Driving assistance method, driving assistance device, and computer program product

Abstract

The invention provides a driving assistance method, which comprises the following steps: when a vehicle runs on a first road, acquiring road environment data ahead in real time by means of an environment sensor of the vehicle, the road environment data at least comprising the first road and a second road branched from the first road, the vehicle plan is switched to run on a second road from a first road according to a preset running route; identifying a road congestion event located on the second road by means of a trained machine learning model based on the road environment data, the road congestion event causing the vehicle not to continue to travel along the second road; upon identification of a road jam event, an alternative travel route is planned for the vehicle to avoid the road jam event. The invention further provides a driving assistance device and a computer program product. According to the invention, by using the artificial intelligence technology, the limitation of the traditional sensing distance and visual angle is broken through, the opportunity of congestion identification is greatly advanced, and the passing efficiency and driving experience of the vehicle in the urban branch scene are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a driving assistance method, and also relates to a driving assistance device and a computer program product. BACKGROUND

[0002] With the continuous development of driving assistance technology, the traffic capacity of vehicles in complex urban scenarios has been significantly enhanced. However, in certain scenarios, driving assistance technology still faces many challenges. For example, when a vehicle is driving on a main road and enters a narrow one-way branch or auxiliary road, the front may be blocked by temporary parked vehicles, loading and unloading personnel and other unexpected factors, forcing the vehicle to back up and reselect a route.

[0003] Currently, the prior art has proposed to use traffic road condition information received from the outside to identify the blockage in advance, but this scheme often has delays and errors such as "blockage has been removed but still prompts no entry" or "blockage has occurred but has not been reported". In addition, the prior art also proposes to use vehicle-mounted sensors to detect unexpected road blockages. However, due to the limited sensing distance and field of view of the sensors, it is often only possible to trigger an alarm when approaching the blockage point, at which time it is difficult to re-plan the route.

[0004] In summary, for temporary blockages in urban branch roads or auxiliary roads, existing driving assistance solutions still have obvious deficiencies in identification timeliness and dynamic path adjustment. SUMMARY

[0005] The purpose of the present application is to provide a driving assistance method, a driving assistance device and a computer program product to at least solve some of the problems in the prior art.

[0006] According to a first aspect of the present application, a driving assistance method is provided, the driving assistance method comprising the following steps:

[0007] Step S1, during driving of a vehicle on a first road, real-time acquisition of road environment data in front of the vehicle by means of an environmental sensor of the vehicle, the road environment data at least including the first road and a second road diverging from the first road, the vehicle planning to switch from the first road to the second road for driving;

[0008] Step S2, based on the road environment data, identification of a road blockage event located on the second road by means of a trained machine learning model, the road blockage event causing the vehicle to be unable to continue driving along the second road; and

[0009] Step S3, in the case of identification of the road blockage event, planning of an alternative driving route for the vehicle to avoid the road blockage event.

[0010] The application particularly includes the following technical concept: during a driving stage of a vehicle along a first road, a trained machine learning model is used to determine in real time whether a second road is blocked, so that the vehicle can learn the risk in advance before entering the road section affected by the blockage. By making full use of artificial intelligence technology, the limitations of traditional sensing distance and angle of view are broken through, the timing of blockage recognition is greatly advanced, sufficient time is left for path re-planning, and the number of alternative detour routes is significantly increased. Overall, the embarrassment and time-consuming of being forced to reverse due to late recognition of the blockage point are avoided, and the traffic efficiency and user driving experience in the urban branch / secondary road scene are significantly improved.

[0011] In an example embodiment, the first road is a main road, and the second road is a secondary road or a branch road; or, the first road is a highway, and the second road is an exit ramp.

[0012] In an example embodiment, the driving assistance method further comprises pre-training the machine learning model in the following manner: obtaining an initial training data set, the initial training data set comprising historical road environment data and corresponding road blockage event annotation information under multiple road types and traffic conditions; and pre-training the machine learning model with the initial training data set until the pre-set termination condition is reached to stop training.

