Vehicle control method and device, electronic equipment and storage medium

Abstract

The invention discloses a vehicle control method and device, electronic equipment and a storage medium. The method comprises the following steps: detecting target state information of a target vehicle to indicate whether an obstacle exists in front of the target vehicle when the target vehicle travels according to a target route; in response to the target state information, displaying that an obstacle exists, calling a first recognition model to perform obstacle type recognition on the obstacle in front of the target vehicle; in response to the condition that the confidence of the obstacle type recognition result is not greater than the preset confidence, calling a second recognition model to perform obstacle type recognition on an obstacle in front of the target vehicle; and performing obstacle avoidance control on the target vehicle according to the obstacle type identification result of the obstacle ahead of the target vehicle output by the second identification model. According to the technical scheme, the generalization and robustness of obstacle recognition can be enhanced by using the target route road scene information, so that the obstacle avoidance control scheme is optimized, and the vehicle driving safety and traffic efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, and in particular to a vehicle control method and device, an electronic device, and a storage medium. BACKGROUND

[0002] The fixed route commercial vehicle scene is one of the important fields for the first landing application of automatic driving technology due to the characteristics of relatively fixed driving path and controllable environment complexity. At present, the automatic driving technology scheme for fixed route commercial vehicles mostly directly applies the design idea of general automatic driving system, and often simply relies on a target detection model such as YOLO in the construction of the perception system, ignoring the unique advantages of pre-perception of environmental features and the repeatability of key perception targets in the fixed route scene, so that the design of the perception system cannot be deeply adapted to the characteristics of the scene, and thus the generalization ability and robustness of the existing perception scheme in the actual application process are low, which is difficult to fully meet the high performance requirements of fixed route commercial vehicles for automatic driving perception systems. SUMMARY

[0003] The present application provides a vehicle control method to solve the problems of low utilization rate of scene information, insufficient generalization and robustness of obstacle type identification for fixed route vehicles.

[0004] According to an aspect of the present application, a vehicle control method is provided, which comprises:

[0005] detecting target state information of a target vehicle, the target state information being used to indicate whether there is an obstacle in front of the target vehicle when the target vehicle travels along a target route;

[0006] in response to the target state information showing that there is an obstacle, calling a first identification model to identify the type of the obstacle in front of the target vehicle, the first identification model being obtained by model training based on first type obstacle sample data forming first training data, the first type obstacle being an obstacle with a frequency of occurrence greater than a first preset value in a road scene of the target route;

[0007] in response to the confidence of the obstacle type identification result being not greater than a preset confidence, calling a second identification model to identify the type of the obstacle in front of the target vehicle, the second identification model being obtained by reference zero learning training based on second training data formed by the first type obstacle sample data and second type obstacle sample data, the second type obstacle sample being an obstacle of a type not contained in the first training data, the reference zero learning causing the first type obstacle and the second type obstacle to form an association in the same semantic space, the second type obstacle being an obstacle with a frequency of occurrence less than a second preset value in the road scene of the target route and being constructed by simulation, the first preset value being greater than the second preset value;

[0008] According to the obstacle type recognition result of the obstacle in front of the target vehicle output by the second recognition model, the target vehicle is controlled to avoid the obstacle.

[0009] According to another aspect of the present application, a vehicle control device is provided, which comprises:

[0010] a state information detection module configured to detect target state information of a target vehicle, the target state information being used to indicate whether there is an obstacle in front of the target vehicle when the target vehicle travels along a target route;

[0011] a first recognition module configured to, in response to the target state information indicating that there is an obstacle, call a first recognition model to recognize the obstacle type of the obstacle in front of the target vehicle, the first recognition model being obtained by model training based on first type obstacle sample data and first training data, the first type obstacle being an obstacle with a frequency of occurrence greater than a first preset value in a road scene of the target route;

[0012] a second recognition module configured to, in response to a confidence level of the obstacle type recognition result being not greater than a preset confidence level, call a second recognition model to recognize the obstacle type of the obstacle in front of the target vehicle, the second recognition model being obtained by reference zero-order learning training based on second type obstacle sample data and second training data formed based on the first type obstacle sample data, the second type obstacle sample being an obstacle of a type not contained in the first training data, the reference zero-order learning causing the first type obstacle and the second type obstacle to form an association in a same semantic space, the second type obstacle being an obstacle constructed by simulation and having a frequency of occurrence less than a second preset value in the road scene of the target route, the first preset value being greater than the second preset value;

[0013] an obstacle avoidance control module configured to, according to the obstacle type recognition result of the obstacle in front of the target vehicle output by the second recognition model, control the target vehicle to avoid the obstacle.

[0014] According to another aspect of the present application, an electronic device is provided, which comprises:

[0015] at least one processor; and

[0016] a memory connected to the at least one processor in communication; wherein

[0017] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the vehicle control method according to any one of the embodiments of the present application.

[0018] According to another aspect of the present application, there is provided a computer readable storage medium storing computer instructions for causing a processor to implement the vehicle control method according to any of the embodiments of the present application when executed.

[0019] According to another aspect of the present application, there is provided a computer program product comprising a computer program for implementing the vehicle control method according to any of the embodiments of the present application when executed by a processor.

[0020] The technical solution of the embodiments of the present application can realize early perception of whether there is an obstacle in front of the target vehicle when the target vehicle travels along the target route according to the target state information of the target vehicle, and can improve the forward-looking and response speed of environmental perception. The first identification model obtained by model training based on the first type of obstacle sample data is used to identify the type of obstacle, which can realize accurate type identification of regular obstacles on the target route. If the confidence of the obstacle type identification result of the first identification model is not greater than the preset confidence, the second identification model obtained by reference zero-order learning training based on the first type of obstacle sample data and the second type of obstacle sample data is used to identify the type of obstacle, which can realize type identification of irregular obstacles on the target route. The obstacle avoidance control according to the output result of the second identification model can avoid the driving safety hazards caused by blind braking or turning, can improve the rationality and stability of the obstacle avoidance decision, and can improve the driving safety of the target vehicle. Based on the above technical solution, the road scene information of the target route can be used to improve the accuracy of obstacle identification, and the hierarchical obstacle type identification strategy of the first identification model and the second identification model can solve the problems of obstacle identification lag, high misjudgment rate and slow response, enhance the generalization and robustness of obstacle identification, and thus optimize the obstacle avoidance control scheme to improve the safety and efficiency of vehicle driving.

