efficiency. Furthermore, the determined initial search range is adjusted based on the heading angle of the vehicle, so that the adjusted target search range is consistent with the driving direction of the vehicle, thereby ensuring a larger search range in the driving direction of the vehicle, and improving the robustness of the candidate map image searched.

[0008] Optionally, the determining the initial search range based on the vehicle position and the pose covariance matrix comprises: performing eigenvalue decomposition on the pose covariance matrix to obtain an X-axis direction offset distance and a Y-axis direction offset distance; determining the initial search range based on the X-axis direction offset distance and the Y-axis direction offset distance with the vehicle position as the center.

[0009] In this embodiment, the X-axis direction offset distance and the Y-axis direction offset distance are determined by performing eigenvalue decomposition on the pose covariance matrix, and then the initial search range is determined in combination with the vehicle position, so as to dynamically constrain the search range. The search range is smaller in the direction with smaller offset distance, and compared with the way of determining the search range with a fixed radius, the search efficiency is improved.

[0010] Optionally, the method further comprises: The candidate map image is at least one, each candidate map image corresponds to one map feature, each map feature and the road condition feature both include a plurality of feature points, and the feature matching result includes a feature point pair matched by each map feature and the road condition feature. The determining the current positioning result of the vehicle based on the feature matching result comprises: acquiring a point pair number of the feature point pair matched by each map feature and the road condition feature in the feature matching result; determining, based on the point pair number, a map feature corresponding to a maximum value of the point pair number as a target map feature from the map features; determining the current positioning result of the vehicle based on a target feature point pair matched by the target map feature and the road condition feature.

[0011] In this embodiment, by obtaining the number of point pairs of each map feature and road feature matched in the feature matching result, the map feature with the largest number of point pairs is selected as a target map feature, and the feature point pair matched between the target map feature and the road feature is used for repositioning. In this way, the feature point pair with high confidence can be screened out for repositioning of the vehicle position, and the accuracy of the repositioning result is improved.

[0012] Optionally, the determining the current positioning result of the vehicle based on the feature matching result comprises: determining a candidate positioning result of the vehicle based on the first feature point pair in the feature matching result; performing confidence checking on the candidate positioning result based on the candidate positioning result and the first feature point pair to determine a first confidence of the candidate positioning result; in a case where the first confidence is greater than or equal to a first threshold, determining the current positioning result of the vehicle based on the candidate positioning result.

[0013] In this embodiment, for the candidate positioning result determined based on the first feature point pair in the feature matching result, the confidence of the candidate positioning result is first checked using the first feature point pair, and then the current positioning result of the vehicle is determined based on the candidate positioning result after the checking is passed. In this way, the reliability of the candidate positioning result is ensured, and thus the accuracy of the final positioning result is improved.

[0014] Optionally, the method further comprises: in a case where the confidence is less than the first threshold, determining a second feature point pair from the feature matching result, the number of the second feature point pair being less than the number of the first feature point pair; performing confidence checking on the candidate positioning result based on the second feature point pair to determine a second confidence of the candidate positioning result; in a case where the second confidence is greater than or equal to a second threshold, determining the current positioning result of the vehicle based on the candidate positioning result.

[0015] In this embodiment, when the confidence checking of the candidate positioning result based on the first feature point pair fails, a second feature point pair with a number less than that of the first feature point pair is further obtained, and the confidence of the candidate positioning result is checked based on the second feature point pair. After the checking is passed, the current positioning result of the vehicle is determined based on the candidate positioning result. By using the second feature point pair with a smaller number for confidence checking, a lightweight checking is realized, the calculation resources are controlled while the accuracy is ensured, and robustness and efficiency are both considered.

[0016] Optionally, the method further comprises: in a case where the second confidence is less than the second threshold, obtaining a heading angle in the initial positioning information; based on the heading angle, adjusting a first search range used for determining the candidate map image to obtain a second search range, the second search range being greater than the first search range; based on the second search range, determining a new candidate map image from the map image library, and determining a current positioning result of the vehicle based on the new candidate map image and the current road condition image.

[0017] In the embodiment, by expanding the search range of the candidate map image in the driving direction of the vehicle when the confidence check fails and re-searching a new candidate map image for repositioning, the success rate of repositioning can be improved.

[0018] Optionally, the determining the current positioning result of the vehicle based on the candidate positioning result comprises: based on the candidate positioning result and the initial positioning information, performing pose fusion to determine the current positioning result of the vehicle.

[0019] In the embodiment, by performing pose fusion based on the candidate positioning result and the initial positioning information to determine the current positioning result of the vehicle, the candidate positioning result can provide a positioning result with higher accuracy, and the initial positioning result can ensure better smoothness, so that by fusing the two to determine the current positioning result of the vehicle, the accuracy and smoothness of the final positioning result can be ensured.

[0020] In a second aspect, the embodiments of the present disclosure provide an intelligent driving repositioning device, comprising: a first acquisition module configured to acquire a current road condition image collected by a vehicle, and initial positioning information matching a timestamp of the current road condition image and a positioning coding feature corresponding to the initial positioning information; a second acquisition module configured to determine a candidate map image from a map image library based on the initial positioning information; a feature extraction module configured to perform feature extraction on the current road condition image and the candidate map image respectively to obtain a road condition feature corresponding to the current road condition image and a map feature corresponding to the candidate map image; a third acquisition module configured to input the positioning coding feature, the road condition feature and the map feature into a pre-trained attention feature matching model, and acquire a feature matching result of the road condition feature and the map feature output by the attention feature matching model; a positioning module configured to determine a current positioning result of the vehicle based on the feature matching result.

[0021] In a third aspect, the embodiments of the present disclosure provide a vehicle, comprising: a processing module; a storage module for storing executable instructions of the processing module; and the processing module is configured to read the executable instructions from the storage module and execute the executable instructions to implement the intelligent driving repositioning method according to the first aspect.

[0022] In a fourth aspect, the embodiments of the present disclosure provide a computer readable storage medium, which stores a computer program for implementing the intelligent driving repositioning method according to the first aspect.

[0023] Through the above technical solution, the intelligent driving repositioning method, the vehicle and the storage medium provided by the present disclosure, by acquiring the current road image, the initial positioning information matched with the timestamp of the current road image and the positioning encoding feature corresponding to the initial positioning information, inputting the positioning encoding feature, the image feature of the current road image and the image feature of the corresponding candidate map image into the attention feature matching model for feature matching, guiding the attention feature matching model to pay attention to the region consistent with the initial positioning information, realizing the deep interaction of the image feature and the initial positioning information constraint, improving the accuracy of feature matching, so that when the current positioning result of the vehicle is determined based on the feature matching result output by the attention feature matching model, the robustness of the repositioning accuracy can be improved based on the accurate feature matching result, and by determining the candidate map image from the map image library based on the initial positioning information for feature matching with the current road image, the repositioning range is reduced, and the efficiency of repositioning can be improved.