[0013] In an example embodiment, the driving assistance method further comprises re-training the machine learning model in the following manner: during actual driving of the vehicle, inputting real-time acquired road environment data into the pre-trained machine learning model to identify a road blockage event located on the second road; for the identified road blockage event, extracting additional road environment data acquired by the vehicle with the environmental sensor while driving on the first road and before entering the second road to construct an incremental training data set; and re-training the pre-trained machine learning model online with the incremental training data set.

[0014] In an example embodiment, the pre-training of the machine learning model is performed in a test stage before the vehicle is shipped; and / or, the re-training of the machine learning model is performed during daily commuting use of the vehicle.

[0015] In an example embodiment, the alternative driving route includes: guiding the vehicle to continue driving along the first road without entering the second road; or, guiding the vehicle to enter the second road and then switch to driving on a third road upstream of the location where the road blockage event occurs.

[0016] In an example embodiment, the real-time acquired road environment data is collected by the environmental sensor of the vehicle within a preset time period before the current time, and the length of the preset time period is fixed at tens of seconds or dynamically adjusted according to the real-time driving speed of the vehicle.

[0017] In one example embodiment, the driving assistance method further comprises the steps of: obtaining traffic condition information via a communication network when the vehicle is driving on the first road, and checking whether the traffic condition information indicates that there is a road congestion event on the second road; verifying the traffic condition information with the recognition result in step S2 by means of the machine learning model; and reporting the recognition result in step S2 to update the traffic condition information if the traffic condition information is inconsistent with the recognition result in step S2.

[0018] According to a second aspect of the present application, a driving assistance apparatus is also provided, which comprises a memory and a processor, the memory storing computer program instructions, when the computer program instructions are executed by the processor, the processor can execute the driving assistance method according to the first aspect of the present application.

[0019] According to a third aspect of the present application, a computer program product is also provided, which comprises computer program instructions, wherein the computer program instructions, when executed by one or more processors, enable the one or more processors to execute the driving assistance method according to the first aspect of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0020] The principles, features and advantages of the present application can be better understood by referring to the following detailed description of the application, taken in conjunction with the accompanying drawings. The drawings include:

[0021] Figure 1 A block diagram of a vehicle comprising a driving assistance apparatus according to one example embodiment of the present application is shown;

[0022] Figure 2 A flowchart of a driving assistance method according to one example embodiment of the present application is shown;

[0023] Figure 3 A flowchart of two additional steps of the driving assistance method shown; Figure 2

[0024] A flowchart of a driving assistance method according to another example embodiment of the present application is shown; Figure 4

[0025] A schematic diagram showing the application of the driving assistance method according to the present application in one example scenario is shown; and Figure 5

[0026] Figure 6 A schematic diagram showing the application of the driving assistance method according to the present application in another example scenario is shown. DETAILED DESCRIPTION

[0027] ​In order to make the technical problems to be solved, technical solutions and beneficial technical effects of the present application clearer, the present application will be further described in detail below in conjunction with the drawings and a plurality of exemplary embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the protection scope of the present application.

[0028] Figure 1 A block diagram of a vehicle 1 according to an exemplary embodiment of the present application is shown, which comprises a driving assistance device 10.

[0029] The vehicle 1 can support partial automatic driving, full automatic driving, and also manual driving mode. The driving assistance device 10 of the vehicle, for example, comprises a processor 12 and a memory 11, and the memory 11 stores computer program instructions, which can be stored in a computer readable storage medium such as a hard disk, a memory, a flash card, etc. The processor 12 can be a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), or other general-purpose processor. When the processor 12 executes the computer program instructions in the memory 11, a driving assistance method for determining the vehicle 1 can be implemented, which will be further described below. The driving assistance device 10 itself can be deployed locally in the vehicle 1 or on a cloud server to adapt to different application scenarios and needs.

[0030] The driving assistance device 10 can be connected to a plurality of sensors and actuators 13, 14, 15, 16, 20 of the vehicle 1 through an in-vehicle network (such as CAN bus, FlexRay, MOST, etc.) in a wired or wireless manner.