[0021] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Figure 1is a flow chart of a vehicle control method according to an embodiment of the present application;

[0024] Figure 2 is a flow chart of a vehicle control method according to an embodiment of the present application;

[0025] Figure 3 is a structural schematic diagram of a vehicle control device according to an embodiment of the present application;

[0026] Figure 4 is a structural schematic diagram of an electronic device implementing a vehicle control method according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the personnel in the technical field better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the personnel in the field without creative labor should belong to the protection scope of the present application.

[0028] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0029] Embodiment One

[0030] Figure 1 A flow chart of a vehicle control method according to an embodiment of the present application is provided, the embodiment can be applicable to the case of needing to control the fixed route vehicle to avoid obstacles, the method can be executed by a vehicle control device, the vehicle control device can be realized in the form of hardware and / or software, and the vehicle control device can be configured in any electronic device with network communication function. As shown in the figure, the method comprises: Figure 1

[0031] S110, detecting target state information of the target vehicle, the target state information being used to indicate whether there is an obstacle in front of the target vehicle when the target vehicle travels according to a target route.​

[0032] By continuously monitoring and analyzing the information output by the onboard multi-sensors of the target vehicle, it can instantly determine whether there are obstacles in front of the target vehicle as it travels along a preset target route, providing triggering conditions for subsequent obstacle category recognition operations. Avoiding the invocation of the recognition model when the target vehicle does not perceive an obstacle reduces computational costs and latency. Simultaneously, it ensures that once an obstacle is detected, subsequent obstacle type recognition is immediately triggered, and the target vehicle is controlled to avoid obstacles based on the recognition results, improving the real-time performance and safety of the overall vehicle response.

[0033] For example, the target status information of the target vehicle can be detected by simultaneously processing camera images, LiDAR point clouds and millimeter-wave radar echoes, performing frame-by-frame segmentation and clustering of the area in front of the target route, distinguishing suspected obstacle targets from drivable areas, and generating status information based on spatial occupancy relationships. When any sensor channel detects status information that meets the preset size and distance thresholds, it is determined that an obstacle exists and the target status information is output.

[0034] S120. In response to the target status information showing the presence of an obstacle, the first recognition model is invoked to identify the type of obstacle in front of the target vehicle. The first recognition model is obtained by training the model based on the first training data formed by the first type of obstacle sample data. The first type of obstacle is an obstacle that appears more frequently than a first preset value in the road scene of the target route.

[0035] The first type of obstacle can refer to obstacles that appear more frequently than a first preset value in the target route road scene. The first type of obstacle can at least cover various common obstacles that the target vehicle is likely to encounter while traveling along the target route. For example, the first type of obstacle can include rocks, puddles, pedestrians and vehicles, etc.

[0036] The first recognition model can refer to an obstacle classification model trained through supervised learning using first training data comprised of first-type obstacle sample data. Thanks to the first training data covering a sufficient number of first-type obstacle samples, the first recognition model possesses high recognition accuracy and high processing efficiency for first-type obstacles. For example, the first recognition model can refer to a pre-trained YOLOv8 model capable of recognizing common obstacles.

[0037] Once the target status information confirms the presence of an obstacle, the first recognition model is invoked to determine the type of the obstacle ahead. This enables rapid and accurate identification of common obstacles, reducing computational overhead. Subsequent extended recognition processes are only triggered when the confidence level is insufficient, reducing unnecessary secondary recognition overhead and thus improving the real-time performance of the vehicle control process.

[0038] As an optional but non-limiting implementation solution, in response to the confidence of the obstacle type identification result being greater than a preset confidence, the target vehicle is controlled to avoid obstacles according to the obstacle type identification result of the obstacle in front of the target vehicle that is output by the first identification model.

[0039] The obstacle type identification result output by the first identification model can refer to a type label output by the first identification model after classifying high-frequency common obstacles and a confidence belonging to the output type. The preset confidence can refer to a pre-set obstacle type identification result reliability threshold, which is used to determine whether the obstacle type result output by the first identification model is reliable enough to directly drive the subsequent vehicle obstacle avoidance control. For example, the preset confidence is 0.8.

[0040] When the confidence returned by the first identification model is higher than the preset confidence, the obstacle type indicated by the obstacle type result output by the first identification model is directly taken as the final obstacle type, and the corresponding obstacle avoidance control operation is determined according to the determined obstacle type, and the second identification model is no longer enabled, avoiding additional calculation delay and power consumption caused by secondary identification.

[0041] S130, in response to the confidence of the obstacle type identification result being not greater than a preset confidence, calling a second identification model to identify the obstacle type of the obstacle in front of the target vehicle, the second identification model being obtained by reference zero learning training based on second training data formed by first type obstacle sample data and second type obstacle sample data, the second type obstacle sample being an obstacle of a type not contained in the first training data that is simulated and constructed, the reference zero learning causing the first type obstacle and the second type obstacle to form an association in the same semantic space, the second type obstacle being an obstacle that appears in a road scene of a target route with a frequency less than a second preset value and is simulated and constructed, the first preset value being greater than the second preset value.

[0042] The second type obstacle can refer to a rare type obstacle that appears in a road scene of a target route with a frequency less than a second preset value and is not contained in the first training data. Due to the low frequency, it is difficult to obtain a large amount of second type obstacle sample data by collection, and therefore a small amount of second type obstacle sample data is simulated and constructed by a simulation engine based on physical and geometric constraints, which can be used to participate in establishing a cross-class mapping relationship with the first type obstacle in the reference zero learning training phase. For example, the second type obstacle can include food, abandoned furniture, etc.

[0043] The second training data can refer to a training data set for training of the second identification model, which is composed of the first type obstacle samples and the second type obstacle samples constructed by simulation, wherein the quantity of the first type obstacle samples is greater than the quantity of the second type obstacle samples, and the ratio of the quantity of the first type obstacle samples to the quantity of the second type obstacle samples is greater than or equal to a preset zero-order learning sample ratio threshold.