[0024] The above description is only a summary of the technical solutions of the present disclosure. In order to enable one skilled in the art to better understand the technical means of the present disclosure, the contents of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present disclosure more obvious and easy to understand, the specific embodiments of the present disclosure are described below. BRIEF DESCRIPTION OF DRAWINGS

[0025] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent and more readily appreciated when considered in connection with the following detailed description, with reference to the accompanying drawings, in which:

[0026] Figure 1 A flowchart of an intelligent driving repositioning method provided by an exemplary embodiment of the present disclosure is shown in the figure; Figure 2 A network structure diagram of an attention feature matching model provided by an exemplary embodiment of the present disclosure is shown in the figure; Figure 3 A flowchart of an intelligent driving repositioning method provided for another exemplary embodiment of the present disclosure is shown; Figure 4 A flowchart of an intelligent driving repositioning method provided for another exemplary embodiment of the present disclosure is shown; Figure 5 A flowchart of an intelligent driving repositioning method provided for another exemplary embodiment of the present disclosure is shown; Figure 6 A structural diagram of an intelligent driving repositioning device provided for an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0027] Embodiments of the present disclosure will be described in more detail by referring to the drawings. Although certain embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein, but rather the embodiments are provided to more thoroughly and completely understand the present disclosure. It is understood that the drawings and embodiments of the present disclosure are for exemplary purposes only and are not intended to limit the scope of protection of the present disclosure.

[0028] It should be understood that each step recited in the method embodiments of the present disclosure can be executed in different orders and / or in parallel. In addition, the method embodiments can include additional steps and / or omit the execution of the steps shown. The scope of the present disclosure is not limited in this respect.

[0029] The term “comprising” and variations thereof as used herein are open-ended, that is, “including but not limited to”. The term “based on” is “based, at least in part, on”. The term “one embodiment” means “at least one embodiment”; the term “another embodiment” means “at least one additional embodiment”; the term “some embodiments” means “at least some embodiments”. Related definitions are given below in the description of the embodiments.

[0030] It should be noted that the terms “first”, “second”, and the like used in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0031] It should be noted that the terms “one”, “multiple” used in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that “one or more” should be understood unless otherwise explicitly stated in the context.

[0032] The names of the messages or information exchanged between the devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0033] Before explaining the specific embodiments of the present disclosure, the professional terms and English abbreviations that the present disclosure may involve are explained as follows: GNSS: Navigation system that provides position, speed, etc. information for the carrier through satellite signals (such as Global Positioning System, Beidou, etc.), which can provide global positioning reference in intelligent driving, but is easily affected by high-rise shelter and multipath effect, leading to positioning drift.

[0034] IMU: A sensor composed of an accelerometer and a gyroscope, which can measure the acceleration and angular velocity of the carrier, and calculate the pose change by integration. The short-term accuracy is high, but the error will accumulate over time and needs to be used with other sensors for fusion.

[0035] SIFT: Scale-Invariant Feature Transform, the full name in English is Scale-Invariant Feature Transform. A classic image local feature extraction algorithm, with scale, rotation, and illumination invariance. By detecting key points in the image and generating feature descriptors, feature matching between different images is achieved, which is used to build the association between images in traditional visual relocalization, but the robustness is limited in large-scale scenes or complex environments.

[0036] In the intelligent driving system, the vehicle needs to rely on the relocalization function to quickly recover the accurate pose in the case of positioning initialization failure, short shelter or long-term failure, etc. In related technologies, the visual relocalization method mainly includes the following two types: The first is a feature point extraction and matching based method, which realizes inter-frame matching by detecting key points and extracting descriptors. Although the robustness is improved, there are still deficiencies in matching efficiency and accuracy in complex environments such as large-scale scenes, repeated structures and light changes. And due to the lack of effective prior screening, it usually needs to be matched in the full map range, and the calculation overhead grows rapidly with the map size, which is difficult to meet the real-time requirements of intelligent driving systems.

[0037] The second is a deep learning driven visual matching method, such as building high-dimensional semantic association between image descriptors based on graph neural network (GNN) or attention mechanism (Transformer). This method has high matching accuracy, but also lacks effective prior screening, and needs to compare features in the full map range, resulting in high calculation cost and slow inference speed, which is not conducive to real-time deployment in resource-constrained vehicle environments.

[0038] At present, in related technologies, some researches try to use multi-sensor data of the vehicle for coarse positioning to narrow the matching range, but generally only stay in position constraint, lack effective fusion mechanism and dynamic matching screening strategy, and are difficult to resist various sensor errors, resulting in large matching range and insufficient relocalization performance, and low efficiency and accuracy in complex environments.

[0039] The intelligent driving repositioning scene generally faces the characteristics of large map size, complex environment, and multiple sensors available. The present disclosure aims to solve the problems of imbalance between matching efficiency and accuracy of related visual repositioning algorithms in complex environments, high computational cost of deep learning methods, and insufficient utilization of multi-modal fusion. By fusing GNSS, IMU, wheel speed sensors, and other multi-modal information, coarse positioning is achieved to obtain coarse positioning and uncertainty. In the attention mechanism feature matching network, the coarse positioning prior information is spliced with the image feature point encoding to guide the attention feature matching model to focus on the area consistent with the prior, and an intelligent driving visual repositioning scheme that balances efficiency and robustness is achieved.

[0040] The intelligent driving repositioning method, vehicle, and storage medium provided by the present disclosure will be explained in detail below with reference to the accompanying drawings.

[0041] Figure 1 A flowchart of the intelligent driving repositioning method provided by an exemplary embodiment of the present disclosure is shown. The method can be executed by the intelligent driving repositioning device provided by the embodiment of the present disclosure. The device can be realized by software and / or hardware, and can be integrated in a vehicle.

[0042] As shown in Figure 1 The intelligent driving repositioning method can include the following steps: Step 101, acquiring a current road condition image collected by a vehicle, initial positioning information matched with a timestamp of the current road condition image, and positioning encoding features corresponding to the initial positioning information.

[0043] The current road condition image is a road condition image of the current location collected by a vehicle-mounted camera during vehicle driving. The timestamp corresponding to the current road condition image is the time when the vehicle-mounted camera collects the current road condition image.

[0044] In this embodiment, for the current road condition image, the initial positioning information matched with the timestamp of the current road condition image can be determined according to the timestamp of the current road condition image.