[0044] The second identification model can refer to an obstacle identification model trained in a reference zero-order learning paradigm, which is used for training of the second training data containing both the first type obstacle samples and the second type obstacle samples obtained by simulation, and establishing a mapping relationship between common obstacle types and rare obstacle types in the same shared semantic space.

[0045] The reference zero-order learning can refer to a training mechanism for establishing a mapping relationship between known high-frequency categories and unknown low-frequency categories in a shared semantic space. Through shared space projection, the model learns to map unknown type attributes to the same space, so that the model can complete identification by means of semantic mapping even if it has not directly learned rare types.

[0046] When the confidence output by the first identification model is lower than a preset confidence, the second identification model is called to identify the obstacle type of the obstacle in front of the target vehicle, so as to make up for the defect that the first identification model covers insufficiently for the second type obstacle. The second identification model has the ability to migrate and infer from common obstacle types to rare obstacle types, so as to extend the range of identifiable obstacle types of the identification model without additional collection of rare obstacle data in real vehicles, reduce the potential safety risk caused by missing rare obstacles, and provide more comprehensive and reliable type basis for subsequent obstacle avoidance decisions.

[0047] As an optional but not limited implementation scheme, the second identification model is generated in the following way, including steps A1-A2:

[0048] Step A1, for each reference obstacle sample data, a multi-dimensional feature vector corresponding to the reference obstacle sample data is extracted from the reference obstacle sample data according to a plurality of feature dimensions, the plurality of feature dimensions including color feature dimension, texture feature dimension, geometric feature dimension and reflection feature dimension, the reference obstacle sample data being obstacle sample data in the first type obstacle sample data and the second type obstacle sample data, and the reference obstacle sample data being data collected by using a vehicle-mounted multi-sensor on the obstacle under different illuminations and angles.

[0049] The reference obstacle sample data can refer to multi-modal raw data and labeling results of the first type obstacle and the second type obstacle collected by using a vehicle-mounted camera, a laser radar and a millimeter wave radar under different illumination conditions and different viewing angles on a target route.

[0050] The feature dimension can refer to a feature component independently characterizing one or more attribute features of the obstacle. The multi-dimensional feature vector corresponding to each reference obstacle sample data can include a color feature dimension, a texture feature dimension, a geometric feature dimension, and a reflection feature dimension.

[0051] By synchronously capturing the appearance, material, shape, and radar scattering characteristics of the obstacle through multiple feature dimensions, different types of obstacles can establish more discriminative inter-class boundaries in the same shared semantic space, while reducing the risk of information loss caused by light or occlusion of a single sensor, thereby providing a stable, comprehensive, and strictly aligned sample representation basis in the same shared semantic space for zero-shot learning.

[0052] For example, the color feature dimension can include the RGB space mean, the HSV space saturation, and the color variance, the texture feature dimension can include the gray level co-occurrence matrix energy, the gray level co-occurrence matrix entropy, and the gray level co-occurrence matrix contrast; the geometric feature dimension can include the obstacle area, the obstacle point cloud density, the obstacle maximum height, and the obstacle volume; and the reflection feature dimension can include the laser radar reflection intensity mean and the millimeter wave radar scattering cross section value.

[0053] For the color feature dimension, the RGB space mean can be used to describe the color of the obstacle; the HSV space saturation can be used to distinguish between bright color objects and dark color objects, when the HSV space saturation is greater than a preset saturation threshold, the obstacle can be considered to belong to a bright color object, and when the HSV space saturation is not greater than the preset saturation threshold, the obstacle can be considered to belong to a dark color object.

[0054] For the texture feature dimension, the gray level co-occurrence matrix energy can be used to reflect the uniformity of the gray level distribution and the texture roughness of the obstacle, the greater the gray level co-occurrence matrix energy, the more regular the obstacle texture, and the smaller the gray level co-occurrence matrix energy, the more random and complex the obstacle texture; the gray level co-occurrence matrix entropy can be used to reflect the complexity of the obstacle information, high entropy can represent that the obstacle texture contains high noise or presents irregular texture, and low entropy can represent uniform texture; the gray level co-occurrence matrix contrast can reflect the clarity and groove depth of the obstacle texture, and further distinguish the obstacle edge, the greater the contrast, the more clear the obstacle texture and the deeper the groove, and the smaller the contrast, the more blurred the obstacle texture.

[0055] For the geometric feature dimension, each information can be obtained according to the point cloud information obtained by the radar, the obstacle occupied area can be calculated according to the point cloud projection; the obstacle point cloud density can reflect the surface detectability of the object, for example, there is liquid at the water hole, so it presents a high-density point cloud distribution; the maximum height of the obstacle can refer to the absolute height relative to the road surface, which is used to distinguish the height type of the obstacle, for example, when the maximum height of the obstacle is greater than a preset height threshold, the obstacle can be considered as a tall obstacle, and when the maximum height of the obstacle is not greater than the preset height threshold, the obstacle can be considered as a small obstacle, and a plurality of height thresholds can be set, so as to divide the obstacles into several levels.

[0056] For the reflection feature dimension, the mean value of the laser radar reflection intensity can be used to reflect the surface optical reflectivity in the near-infrared band, which is sensitive to the color and gloss of the obstacle; the scattering cross-section value of the millimeter wave radar can be used to represent the equivalent electromagnetic scattering cross-section of the obstacle in the millimeter wave frequency band, which is sensitive to the conductive area and edge curvature of the obstacle.

[0057] Step A2, based on the multi-dimensional feature vectors corresponding to the plurality of reference obstacle sample data and the obstacle type labels corresponding to the reference obstacle sample data, performing zero-order learning training on the to-be-trained identification model to obtain a second identification model, the to-be-trained identification model being constructed based on an ensemble learning model framework.