[0045] The initial positioning information can include the position, attitude, and uncertainty of the vehicle. The uncertainty includes the uncertainty of the position and the uncertainty of the attitude, which is used to represent the error of the position and the attitude. For example, the uncertainty can be represented by a covariance. It can be understood that the position of the vehicle in a three-dimensional space can be represented by XYZ three-axis coordinates, and the corresponding uncertainty includes the uncertainty in the X-axis direction, the uncertainty in the Y-axis direction, and the uncertainty in the Z-axis direction. The attitude of the vehicle usually includes the pitch angle, the roll angle, and the heading angle, and the corresponding uncertainty includes the uncertainty corresponding to the three rotational degrees of freedom, respectively. Therefore, when the initial positioning information includes the position and the attitude, the uncertainty can be represented as a pose covariance matrix containing six elements.

[0046] In this embodiment, when obtaining the initial positioning information matching the timestamp of the current road image, the target data corresponding to the timestamp can be obtained from the data collected by the GNSS, IMU, wheel speed sensor, lidar data, millimeter wave radar, etc. Then, the target data is processed by Kalman filtering, factor graph optimization, particle filtering, extended Kalman filtering, unscented Kalman filtering, etc. to obtain the coarse positioning pose estimation result, i.e., the initial positioning information. These data sources can provide stable spatial positioning reference or auxiliary information. After fusion with other sensor data, the reliability of coarse positioning can be improved, and the robustness of multi-modal fusion can be enhanced. These methods process probability distribution by means of different mathematical models, can efficiently deal with state estimation problems of nonlinear and non-Gaussian systems, have strong adaptability in complex sensor noise characteristics, and can stably output accurate coarse positioning pose estimation results. Among them, Kalman filtering can update state estimation in real time based on sensor noise characteristics, and factor graph optimization can comprehensively adjust multi-time and multi-source data to output the coarse pose estimation corresponding to the current road image and its uncertainty representation, providing accurate spatial prior constraint for subsequent candidate map image screening and feature matching.

[0047] It should be noted that coarse positioning based on GNSS, IMU, wheel speed sensor, etc. is a relatively mature technology, and the present disclosure does not describe it in detail.

[0048] Then, after obtaining the initial positioning information of coarse positioning, the initial positioning information can be encoded to obtain positioning encoding features. The encoding of the initial positioning information to obtain the positioning encoding features can be realized by means of hash encoding, or by means of training an encoding model, which is not limited in the present disclosure.

[0049] Step 102, based on the initial positioning information, determining the candidate map image from the map image library.

[0050] Among them, the candidate map image obtained from the map image library is a map image adjacent to the current road image in space, which is used for feature matching with the current road image. The number of candidate map images is much smaller than the number of map images contained in the map image library. Therefore, the present scheme selects the candidate map image to match with the current road image, which narrows down the matching range and greatly improves the matching efficiency and reduces the consumption of computing resources.

[0051] As an example, when determining the candidate map image from the map image library, the spatially adjacent map images can be quickly retrieved from the map image library as the candidate map images according to the determined initial positioning information and the set spatial range threshold, which can be dynamically adjusted according to the environmental complexity, thereby significantly compressing the matching search space and greatly improving the overall operation efficiency while ensuring the effectiveness of the candidate map image.

[0052] As an example, when determining the candidate map image from the map image library, the candidate map image can be selected from the partial map images according to the determined initial positioning information, and then selected from the partial map images by combining the feature similarity screening, the scene category matching-based screening and other strategies. These strategies can reduce the omission of candidate images and improve the robustness of the screening result in a dynamic environment.

[0053] It should be noted that in the present embodiment, the map image library is pre-constructed, and the map image library stores map images, each of which corresponds to a unique standard pose. Alternatively, the map image library can also directly store the map features corresponding to the map images. In this case, there is no need to extract the features of the candidate map image during repositioning, thereby saving the computing resources.

[0054] In step 103, feature extraction is performed on the current road condition image and the candidate map image respectively to obtain road condition features corresponding to the current road condition image and map features corresponding to the candidate map image.

[0055] In the present embodiment, for the obtained current road condition image and candidate map image, feature extraction is respectively performed on each image to obtain the features corresponding to the current road condition image (referred to as road condition features for ease of description and distinction) and the features corresponding to the candidate map image (referred to as map features for ease of description and distinction). The road condition features and the map features each contain a plurality of feature points, and each feature point includes corresponding feature information and position information.

[0056] It should be noted that when performing feature extraction on the current road condition image and the candidate map image, the currently commonly used feature extraction algorithm can be used to extract the image features, such as the SIFT algorithm. Alternatively, a plurality of different feature extraction algorithms can be integrated in the vehicle, and different feature extraction algorithms can be adaptively selected or fused based on the scene light conditions, image texture richness, etc., to achieve the best balance between matching robustness and computational real-time performance.

[0057] In step 104, the positioning encoding features, the road condition features and the map features are input into the pre-trained attention feature matching model, and a feature matching result of the road condition features and the map features output by the attention feature matching model is obtained.

[0058] The attention feature matching model is a feature matching model based on an attention mechanism, and the attention feature matching model is obtained through pre-training.

[0059] In this embodiment, a self-attention and cross-attention mechanism can be introduced to construct a deep feature matching network, and a training sample is constructed, the training sample including a sample map feature, a sample road feature, and a sample positioning encoding feature, and the sample map feature and the sample road feature in the training sample are matched and labeled between characteristic points. The deep feature matching network is trained using the labeled training sample, and when the number of iterations reaches a threshold or the model converges, the training is completed, and a trained attention feature matching model is obtained. The trained attention feature matching model can output the matching relationship between the feature points of the input two frames of features, that is, output the matched feature point pairs, and each feature point pair includes one feature point on each of the two input frames of features. The self-attention mechanism can focus on the significant feature regions of the road feature and the map feature, and the cross-attention mechanism can establish semantic association between the two frames of features. Through semantic enhancement and significant region modeling of local features, the matching accuracy is significantly improved in scenes with repetitive textures, changes in illumination, or occlusions.

[0060] In this embodiment, the positioning encoding feature, the road feature, and the map feature are input into the pre-trained attention feature matching model for feature extraction. The attention feature matching model associates the positioning encoding feature, the road feature, and the map feature through the attention mechanism, matches the road feature and the map feature under the constraint of the positioning encoding feature, and outputs a feature matching result. The feature matching result is the matching relationship between the feature points of the road feature and the map feature, including a plurality of matched feature point pairs, for example, a feature point A in the road feature matches a feature point B in the map feature, and the feature matching result includes the association relationship between the feature point and the matched feature point B. In the attention network, not only the self-attention and cross-attention between image features are performed, but also the positioning encoding feature of the initial positioning information and the image feature point encoding are spliced to realize cross-modal interaction, guide the feature matching to focus on the region consistent with the prior distribution (i.e., the initial positioning information), and make the feature matching more accurate, especially in repetitive structure, illumination change, and weak texture scenes, thereby significantly improving the robustness of the feature matching result.

[0061] As an example, the network structure of the attention feature matching model is as follows: Figure 2As shown, the attention feature matching model includes a feature point encoding module, an attention training network, and a feature point matching optimization module. The feature point encoding module is configured to encode position information in the input positioning encoding features and map features to output map position encoding information, and to encode position information in the input positioning encoding features and road condition features to output road condition position encoding information. The map position encoding information is spliced with feature information in the map features to obtain map feature information corresponding to the map features, and the road condition position encoding information is spliced with feature information in the road condition features to obtain road condition feature information corresponding to the road condition features. Then, the map feature information and the road condition feature information are input into the attention training network for key feature extraction to obtain a map feature point descriptor and a road condition feature point descriptor. Then, the map feature point descriptor and the road condition feature point descriptor are input into the feature point matching optimization module for matching to output a feature point matching relationship. By using the feature point encoding module to encode the position information in the positioning encoding features and the image features, the interaction of multi-modal information is realized, the feature matching is guided to focus on the region consistent with the prior distribution (i.e., the initial positioning information), and the feature matching is more accurate.

[0062] In this embodiment, the attention feature matching model is trained by introducing the self-attention and cross-attention mechanisms. The combination of the self-attention and cross-attention mechanisms can accurately capture the correlation between features while reducing the computational complexity, and is suitable for the resource-limited vehicle-mounted environment.

[0063] In step 105, the current positioning result of the vehicle is determined based on the feature matching result.

[0064] In this embodiment, after obtaining the feature matching result between the road condition features and each frame of map features, the current positioning result of the vehicle can be obtained by repositioning based on the obtained feature matching result.

[0065] As an example, the map feature with the most matched feature point pairs can be selected according to the feature matching result between the road condition features and each frame of map features, and the feature point pairs matched with the map feature are used for repositioning to obtain the current positioning result of the vehicle. For example, the selected feature point pairs can be first removed by using a random sample consensus (RANSAC) algorithm, and then the relative pose between the two frames is calculated by using a perspective N-point positioning (PnP) geometric algorithm. Then, according to the relative pose and the standard pose corresponding to the candidate map image where the selected feature point is located, the final accurate and stable repositioning pose can be determined, i.e., the current positioning result of the vehicle is obtained.

[0066] It should be noted that in the embodiment, the specific implementation of matching the feature point pairs based on the road condition features and the map features to reposition to determine the vehicle position belongs to the commonly used visual repositioning means, which is not the main improvement point of the present disclosure, and therefore the specific implementation of this part will not be described herein.

[0067] The intelligent driving repositioning method provided by the embodiments of the present disclosure can improve the robustness of the repositioning accuracy by obtaining the current road condition image, the initial positioning information matched with the timestamp of the current road condition image, and the positioning encoding features corresponding to the initial positioning information, inputting the positioning encoding features, the image features of the current road condition image, and the image features of the corresponding candidate map image into the attention feature matching model for feature matching, guiding the attention feature matching model to focus on the region consistent with the initial positioning information, realizing the deep interaction of the image features and the initial positioning information constraint, and improving the accuracy of the feature matching, so that when the current positioning result of the vehicle is determined based on the feature matching result output by the attention feature matching model, the robustness of the repositioning accuracy can be improved based on the accurate feature matching result, and the efficiency of the repositioning can be improved by determining the candidate map image from the map image library based on the initial positioning information and performing feature matching between the candidate map image and the current road condition image.

[0068] In the related art, when determining the map frame matched with the current frame, the search range is determined with the position of the coarse positioning as the center and a preset length as the radius, and the map images in the search range are selected as the candidates to be matched with the current frame. The search radius of this selection method is fixed, and a large area in the non-vehicle driving direction also needs to be searched, which affects the search efficiency. To solve this problem, in the scheme of the present disclosure, the pose covariance matrix is obtained when coarse positioning is performed, and the search range is determined based on the pose covariance matrix, so that the search range is dynamically changed, and the efficiency is higher than that of the traditional fixed-radius search method. The pose covariance matrix includes the covariance of the position of the vehicle and the covariance of the attitude, which are used to represent the errors of the position and the attitude. The covariance of the position includes the covariances corresponding to the X-axis, Y-axis and Z-axis directions, respectively. The covariance of the attitude includes the covariances corresponding to the pitch angle, roll angle and heading angle, respectively. Thus, in an optional embodiment of the present disclosure, the initial positioning information includes the position of the vehicle, the attitude of the vehicle and the pose covariance matrix, and the attitude of the vehicle at least includes the heading angle of the vehicle, as shown in Figure 3 On the basis of the foregoing embodiment, step 102 can include the following sub-steps: Step 201, determining an initial search range based on the position of the vehicle and the pose covariance matrix.

[0069] When the initial search range is determined based on the position of the vehicle and the pose covariance matrix, the X-axis direction covariance and the Y-axis direction covariance in the pose covariance matrix are mainly used for the determination.

[0070] As an example, a search distance threshold can be preset, and the X-axis direction covariance and the Y-axis direction covariance in the pose covariance matrix are multiplied by the search distance threshold respectively to obtain the search distances corresponding to the X-axis direction and the Y-axis direction respectively, and then the initial search range is determined with the vehicle position as the center and the search distances corresponding to the X-axis direction and the Y-axis direction as the search radii in the corresponding directions. It can be understood that when the search distances corresponding to the X-axis direction and the Y-axis direction are equal, the determined initial search range is a circle with the vehicle position as the center and the search radius as the radius; when they are not equal, the determined initial search range is an ellipse, the center of which is the vehicle position, the major axis of which is the larger one of the search distances corresponding to the X-axis direction and the Y-axis direction, and the minor axis of which is the smaller one.

[0071] As an example, the pose covariance matrix can be eigen-decomposed to obtain the X-axis direction offset distance and the Y-axis direction offset distance. In this case, a common eigen-decomposition algorithm can be used to eigen-decompose the pose covariance matrix to obtain an eigenvalue and an eigenvector, and the element value corresponding to the X-axis in the eigenvector is taken as the X-axis direction offset distance, and the element value corresponding to the Y-axis in the eigenvector is taken as the Y-axis direction offset distance. Then, the initial search range is determined based on the X-axis direction offset distance and the Y-axis direction offset distance with the vehicle position as the center. It can be understood that when the X-axis direction offset distance and the Y-axis direction offset distance are not equal, the determined initial search range is an ellipse. In this way, the X-axis direction offset distance and the Y-axis direction offset distance are determined by eigen-decomposing the pose covariance matrix, and then the initial search range is determined in combination with the vehicle position, so that the search range is dynamically constrained, the search range is smaller in the direction with smaller offset distance, and the search efficiency is improved compared with the method of determining the search range with a fixed radius.