[0058] Zero-order learning training can refer to a paradigm of cross-class transfer learning in which, in the training stage, the feature vectors and labels of visible categories are relied on, the semantic description and visual features are mapped to the same shared space, and a mapping function from the feature space to the category label is learned and obtained, and in the application stage, the model obtained by training identifies the invisible categories according to the learned mapping function from the feature space to the category label. The to-be-trained identification model can refer to an identification model constructed based on an ensemble learning model framework. For example, the to-be-trained model can refer to an extreme gradient boosting tree.

[0059] The multi-dimensional feature vectors corresponding to the plurality of reference obstacle sample data and the obstacle type labels corresponding to the reference obstacle sample data are input into the to-be-trained identification model as training data, the classifier parameters are optimized under the ensemble learning model framework, the mapping function of the feature space constituted by the plurality of feature vectors and the obstacle type label is established, so that the model gradually realizes the knowledge transfer across the obstacle categories, and finally converges to the second identification model.

[0060] The second recognition model training process does not require a large amount of rare obstacle sample data, and can learn a mapping function from an obstacle feature space to an obstacle type under an ensemble learning model framework by using a large amount of first-type obstacle samples and a small amount of second-type obstacle samples, thereby solving the problem of difficult data collection and labeling of rare obstacles. The second recognition model is based on an ensemble learning model, can effectively process structured features, and has strong fitting capability for complex nonlinear relationships between features. Therefore, the mapping function from the obstacle feature space to the obstacle type learned by the second recognition model has strong generalization capability compared with the first recognition model, so that the model can still maintain a low false positive rate and a high recall rate for rare obstacle types. Moreover, the ensemble learning model has high training efficiency, and the model size is controllable, which is suitable for a vehicle-mounted environment. In addition, the second recognition model has scalability. If a new obstacle type is added later, only the multi-dimensional feature vector examples of the new obstacle type need to be provided, and the second recognition model can be directly used for recognition without the need to retrain the entire model, thereby improving the model deployment efficiency. The multi-dimensional feature vector examples of the new obstacle type need to be consistent with the attribute dimensions, meanings and orders of the multi-dimensional feature vectors in the second recognition model training data.

[0061] As an optional but not limited implementation scheme, calling the second recognition model to identify the obstacle type of the obstacle in front of the target vehicle includes steps B1-B2:

[0062] Step B1, determining a real-time obstacle feature vector corresponding to the obstacle in front of the target vehicle.

[0063] The real-time obstacle feature vector can refer to a feature vector representation obtained by instantaneously synchronously collecting, by a vehicle-mounted multi-sensor, after detecting the obstacle in front, and strictly extracting features according to the same extraction method as the multi-dimensional feature vector in the second recognition model training phase. The real-time obstacle feature vector is isomorphic to the multi-dimensional feature vector in the second recognition model training data, that is, the feature attribute dimensions, meanings and orders are strictly consistent.

[0064] The obstacle in front of the vehicle can refer to an entity object occupying the driving space of the target vehicle existing in a preset perception area in front of the target vehicle when the target vehicle drives according to a preset target route. The preset perception area can refer to the perception range of the vehicle-mounted multi-sensor configured by the target vehicle.

[0065] The real-time obstacle feature vector is synchronously collected and complementarily fused by the vehicle-mounted multi-sensor, and is strictly isomorphic to the second recognition model training data in terms of feature attribute dimensions, meanings and orders, can directly realize semantic space alignment, and can avoid feature vector calibration or post-processing.

[0066] In step B2, the real-time obstacle feature vector is matched with the to-be-matched feature vectors corresponding to different obstacle types by the second identification model, and the obstacle type of the obstacle in front of the target vehicle is determined according to the similarity matching result. The to-be-matched feature vectors corresponding to different obstacle types are obtained by learning and counting the second training data formed by the first type obstacle sample data and the second type obstacle sample data when the second identification model is trained.

[0067] The to-be-matched feature vector can refer to a class prototype feature vector generated for each obstacle type in the same shared semantic space after the second identification model counts the feature distribution of the first type and the second type obstacle sample in the zero-time learning training stage. Each attribute value contained in the to-be-matched feature vector can be obtained from the statistical quantity of the corresponding attribute of the plurality of corresponding obstacle type samples. For example, the average of the attribute values of the plurality of corresponding obstacle type samples can be used, and the result obtained by aggregating the attribute values of the plurality of corresponding obstacle type samples through a deep learning model. The to-be-matched feature vector can summarize the statistical distribution of the corresponding obstacle type in the multi-dimensional feature space, and can be used as a feature vector similarity matching reference.

[0068] The similarity matching can refer to projecting the real-time obstacle feature vector and each to-be-matched feature vector into the same shared space, calculating the closeness between the real-time obstacle feature vector and each to-be-matched feature vector, and selecting the obstacle type with the maximum closeness as the obstacle type of the obstacle in front of the target vehicle. For example, the closeness between the real-time obstacle feature vector and each to-be-matched feature vector can be calculated by calculating the cosine similarity, Euclidean distance or Mahalanobis distance.

[0069] The second identification model matches the real-time obstacle feature vector with the to-be-matched feature vectors corresponding to different obstacle types, obtains the maximum similarity and the obstacle type corresponding to the maximum similarity. If the maximum similarity is greater than a preset similarity threshold, the obstacle type corresponding to the maximum similarity is taken as the obstacle type of the obstacle in front of the target vehicle, and the cross-class identification of common obstacles and rare obstacles is realized. If the maximum similarity is not greater than the preset similarity threshold, it is considered that the obstacle recognition confidence is too low, and the obstacle type is output as an unknown obstacle to ensure the safety of the target vehicle.

[0070] The second recognition model can align the attribute features possessed by each type of obstacle in the same shared semantic space, that is, establish a multi-dimensional feature space, and realize semantic bridging between different types of obstacles by establishing a mapping function between the multi-dimensional feature space and the obstacle type, so that the obstacle type of the obstacle in front of the target vehicle can be determined by similarity matching between the real-time obstacle feature vector and the to-be-matched feature vector corresponding to different obstacle types, cross-class recognition from common obstacles to rare obstacles can be realized, and the generality of obstacle recognition can be improved.

[0071] In S140, an obstacle avoidance control is performed on the target vehicle according to the obstacle type recognition result of the obstacle in front of the target vehicle output by the second recognition model.