[0072] In step 202, the initial search range is adjusted based on the heading angle in the vehicle pose to obtain a target search range.

[0073] It can be understood that the heading angle of the vehicle represents the driving direction of the vehicle. In this embodiment, after the initial search range is determined, the initial search range is further adjusted based on the heading angle in the vehicle pose, and the initial search range is rotated according to the heading angle of the vehicle to obtain the target search range, so that the main direction (i.e. the direction of the major axis of the ellipse) of the target search range is adjusted to be consistent with the heading angle of the vehicle, and a larger search range in the driving direction of the vehicle is ensured to ensure the relevance of the candidate map image and the current road image.

[0074] In step 203, the map image library is searched based on the target search range to obtain a candidate map image.

[0075] In this embodiment, after the target search range is determined, the map image in the map image library within the target search range can be searched based on the target search range, and the map image in the map image library with a standard position within the target search range is obtained as a candidate map image.

[0076] The intelligent driving repositioning method provided in the embodiments of the present disclosure can obtain the vehicle position, the vehicle attitude, and the pose covariance matrix during coarse positioning, determine the initial search range based on the vehicle position and the pose covariance matrix, dynamically constrain the search range based on the pose covariance matrix, improve the search efficiency, adjust the determined initial search range based on the heading angle of the vehicle, make the adjusted target search range consistent with the driving direction of the vehicle, and thus ensure a large search range in the driving direction of the vehicle, thereby improving the robustness of the candidate map image searched.

[0077] In this embodiment, the determined candidate map image can be one or multiple, each candidate map image corresponds to one map feature, and each map feature and the road condition feature both include multiple feature points. The feature matching result includes the feature point pairs matched by each map feature and the road condition feature. When repositioning is performed based on the feature matching result, one map feature can be selected from the map features for repositioning according to the number of feature point pairs matched by each map feature and the road condition feature, so as to avoid the consumption of computing resources caused by using all feature point pairs matched by the map features and the road condition feature for repositioning. Thus, in an optional embodiment of the present disclosure, when the current positioning result of the vehicle is determined based on the feature matching result, the number of point pairs of the feature point pairs matched by each map feature and the road condition feature in the feature matching result can be obtained first, and based on the number of point pairs, the map feature corresponding to the maximum number of point pairs is determined as the target map feature, that is, the map feature with the most successfully matched feature points is selected as the target map feature, and then the target feature point pairs matched by the target map feature and the road condition feature are used to determine the current positioning result of the vehicle. For example, assuming that the number of point pairs of the feature point pairs matched by the map feature 1 and the road condition feature in the feature matching result is Q1, the number of point pairs of the feature point pairs matched by the map feature 2 and the road condition feature is Q2, and the number of point pairs of the feature point pairs matched by the map feature 3 and the road condition feature is Q3, Q1>Q3>Q2, the map feature 1 is determined as the target map feature, and the target feature point pairs matched by the map feature 1 and the road condition feature are used to determine the current positioning result of the vehicle.

[0078] In the embodiments of the present disclosure, by acquiring the number of point pairs of the feature point pairs matched by each map feature in the feature matching result and the road feature, the map feature with the largest number of point pairs is selected as a target map feature, and the feature point pairs matched by the target map feature and the road feature are used for repositioning. In this way, the feature point pairs with high confidence can be screened out for repositioning of the vehicle position, and the accuracy of the repositioning result is improved.

[0079] In order to ensure the confidence of the repositioning result and improve the positioning accuracy, in an optional embodiment of the present disclosure, the positioning result calculated according to the feature matching result can be first subjected to confidence checking, and if the confidence checking is passed, the positioning result is determined as the final current positioning result, thereby ensuring the reliability of the finally determined positioning result. Therefore, as shown in FIG. 5, on the basis of the foregoing embodiments, step 105 can include the following sub-steps: Figure 4 Step 301: determining a candidate positioning result of the vehicle based on the first feature point pair in the feature matching result.

[0080] The first feature point pair is a feature point pair matched by a map feature point in the same map feature and a road feature point in the road feature.

[0081] In the present embodiment, the candidate positioning result of the vehicle is determined based on the feature point pair (referred to as the first feature point pair for the sake of description and distinction) matched by the same map feature and the road feature. When selecting the map feature used to determine the candidate positioning result of the vehicle, one map feature can be randomly selected from at least one map feature, or the map feature with the largest number of feature point pairs matched by the road feature can be selected according to the number of feature point pairs matched by the road feature on each map feature, and used to determine the candidate positioning result of the vehicle.

[0082] It should be noted that in the present embodiment, the specific implementation of determining the candidate positioning result of the vehicle based on the first feature point pair in the feature matching result can refer to the implementation of repositioning the vehicle position by matching the feature point pairs in the commonly used visual repositioning technology, and the present disclosure will not be described herein.

[0083] Step 302: performing confidence checking based on the candidate positioning result and the first feature point pair, and determining a first confidence of the candidate positioning result.

[0084] In the present embodiment, in order to ensure the accuracy of the final positioning result, the confidence of the obtained candidate positioning result can be checked in combination with the first feature point pair.

[0085] ​Exemplarily, the coordinates of each road feature point in the first feature point pair can be acquired, the coordinates of each road feature point are geometrically re-projected with the coordinates in the candidate positioning result to obtain new coordinates corresponding to each road feature point, then for each road feature point, the obtained new coordinates are compared with the coordinates of the matched map feature point, if consistent, the number of checked feature point pairs is accumulated by 1 to obtain the total number of checked feature point pairs in the first feature point pair, and the ratio of the total number of checked feature point pairs to the total number of point pairs contained in the first feature point pair is determined as the confidence of the candidate positioning result (for the convenience of description and distinction, referred to as the first confidence), wherein the initial value of the number of checked feature point pairs is 0.

[0086] It should be noted that in the present embodiment, the coordinate conversion by geometric re-projection is a relatively mature technology in the current SLAM technology, and the present disclosure does not repeat it.

[0087] Step 303, in the case where the first confidence is greater than or equal to the first threshold, determining the current positioning result of the vehicle based on the candidate positioning result.

[0088] The specific value of the first threshold can be set according to actual needs, for example, the first threshold is set to 0.8.

[0089] In the present embodiment, after obtaining the first confidence of the candidate positioning result, the first confidence can be compared with the first threshold, if the first confidence is greater than or equal to the first threshold, it indicates that the determined candidate positioning result is reliable, and the current positioning result of the vehicle can be determined based on the candidate positioning result, for example, the candidate positioning result can be directly determined as the current positioning result of the vehicle.

[0090] Step 304, in the case where the confidence is less than the first threshold, determining the second feature point pair from the feature matching result, the number of the second feature point pair is less than the number of the first feature point pair.

[0091] The second feature point pair is a feature point pair matched between a map feature point in the same map feature and a road feature point in the road feature.

[0092] In the present embodiment, the first confidence corresponding to the candidate positioning result is compared with the first threshold, if the first confidence is less than the first threshold, the second feature point pair is determined from the feature matching result, the second feature point pair is not completely the same as the first feature point pair, and the acquired map feature point in the second feature point pair should belong to the same map feature. The number of the selected second feature point pair can be less than the number of the first feature point pair to achieve lightweight re-projection matching verification.

[0093] At step 305, the confidence of the candidate positioning result is verified based on the second feature point pair, and the second confidence of the candidate positioning result is determined.

[0094] In this embodiment, after the second feature point pair is determined, the confidence of the candidate positioning result is further verified based on the second feature point pair, and the second confidence corresponding to the candidate positioning result is obtained.

[0095] For example, the coordinates of each road condition feature point in the second feature point pair can be obtained, the coordinates of each road condition feature point are geometrically reprojected with the coordinates in the candidate positioning result, the new coordinates corresponding to each road condition feature point are obtained, then for each road condition feature point, the obtained new coordinates are compared with the coordinates of the matched map feature point, if they are consistent, the number of verified feature point pairs is accumulated by 1, the total number of verified feature point pairs in the second feature point pair is obtained, and the ratio of the total number of verified feature point pairs to the total number of point pairs contained in the second feature point pair is determined as the second confidence of the candidate positioning result.

[0096] At step 306, in the case that the second confidence is greater than or equal to the second threshold, the current positioning result of the vehicle is determined based on the candidate positioning result.

[0097] The specific value of the second threshold can be set according to actual needs, and the second threshold can be the same as or different from the first threshold, which is not limited in the present disclosure.

[0098] In this embodiment, after the second confidence of the candidate positioning result is determined, the second confidence can be compared with the second threshold, if the second confidence is greater than or equal to the second threshold, it indicates that the determined candidate positioning result is reliable, and the current positioning result of the vehicle can be determined based on the candidate positioning result, for example, the candidate positioning result can be directly determined as the current positioning result of the vehicle.

[0099] In an optional embodiment of the present disclosure, if the second confidence is less than the second threshold, it indicates that the candidate positioning result is unreliable, in this case, the search range of the map image can be adjusted according to the driving direction of the vehicle, the search range is expanded in the driving direction of the vehicle, and the candidate map image is re-searched from the map image library based on the expanded search range for repositioning. Specifically, in the case that the second confidence is less than the second threshold, the heading angle in the initial positioning information can be obtained, and the first search range used to determine the candidate map image is adjusted based on the obtained heading angle to obtain a second search range, the second search range is greater than the first search range, and then the new candidate map image is determined from the map image library based on the second search range, and the current positioning result of the vehicle is determined based on the new candidate map image and the current road condition image.

[0100] Generally, the heading angle of the vehicle represents the driving direction of the vehicle, and therefore, in this embodiment, the heading angle of the vehicle is obtained from the initial positioning information, and the first search range used for determining the candidate map image last time is adjusted based on the heading angle of the vehicle to obtain a second search range greater than the first search range, so as to expand the search range in the driving direction of the vehicle, and then more new candidate map images can be determined from the map image library based on the expanded second search range, and the new candidate map images and the current road condition image are used for repositioning to obtain the current positioning result of the vehicle. It can be understood that the process of repositioning based on the new candidate map image and the current road condition image to determine the current positioning result of the vehicle is the same as the process of repositioning based on the obtained candidate map image and the current road condition image last time, and the map features of the new candidate map image also need to be extracted for feature matching with the road condition features, and the above repositioning and confidence checking process is repeated, if the confidence checking passes, the repositioning is completed, and the current positioning result of the vehicle is obtained; if the confidence of the re-determined positioning information is still insufficient, the repositioning is expanded to the global. Thus, by expanding the search range of the candidate map image in the driving direction of the vehicle and re-searching new candidate map images for repositioning when the confidence checking fails, the success rate of repositioning can be improved.

[0101] The intelligent driving repositioning method of the embodiments of the present disclosure, for the candidate positioning result determined based on the first feature point pair in the feature matching result, first checks the confidence of the candidate positioning result by using the first feature point pair, and then determines the current positioning result of the vehicle based on the candidate positioning result after the checking passes, thereby ensuring the reliability of the candidate positioning result, and improving the accuracy of the final positioning result. In addition, when the confidence checking of the candidate positioning result based on the first feature point pair fails, a second feature point pair with a smaller number of feature points than the first feature point pair is further obtained, the confidence of the candidate positioning result is checked based on the second feature point pair, and then the current positioning result of the vehicle is determined based on the candidate positioning result after the checking passes. By using the second feature point pair with fewer feature points for confidence checking, a lightweight checking is realized, the calculation resources are controlled while the accuracy is ensured, and robustness and efficiency are considered.

[0102] In an optional embodiment of the present disclosure, based on the foregoing embodiments, when the current positioning result of the vehicle is determined based on the candidate positioning result, pose fusion can be performed based on the candidate positioning result and the initial positioning information to determine the current positioning result of the vehicle.

[0103] The pose fusion refers to comprehensive processing of the pose information estimated in different ways to obtain more accurate, stable and robust pose results. In the embodiments of the present disclosure, the initial positioning information and the determined candidate positioning result are fused by pose fusion to obtain the final positioning result of the vehicle, i.e., the current positioning result.

[0104] Exemplarily, the candidate positioning result and the initial positioning information can be fused in pose by means of graph optimization (least square method), the initial pose in the initial positioning information is taken as an edge, the accurate positioning in the candidate positioning result is taken as a node, the edge and the node are configured into a graph, and the optimal solution of the graph is solved by means of the least square method, that is, the current positioning result of the vehicle is obtained.

[0105] In the embodiments of the present disclosure, the current positioning result of the vehicle is determined by fusing the candidate positioning result and the initial positioning information in pose, the candidate positioning result can provide a positioning result with higher accuracy, and the initial positioning result can ensure better smoothness, so that the accuracy and smoothness of the final positioning result can be ensured by fusing the two to determine the current positioning result of the vehicle.