[0072] The obstacle type recognition result can refer to the obstacle type output by the second recognition model after similarity matching in the shared semantic space and the confidence degree corresponding to the output obstacle. For obstacle types with high feature vector similarity, it can be considered that they have similar physical attributes, and for obstacles with similar physical attributes, at least they can be broadly classified into the same obstacle type, so the same or similar obstacle avoidance scheme can be used when controlling the vehicle to avoid obstacles. The physical attribute can refer to the physical properties of the obstacle reflected by the multiple attribute features included in the multi-dimensional feature vector of the obstacle.

[0073] The obstacle avoidance control can refer to a control process in which the target vehicle determines the corresponding obstacle avoidance control scheme according to the obstacle type output by the second recognition model, and adjusts the steering, braking and driving force of the vehicle according to the determined obstacle avoidance control scheme to avoid the obstacle. For example, the obstacle avoidance control scheme corresponding to the obstacle type output by the second recognition model can be determined according to the preset reference configuration information, wherein the preset reference configuration information can refer to a mapping table preset to indicate the association between the obstacle type and the vehicle obstacle avoidance scheme.

[0074] According to the obstacle type recognition result of the obstacle in front of the target vehicle output by the second recognition model, the obstacle avoidance control is performed on the target vehicle, which can ensure that even when facing rare obstacles, the vehicle can still take appropriate obstacle avoidance measures based on the classification result, avoid misoperation or excessive caution, and improve the traffic efficiency and driving safety.

[0075] The technical scheme of the embodiment of the present application can perceive in advance whether there is an obstacle in front of the target vehicle when the target vehicle travels according to the target route, can improve the foresight and response speed of environmental perception, the first identification model obtained based on the first type of obstacle sample data is used to identify the type of the obstacle, and the accurate type identification of the regular obstacle on the target route can be realized, if the confidence of the obstacle type identification result of the first identification model is not greater than the preset confidence, the second identification model obtained based on the first type of obstacle sample data and the second type of obstacle sample data is used to identify the type of the obstacle, and the type identification of the irregular obstacle on the target route can be realized, and the obstacle avoidance control is performed according to the output result of the second identification model, the driving safety hazard caused by blind braking or turning can be avoided, the rationality and stability of the obstacle avoidance decision can be improved, and the driving safety of the target vehicle can be improved. Based on the above technical scheme, the road scene information of the target route can be used to improve the accuracy of obstacle identification, and the hierarchical obstacle type identification strategy of the first identification model and the second identification model is used to solve the problems of obstacle identification lag, high misjudgment rate and slow response, to enhance the generalization and robustness of obstacle identification, and to optimize the obstacle avoidance control scheme to improve the safety and efficiency of vehicle driving.

[0076] Embodiment two

[0077] Figure 2 A flowchart of a vehicle control method provided for the second embodiment of the present application, the embodiment further optimizes the process of detecting the target state information of the target vehicle in the above-mentioned embodiment. As shown in the figure, the method comprises the following steps. Figure 2

[0078] S210, obtain the positioning information of the target vehicle, and call the reference road appearance information of the road section where the target vehicle is located according to the positioning information of the target vehicle.

[0079] The positioning information can refer to a data set obtained by fusing and solving a vehicle-mounted global navigation satellite system, an inertial measurement unit and a wheel speed sensor, and can be used to represent the real-time position and attitude of the target vehicle in the earth coordinate system or the local high-precision map. The reference road appearance information can refer to the description of the road geometry, boundary and surface characteristics of the target route in the obstacle-free state.

[0080] ​By associating the real-time positioning information of the target vehicle with the reference road appearance information, the road geometry, boundary and surface feature description of the road section where the target vehicle is located in the obstacle-free state can be obtained in real time. Thus, a priori reference that is strongly related to the real-time geographic position of the target vehicle can be provided for subsequent obstacle detection, which can improve the environmental adaptability and the pertinence and accuracy of subsequent difference calculation.

[0081] As an optional but not limited implementation scheme, the reference road appearance information is generated in the following manner, including steps C1-C2:

[0082] Step C1, obtaining first reference information and second reference information of the target vehicle when the target vehicle is empty driving on the target route, the first reference information including color histogram and texture features extracted from road images of the target route collected under different lighting conditions, and the second reference information including elevation map and normal vector map generated by laser radar point cloud, the elevation map and the normal vector map being used to record the flatness and geometric profile of the road surface of the target route.

[0083] Empty driving can refer to the target vehicle driving along the target route in a state of not carrying passengers or goods, which can reduce the influence of body posture changes on sensor collection, and ensure that the sensor collected data can reflect the intrinsic geometry and appearance of the target route. Empty driving can refer to the target vehicle driving along the target route in a light load state without passengers or goods.

[0084] The first reference information can refer to a color histogram and texture feature set extracted after color space conversion and texture filtering of multiple road images collected by the vehicle-mounted camera of the target vehicle under various lighting conditions such as dawn, noon, dusk and night. The first reference information can be used to describe the visual appearance of the target route road. For example, linear back-projection algorithm can be used to capture texture features in multiple road images.

[0085] The second reference information can refer to the elevation map and normal vector map generated after ground segmentation and gridding of the point cloud information obtained by the laser radar during the same empty driving process. Optionally, the second reference information can refer to the elevation map and normal vector map obtained after mutual correction and fusion of multiple elevation maps and multiple normal vector maps generated during multiple empty driving processes.

[0086] By empty collection to exclude load disturbance, multi-light sampling and point cloud fusion can eliminate accidental errors that may be introduced by a single moment or a single sensor, while providing consistent cross-light, geometric accurate and dimensionally complementary original data basis for subsequent appearance templates, significantly improving the environmental adaptability and long-term stability of the reference road appearance information.

[0087] Step C2, generating reference road appearance information corresponding to each road segment in the target route according to the first reference information and the second reference information.

[0088] The first reference information and the second reference information are spatiotemporally registered and fused under the same spatial coordinate system, and a unified road appearance description covering the entire target route is constructed segment by segment. Through the complementation of the visual and geometric dual modalities, the template contains both color texture information insensitive to illumination changes and elevation normal information sensitive to physical undulations, so that the subsequent difference detection can still maintain stability when facing shadows, reflections or slight water stains, while having high sensitivity when encountering elevation mutations caused by real obstacles, thereby balancing low false negatives and low false positives, and improving the overall reliability and environmental adaptability of the obstacle perception system.

[0089] S220, obtaining target road appearance information in front of the target vehicle during driving of the target vehicle according to the target route.

[0090] The front of the driving can refer to a preset perception area extending in the direction of driving of the target vehicle on the target route based on the real-time position of the target vehicle, wherein the longitudinal range of the preset perception area can be dynamically adjusted according to the vehicle speed and braking demand.

[0091] The target road appearance information can refer to the real-time acquisition of the road geometry, boundary and surface feature description of the road segment in front of the target vehicle during the driving of the target vehicle. The target road appearance information can be obtained by real-time collection of multiple sensors and fusion processing, which can ensure that obstacles or road changes can be quickly captured, providing high-fidelity and low-latency input for difference calculation, to improve the timeliness and reliability of subsequent obstacle detection.

[0092] S230, determining the road appearance difference between the target road appearance information and the reference road appearance information of the road segment where the target vehicle is located, and determining the target state information of the target vehicle according to the road appearance difference.

[0093] The road appearance difference can be used to quantify the degree of inconsistency between the real-time acquired target road appearance and the extracted reference road appearance of the corresponding road segment.

[0094] The target state information can refer to the state identifier generated by comparing the road appearance difference with the preset difference threshold, which is used to indicate whether there is an obstacle in front of the target vehicle during driving of the target vehicle according to the target route. When the road appearance difference is greater than the preset difference threshold, the target state information of the target vehicle indicates that there is an obstacle in front of the target vehicle; when the road appearance difference is not greater than the preset difference threshold, the target state information of the target vehicle indicates that there is no obstacle in front of the target vehicle.

[0095] The road appearance difference degree can be used to realize obstacle detection without pre-training of the obstacle detection model, can reduce the probability of misidentifying road inherent facilities as obstacles, and can improve the perception level of unknown obstacles.

[0096] For example, the road appearance difference degree can be obtained by weighting the color difference degree, the texture difference degree and the geometric difference degree. The color difference degree can be used to calculate the color histogram difference degree between the target road appearance information and the reference road appearance information using the Bhattacharyya coefficient; the texture difference degree can refer to the Euclidean distance between the target road appearance information and the reference road appearance information; and the geometric difference degree can refer to the mean square error of the elevation map between the target road appearance information and the reference road appearance information.

[0097] S240, in response to the target state information indicating that there is an obstacle, calling a first recognition model to identify the type of the obstacle in front of the target vehicle, the first recognition model being obtained by model training based on first-type obstacle sample data forming first training data, the first-type obstacle being an obstacle with a frequency of occurrence greater than a first preset value in a road scene of the target route.

[0098] S250, in response to the confidence of the obstacle type identification result being not greater than a preset confidence, calling a second recognition model to identify the type of the obstacle in front of the target vehicle, the second recognition model being obtained by reference zero-order learning training based on second training data formed by the first-type obstacle sample data and second-type obstacle sample data, the second-type obstacle sample being an obstacle of a type not contained in the first training data, the reference zero-order learning forming an association between the first-type obstacle and the second-type obstacle in the same semantic space, the second-type obstacle being an obstacle with a frequency of occurrence less than a second preset value in the road scene of the target route and being constructed by simulation, the first preset value being greater than the second preset value.

[0099] S260, according to the obstacle type identification result of the obstacle in front of the target vehicle output by the second recognition model, performing obstacle avoidance control on the target vehicle.

[0100] The technical scheme of the embodiment of the present application can quickly find abnormal changes in the road surface in front of the vehicle by positioning the target vehicle and calling the reference road appearance of the corresponding road section as a reference, by the difference between the real-time target road appearance in front of the vehicle and the reference appearance, without relying on a global high-precision map or a complex target detection model, and can reduce the computational load and communication demand, thereby improving the identification efficiency and driving safety of the fixed route vehicle in the face of sudden obstacles; the first identification model obtained by model training based on the first type of obstacle sample data is used to identify the type of the obstacle, which can realize accurate type identification of the regular obstacle on the target route; if the confidence of the obstacle type identification result of the first identification model is not greater than the preset confidence, the second identification model obtained by reference zero-order learning training based on the first type of obstacle sample data and the second type of obstacle sample data is used to identify the type of the obstacle, which can realize type identification of the irregular obstacle on the target route; and obstacle avoidance control is performed according to the output result of the second identification model, which can avoid the driving safety hazards caused by blind braking or turning, improve the rationality and stability of the obstacle avoidance decision, and improve the driving safety of the target vehicle. Based on the above technical scheme, the road scene information of the target route can be used to improve the accuracy of obstacle identification, and the hierarchical obstacle type identification strategy of the first identification model and the second identification model can be used to solve the problems of obstacle identification lag, high misjudgment rate and slow response, enhance the generalization and robustness of obstacle identification, and thus optimize the obstacle avoidance control scheme to improve the safety and efficiency of vehicle driving.

[0101] Embodiment three

[0102] Figure 3 A structural schematic diagram of a vehicle control device provided for the third embodiment of the present application is shown in FIG. 3. As shown in the figure, the device includes: Figure 3

[0103] The state information detection module 310 is configured to detect target state information of the target vehicle, and the target state information is used to indicate whether there is an obstacle in front of the target vehicle when the target vehicle is driving according to the target route.

[0104] The first identification module 320 is configured to, in response to the target state information indicating that there is an obstacle, call a first identification model to identify the type of the obstacle in front of the target vehicle, and the first identification model is obtained by model training based on first type of obstacle sample data forming first training data.