[0106] Figure 5 A flowchart of an intelligent driving repositioning method of a specific embodiment of the present disclosure is shown, as shown in Figure 5 After the image frame (i.e. the current road condition image) is input, feature extraction is performed thereon, and the corresponding prior coarse positioning position and attitude and covariance (i.e. the initial positioning information in the foregoing embodiments) are obtained based on the time stamp of the image frame, the search range is determined based on these information, and the map candidate frame is searched from the map image library, wherein the map features of the map images are stored in the map image library, that is, the searched map candidate frame is the map features. Then, the map features and the features extracted from the image frame are matched in feature points to obtain the matched feature point pairs, and then the accurate pose estimation is performed based on the matched feature point pairs, and the confidence check is performed on the estimated accurate pose, if the check passes, the initial positioning pose is fused and optimized to obtain the final accurate positioning result (i.e. the current positioning result of the vehicle); if the check fails, the prior pose (i.e. the candidate positioning result in the foregoing embodiments) solved by the feature point pairs is combined to perform a light feature point compensation matching check, if the check passes, the initial positioning pose is fused and optimized to obtain the final accurate positioning result, and if the check still fails, the search range is expanded in combination with the vehicle motion direction to search for the map candidate frame again to start a new round of feature point matching.

[0107] In order to realize the above-mentioned embodiments, the present disclosure further provides an intelligent driving repositioning device.

[0108] Figure 6 The structure diagram of the intelligent driving repositioning device provided by an embodiment of the present disclosure is shown, the device is realized in the form of software and / or hardware, and can be integrated in the vehicle.

[0109] As Figure 6As shown, the intelligent driving repositioning apparatus 40 can include a first acquisition module 410, a second acquisition module 420, a feature extraction module 430, a third acquisition module 440, and a positioning module 450.

[0110] The first acquisition module 410 is configured to acquire a current road condition image collected by a vehicle, initial positioning information matched with a timestamp of the current road condition image, and positioning code features corresponding to the initial positioning information. The second acquisition module 420 is configured to determine a candidate map image from a map image library based on the initial positioning information. The feature extraction module 430 is configured to perform feature extraction on the current road condition image and the candidate map image respectively to obtain road condition features corresponding to the current road condition image and map features corresponding to the candidate map image. The third acquisition module 440 is configured to input the positioning code features, the road condition features, and the map features into a pre-trained attention feature matching model, and acquire a feature matching result of the road condition features and the map features output by the attention feature matching model. The positioning module 450 is configured to determine a current positioning result of the vehicle based on the feature matching result.

[0111] Optionally, the initial positioning information includes a vehicle position, a vehicle attitude, and a pose covariance matrix. The second acquisition module 420 includes: A range determination unit configured to determine an initial search range based on the vehicle position and the pose covariance matrix. An adjustment unit configured to adjust the initial search range based on a heading angle in the vehicle attitude to obtain a target search range. A map determination unit configured to search the map image library based on the target search range to obtain the candidate map image.

[0112] Further optionally, the range determination unit is further configured to: perform feature decomposition on the pose covariance matrix to obtain an X-axis direction offset distance and a Y-axis direction offset distance. determine the initial search range based on the X-axis direction offset distance and the Y-axis direction offset distance with the vehicle position as the center.

[0113] Optionally, the candidate map image is at least one, each of the candidate map images corresponds to a map feature, each of the map features and the road condition features includes a plurality of feature points, and the feature matching result includes a plurality of feature point pairs each of which is matched by each of the map features and the road condition features. The positioning module 450 is further configured to: acquire a point pair number of the feature point pairs each of which is matched by each of the map features and the road condition features in the feature matching result. Determine, based on the number of point pairs, a maximum value of the number of point pairs from the map features as a target map feature. Determine, based on the target feature point pairs matched between the target map feature and the road feature, a current positioning result of the vehicle.

[0114] Optionally, the positioning module 450 comprises: A first positioning unit, configured to determine, based on the first feature point pairs in the feature matching result, a candidate positioning result of the vehicle. A first verification unit, configured to perform confidence verification on the candidate positioning result based on the candidate positioning result and the first feature point pairs, to determine a first confidence of the candidate positioning result. A second positioning unit, configured to, in a case where the first confidence is greater than or equal to a first threshold, determine, based on the candidate positioning result, a current positioning result of the vehicle.

[0115] Further optionally, the positioning module 450 further comprises: A point pair determination unit, configured to, in a case where the confidence is less than the first threshold, determine, from the feature matching result, second feature point pairs, the number of the second feature point pairs being less than the number of the first feature point pairs. A second verification unit, configured to perform confidence verification on the candidate positioning result based on the second feature point pairs, to determine a second confidence of the candidate positioning result. The second positioning unit is further configured to, in a case where the second confidence is greater than or equal to a second threshold, determine, based on the candidate positioning result, the current positioning result of the vehicle.

[0116] Optionally, the positioning module 450 further comprises: A direction determination unit, configured to, in a case where the second confidence is less than the second threshold, obtain a heading angle in the initial positioning information. A range expansion unit, configured to adjust, based on the heading angle, a first search range used to determine the candidate map image, to obtain a second search range, the second search range being greater than the first search range. A map searching unit, configured to determine, based on the second search range, a new candidate map image from a map image library, the new candidate map image being used to determine, in combination with the current road image, the current positioning result of the vehicle.

[0117] Further optionally, the second positioning unit is further configured to: Perform pose fusion based on the candidate positioning result and the initial positioning information, to determine the current positioning result of the vehicle. The intelligent driving repositioning device for a vehicle provided in the embodiments of the present disclosure can execute the intelligent driving repositioning method provided in the embodiments of the present disclosure, and has the corresponding function modules and beneficial effects of the execution method. The contents not described in detail in the device embodiments of the present disclosure can be referred to the description in any method embodiment of the present disclosure.

[0118] The embodiment of the present disclosure further provides a vehicle, comprising a processing module; a storage module for storing executable instructions of the processing module; The processing module is used for reading the executable instructions from the storage module and executing the executable instructions to realize the intelligent driving relocation method provided by any embodiment of the present disclosure, so as to achieve the same effect as the above-mentioned realization method.

[0119] In the case of using an integrated unit, the vehicle can comprise a processing module and a storage module. The processing module can be used to control and manage the actions of the vehicle. The storage module can be used to support the vehicle to execute mutual program codes and data.

[0120] The processing module can be a processor or a controller, which can realize or execute various exemplary logical blocks, modules and circuits described in combination with the present disclosure. The processor can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of digital signal processing (DSP) and microprocessors, etc. The storage module can be a memory.

[0121] The embodiment of the present disclosure further provides a computer program product, comprising computer programs / instructions, which, when executed by a processor, cause the processor to execute the above-mentioned related steps to realize the intelligent driving relocation method provided by any embodiment of the present disclosure.