[0105] ​The second identification module 330 is configured to, in response to the confidence degree of the obstacle type identification result being less than a preset confidence degree, call a second identification model to identify the obstacle type of the obstacle in front of the target vehicle, the second identification model being obtained by performing reference zero-order learning training based on second training data formed based on first type obstacle sample data and second type obstacle sample data, the second type obstacle sample being an obstacle of a type not included in the first training data, the reference zero-order learning causing the first type obstacle and the second type obstacle to be associated in a same semantic space, the second type obstacle being an obstacle that appears less frequently than a second preset value in a road scene of the target route and is simulated, the first preset value being greater than the second preset value;

[0106] The obstacle avoidance control module 340 is configured to perform obstacle avoidance control on the target vehicle according to the obstacle type identification result of the obstacle in front of the target vehicle output by the second identification model.

[0107] Optionally, the target state information of the target vehicle is detected, including:

[0108] The positioning information of the target vehicle is acquired, and reference road appearance information of a road segment where the target vehicle is located is called according to the positioning information of the target vehicle;

[0109] The target road appearance information in front of the target vehicle when the target vehicle travels along the target route is acquired;

[0110] The road appearance difference degree between the target road appearance information and the reference road appearance information of the road segment where the target vehicle is located is determined, and the target state information of the target vehicle is determined according to the road appearance difference degree.

[0111] Optionally, the reference road appearance information is generated in the following manner:

[0112] The first reference information and the second reference information of the target vehicle when traveling empty on the target route are acquired, the first reference information including color histograms and texture features extracted from road images of the target route collected under different light conditions, and the second reference information including elevation maps and normal vector maps generated by using a laser radar point cloud, the elevation maps and the normal vector maps being used to record the flatness and geometric profile of a road surface of the target route;

[0113] The reference road appearance information corresponding to each road segment in the target route is generated according to the first reference information and the second reference information.

[0114] Optionally, the device further includes:

[0115] In response to the confidence of the obstacle type identification result being greater than a preset confidence, the target vehicle is controlled to avoid the obstacle in front of the target vehicle according to the obstacle type identification result of the obstacle in front of the target vehicle output by the first identification model.

[0116] On the basis of the above-mentioned embodiments, optionally, the second identification model is generated in the following manner:

[0117] For each reference obstacle sample data, a multi-dimensional feature vector corresponding to the reference obstacle sample data is extracted from the reference obstacle sample data according to a plurality of feature dimensions, the plurality of feature dimensions including color feature dimensions, texture feature dimensions, geometric feature dimensions, and reflection feature dimensions, the reference obstacle sample data being obstacle sample data in first-type obstacle sample data and second-type obstacle sample data, the reference obstacle sample data being collected by using a vehicle-mounted multi-sensor on the obstacle under different illuminations and angles;

[0118] Based on the multi-dimensional feature vectors corresponding to a plurality of reference obstacle sample data and the obstacle type labels corresponding to the reference obstacle sample data, a zero-time learning training is performed on a to-be-trained identification model to obtain the second identification model, the to-be-trained identification model being constructed based on an ensemble learning model framework.

[0119] On the basis of the above-mentioned embodiments, optionally, the second identification model is called to identify the obstacle type of the obstacle in front of the target vehicle, including:

[0120] The real-time obstacle feature vector corresponding to the obstacle in front of the target vehicle is determined.

[0121] The real-time obstacle feature vector is matched with to-be-matched feature vectors corresponding to different obstacle types by the second identification model, and the obstacle type of the obstacle in front of the target vehicle is determined according to the similarity matching result, the to-be-matched feature vectors corresponding to different obstacle types being feature distributions corresponding to different obstacle types learned and counted by the second identification model when the second identification model is trained by using second training data formed by the first-type obstacle sample data and the second-type obstacle sample data.

[0122] The technical scheme of the embodiment of the present application can perceive in advance whether there is an obstacle in front of the target vehicle when the target vehicle travels according to the target route, by detecting the target state information of the target vehicle in real time, and can improve the foresight and response speed of environmental perception; the first identification model obtained by model training based on the first type of obstacle sample data is used to identify the type of the obstacle, so that accurate type identification of the regular obstacle on the target route can be realized; if the confidence of the obstacle type identification result of the first identification model is not greater than the preset confidence, the second identification model obtained by reference zero-order learning training based on the first type of obstacle sample data and the second type of obstacle sample data is used to identify the type of the obstacle, so that type identification of the irregular obstacle on the target route can be realized; and obstacle avoidance control is performed according to the output result of the second identification model, which can avoid the driving safety hazards caused by blind braking or turning, can improve the rationality and stability of the obstacle avoidance decision, and can improve the driving safety of the target vehicle. Based on the above technical scheme, the road scene information of the target route can be used to improve the accuracy of obstacle identification, and the hierarchical obstacle type identification strategy of the first identification model and the second identification model can be used to solve the problems of obstacle identification lag, high misjudgment rate and slow response, to enhance the generalization and robustness of obstacle identification, so as to optimize the obstacle avoidance control scheme and improve the safety and traffic efficiency of vehicle driving.

[0123] The vehicle control device provided in the embodiment of the present application can execute the vehicle control method provided in any of the embodiments of the present application, has the functions and advantages corresponding to the execution of the vehicle control method, and the detailed process is described in the related operation of the vehicle control method in the foregoing embodiments.

[0124] It is worth noting that each unit and module included in the above device is only divided according to the function logic, but is not limited to the above division, as long as the corresponding function can be realized; in addition, the specific name of each functional unit is only for easy mutual distinction, and is not used to limit the protection scope of the embodiment of the present application.

[0125] Embodiment four

[0126] Figure 4A structural diagram of an electronic device 10 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices (e.g., headsets, glasses, watches, etc.), and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present application described and / or claimed in this document.

[0127] As shown, Figure 4 The electronic device 10 includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., connected to the at least one processor 11 in communication, where the memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer programs stored in the read-only memory (ROM) 12 or loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0128] A plurality of components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc., an output unit 17, such as various types of displays, speakers, etc., a storage unit 18, such as a magnetic disk, an optical disk, etc., and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunications networks.

[0129] The processor 11 can be various general and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as the vehicle control method.

[0130] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program in accordance with embodiments of the present application. For example, embodiments of the present application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication unit 19, or installed from the storage unit 18, or installed from the ROM 12. When the computer program is executed by the processor 11, the above-described functions defined in the methods of embodiments of the present application are performed.

[0131] In some embodiments, the vehicle control method can be implemented as a computer program tangibly embodied in a computer readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded onto and / or installed in the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded onto the RAM 13 and executed by the processor 11, one or more steps of the vehicle control method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the vehicle control method by any other suitable means, for example, by means of firmware.

[0132] The various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0133] Computer programs used to implement the methods of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program, when executed by the processor of the machine, implements the functions / acts specified in the flowcharts and / or block diagrams. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0134] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage 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. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0135] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0136] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), blockchain network, and the Internet.

[0137] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service.

[0138] It should be understood that the various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present application can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, which is not limited herein.

[0139] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A vehicle control method, characterized in that, The method includes: The target status information of the target vehicle is detected, and the target status information is used to indicate whether there are obstacles in front of the target vehicle when the target vehicle is traveling along the target route; In response to the target status information indicating the presence of an obstacle, a first recognition model is invoked to identify the type of obstacle in front of the target vehicle. The first recognition model is obtained by training the model based on first training data formed by first type of obstacle sample data. The first type of obstacle is an obstacle that appears more frequently than a first preset value in the road scene of the target route. In response to the obstacle type recognition result having a confidence level not greater than a preset confidence level, a second recognition model is invoked to identify the obstacle type in front of the target vehicle. The second recognition model is obtained by reference zero-time training based on second training data formed by first type obstacle sample data and second type obstacle sample data. The second type obstacle sample is an obstacle whose type is not included in the simulated first training data. The reference zero-time training enables the first type obstacle and the second type obstacle to form an association in the same semantic space. The second type obstacle is an obstacle whose frequency of occurrence in the road scene of the target route is less than a second preset value and is constructed through simulation. The first preset value is greater than the second preset value. Based on the obstacle type identification result of the obstacle in front of the target vehicle output by the second identification model, obstacle avoidance control is performed on the target vehicle.

2. The method according to claim 1, characterized in that, Detect the target vehicle's status information, including: Obtain the location information of the target vehicle, and call the pre-configured reference road appearance information of the road segment where the target vehicle is located based on the location information of the target vehicle; Obtain information about the appearance of the target road ahead of the target vehicle when the target vehicle is traveling along the target route; Determine the road appearance difference between the target road appearance information and the reference road appearance information of the road segment where the target vehicle is located, and determine the target state information of the target vehicle based on the road appearance difference.

3. The method according to claim 2, characterized in that, The reference road appearance information is generated in the following way: The system acquires first and second reference information when the target vehicle is driving empty on the target route. The first reference information includes color histograms and texture features extracted from road images corresponding to the target route collected under different lighting conditions. The second reference information includes elevation maps and normal vector maps generated using lidar point clouds. The elevation maps and normal vector maps are used to record the smoothness and geometric contour of the road surface corresponding to the target route. Based on the first reference information and the second reference information, reference road appearance information corresponding to each road segment in the target route is generated.

4. The method according to claim 1, characterized in that, The method further includes: If the confidence level of the obstacle type identification result is greater than the preset confidence level, obstacle avoidance control is performed on the target vehicle based on the obstacle type identification result of the obstacle in front of the target vehicle output by the first identification model.

5. The method according to claim 1, characterized in that, The second recognition model is generated in the following manner: For each reference obstacle sample data, a multi-dimensional feature vector corresponding to the reference obstacle sample data is extracted from the reference obstacle sample data according to multiple feature dimensions. The multiple feature dimensions include color feature dimension, texture feature dimension, geometric feature dimension and reflection feature dimension. The reference obstacle sample data is obstacle sample data from the first type of obstacle sample data and the second type of obstacle sample data. The reference obstacle sample data is collected from the obstacle using vehicle-mounted multi-sensor data under different lighting and angles. Based on the multidimensional feature vectors corresponding to multiple reference obstacle sample data and the obstacle type labels corresponding to the reference obstacle sample data, the second recognition model is obtained by zero-time training of the recognition model to be trained. The recognition model to be trained is constructed based on the ensemble learning model framework.

6. The method according to claim 1, characterized in that, The second recognition model is invoked to identify the type of obstacle in front of the target vehicle, including: Determine the real-time obstacle feature vector corresponding to the obstacle in front of the target vehicle; The second recognition model performs similarity matching between the real-time obstacle feature vector and the feature vector to be matched corresponding to different obstacle types, and determines the obstacle type of the obstacle in front of the target vehicle based on the similarity matching result. The feature vector to be matched corresponding to different obstacle types is the feature distribution corresponding to different obstacle types learned and statistically obtained by the second recognition model from the second training data formed by the first type obstacle sample data and the second type obstacle sample data during model training.

7. A vehicle control device, characterized in that, The device includes: The status information detection module is used to detect the target status information of the target vehicle. The target status information is used to indicate whether there are obstacles in front of the target vehicle when the target vehicle is traveling along the target route. The first identification module is used to respond to the target status information showing the presence of an obstacle by calling the first identification model to identify the type of obstacle in front of the target vehicle. The first identification model is obtained by training the model based on the first training data formed by the sample data of the first type of obstacle. The first type of obstacle is an obstacle that appears more frequently than a first preset value in the road scene of the target route. The second recognition module is used to call the second recognition model to identify the obstacle type in front of the target vehicle in response to the obstacle type recognition result having a confidence level not greater than a preset confidence level. The second recognition model is obtained by reference zero-time training based on the second training data formed by the first type obstacle sample data and the second type obstacle sample data. The second type obstacle sample is an obstacle whose type is not included in the simulated first training data. The reference zero-time training enables the first type obstacle and the second type obstacle to form an association in the same semantic space. The second type obstacle is an obstacle whose frequency of occurrence in the road scene of the target route is less than a second preset value and is constructed through simulation. The first preset value is greater than the second preset value. The obstacle avoidance control module is used to perform obstacle avoidance control on the target vehicle based on the obstacle type identification result of the obstacle in front of the target vehicle output by the second identification model.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the vehicle control method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the vehicle control method of any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the multipath transmission scheduling method according to any one of claims 1-6.

Images