[0122] According to one or more embodiments of the present disclosure, the present disclosure provides an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; The processor is used to read the executable instructions from the memory and execute the executable instructions to realize the intelligent driving relocation method provided by any embodiment of the present disclosure.

[0123] According to one or more embodiments of the present disclosure, the present disclosure provides a computer readable storage medium, which stores a computer program, and the computer program is used to realize the intelligent driving relocation method provided by any embodiment of the present disclosure.

[0124] It should be noted that the computer-readable medium in the above disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus. In the present disclosure, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take many forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or apparatus. The program code contained in the computer-readable medium can be transmitted by any suitable medium, including but not limited to a wire, a cable, an RF (radio frequency) or the like, or any suitable combination of the above.

[0125] The above computer-readable medium can be contained in the above electronic device; or can exist separately and not be assembled into the electronic device.

[0126] The beneficial effects of the above embodiments can refer to the beneficial effects of the corresponding methods provided above, which will not be repeated here.

[0127] The computer program product of the first aspect can include one or more non-transitory computer-readable media storing instructions that, when executed, cause one or more processors to perform the operations of the method of the first aspect. The computer program product of the first aspect can include a computer-readable medium storing instructions that, when executed, cause one or more processors to perform the operations of the method of the first aspect.

[0128] The units described in the embodiments of the present disclosure can be implemented by software, or by hardware, or by a combination of software and hardware. In some cases, the names of the units do not constitute a limitation on the units themselves.

[0129] The functions described in this description above can be implemented in hardware, software, or any combination thereof. If implemented in software, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media include both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media can be any available media that can be accessed by a general purpose or special purpose computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code means in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Also, functional

[0130] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0131] The above description merely provides an overview of the exemplary embodiments and merely describes techniques in accordance with the principles of the disclosure. It is therefore not exclusive to the above technical features of the disclosed range of the present disclosure, and should also encompass other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with each other to form a technical solution with similar functions disclosed in the present disclosure (but not limited to).

[0132] Furthermore, although operations are depicted in a particular, sequential order, this should not be understood as requiring or implying that the operations are performed in the order illustrated, or that they are performed sequentially, as some operations can not be performed in this order and / or concurrently. Similarly, while certain implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the disclosure. Some features described in the context of separate embodiments can also be implemented together in a single embodiment. Conversely, various features described in the context of a single embodiment can also be implemented in multiple embodiments, separately or in any suitable sub-combination.

[0133] It should also be noted that, in the present disclosure, the terms such as first and second are merely used to distinguish one entity or operation from another, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. In addition, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0134] Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely exemplary forms of implementing the claims. The present disclosure can have various modifications and alterations, and it is understood that any modifications, equivalent replacements, improvements, etc. within the spirit and principle of the present disclosure should be included in the scope of the claims of the present disclosure.

Claims

1. A method for intelligent driving relocation, characterized in that, The method includes: Acquire the current road condition image collected by the vehicle, as well as the initial positioning information matching the timestamp of the current road condition image and the positioning coding feature corresponding to the initial positioning information; Based on the initial positioning information, candidate map images are determined from the map image library; Feature extraction is performed on the current road condition image and the candidate map image respectively to obtain the road condition features corresponding to the current road condition image and the map features corresponding to the candidate map image; The location coding features, the road condition features, and the map features are input into a pre-trained attention feature matching model, and the feature matching results of the road condition features and the map features output by the attention feature matching model are obtained. The current location of the vehicle is determined based on the feature matching results.

2. The intelligent driving repositioning method according to claim 1, characterized in that, The initial positioning information includes vehicle position, vehicle attitude, and pose covariance matrix; The step of determining candidate map images from the map image library based on the initial positioning information includes: Based on the vehicle position and the pose covariance matrix, the initial search range is determined; Based on the heading angle in the vehicle's attitude, the initial search range is adjusted to obtain the target search range; The candidate map images are obtained by searching the map image library based on the target search range.

3. The intelligent driving repositioning method according to claim 2, characterized in that, The determination of the initial search range based on the vehicle position and the pose covariance matrix includes: The pose covariance matrix is ​​subjected to eigenvalue decomposition to obtain the offset distance in the X-axis direction and the offset distance in the Y-axis direction; The initial search range is determined based on the vehicle's location as the center, and the offset distances in the X-axis and Y-axis directions.

4. The intelligent driving relocation method according to claim 1, characterized in that, The candidate map image is at least one, each candidate map image corresponds to a map feature, and each map feature and the road condition feature includes multiple feature points. The feature matching result includes feature point pairs that match each map feature with the road condition feature. Determining the current location of the vehicle based on the feature matching result includes: Obtain the number of feature point pairs that match each map feature with the road condition feature in the feature matching result; Based on the number of point pairs, the map feature corresponding to the maximum number of point pairs among the map features is determined as the target map feature; Based on the target feature point pairs that match the target map features and the road condition features, the current location result of the vehicle is determined.

5. The intelligent driving repositioning method according to claim 1, characterized in that, Determining the current location of the vehicle based on the feature matching result includes: Based on the first feature point pair in the feature matching result, the candidate positioning result of the vehicle is determined; Based on the candidate localization results and the first feature point pair, a confidence level check is performed to determine the first confidence level of the candidate localization results; If the first confidence level is greater than or equal to the first threshold, the current location result of the vehicle is determined based on the candidate location results.

6. The intelligent driving repositioning method according to claim 5, characterized in that, The method further includes: If the confidence level is less than the first threshold, a second feature point pair is determined from the feature matching result, wherein the number of the second feature point pair is less than the number of the first feature point pair. Based on the second feature point pair, the confidence level of the candidate localization result is verified to determine the second confidence level of the candidate localization result; If the second confidence level is greater than or equal to the second threshold, the current location result of the vehicle is determined based on the candidate location results.

7. The intelligent driving relocation method according to claim 6, characterized in that, The method further includes: If the second confidence level is less than the second threshold, obtain the heading angle from the initial positioning information; Based on the heading angle, the first search range used to determine the candidate map image is adjusted to obtain a second search range, which is larger than the first search range. Based on the second search range, new candidate map images are determined from the map image library, and the current location result of the vehicle is determined based on the new candidate map images and the current road condition image.

8. The intelligent driving relocation method according to any one of claims 5-7, characterized in that, Determining the current location of the vehicle based on the candidate location results includes: Based on the candidate positioning results and the initial positioning information, pose fusion is performed to determine the current positioning result of the vehicle.

9. A vehicle, characterized in that, include: Processing module; A storage module for storing executable instructions of the processing module; The processing module is configured to read the executable instructions from the storage module and execute the executable instructions to implement the intelligent driving repositioning method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for implementing the intelligent driving repositioning method according to any one of claims 1-8.

Images