[0031] The driving assistance device 10 can be connected to a plurality of environmental sensors of the vehicle 1 to obtain real-time road environment data in front. For example, the driving assistance device 10 can be connected to a camera 13 for capturing images of the road environment in front. In addition, the driving assistance device 10 can also be connected to other environmental sensors such as laser radar sensors, millimeter wave radar sensors and ultrasonic radar sensors to realize multi-modal perception of the road environment. In the driving assistance device 10, for example, trained machine learning models are provided or the driving assistance device 10 can access them through a network, which are trained, for example, to recognize a road blockage event on a second road in advance according to the above-mentioned environmental perception data when the vehicle 1 is still driving on a first road. The model, for example, supports multi-modal input and can fuse different types of perception data to improve the detection accuracy and robustness of the blockage event.

[0032] The driving assistance device 10 can also be connected to a navigation positioning unit 14 of the vehicle 1 to obtain real-time vehicle positioning, read a preset driving route, and extract road shape information from an embedded electronic map. Based on this, the driving assistance device 10 can determine whether the vehicle 1 is about to enter a second road from a first road, thereby triggering the identification of a potential road blockage event on the second road in a timely manner. Once a blockage is detected, the route can also be immediately re-planned in combination with the real-time position of the vehicle and the road shape.

[0033] The driving assistance device 10 can also access a communication network through a communication interface 15 to receive real-time traffic condition information published by an external road supervision platform. These external information can serve as a supplementary data source, which is fused or cross-verified with the perception data obtained by the driving assistance device 10 based on the environmental sensors of the vehicle 1. For example, when the environmental perception data suggests that there may be a road blockage event ahead, the external traffic condition information can further confirm the existence, severity, and specific cause of the road blockage event, and vice versa.

[0034] The driving assistance device 10 can also be connected to a data recorder 16 of the vehicle 1, which continuously saves raw data collected by various environmental sensors, including images, point clouds, radar echoes, and corresponding time stamps and positioning information. When the machine learning model needs to be retrained, the driving assistance device 10 can retrieve additional road environment data saved in the data recorder 16, filter and label them as needed to form an incremental training set, thereby realizing online optimization of the model performance. In other embodiments, the vehicle data recorder 16 can be arranged integrally with the environmental sensors of the vehicle 1, or the environmental sensors themselves can also be equipped with storage devices to meet different design requirements and application scenarios.

[0035] The driving assistance device 10 can also be connected to a driving execution device 20 of the vehicle to be able to perform lateral control and longitudinal control of the vehicle 1. The driving execution device 20 includes, for example, key components such as a powertrain, a transmission system, a steering system, and a braking system. The driving execution device 20 can respond to the driving route or lateral / longitudinal control instructions re-planned by the driving assistance device 10 to perform corresponding vehicle 1 control operations.

[0036] It should be understood that Figure 1 The number and types of sensors or actuators connected to the driving assistance device 10 shown in FIG. 1 are only examples, and this document is not intended to limit them. In actual applications, other types or numbers of sensors and actuators can be used in the vehicle 1 to meet specific needs and conditions.

[0037] It should also be understood that the connection relationship and function division of each module can be adjusted as required without departing from the core idea of the present application, and these adjustments are within the protection scope of the present application. For example, the driving assistance device 10 is not limited to the configuration shown in the figures, and can also be at least partially integrated with the driving execution device 20 and / or the navigation positioning unit 14. Figure 1

[0038] Figure 2 A flowchart of a driving assistance method according to an example embodiment of the present application is shown. The method exemplarily includes steps S1, S2 and S3, and can be implemented, for example, in the case of the driving assistance device 10 shown. Figure 1

[0039] In step S1, during driving of the vehicle on a first road, road environment data in front of the vehicle is acquired in real time by means of an environmental sensor of the vehicle, the road environment data at least including the first road and a second road branching out from the first road, and the vehicle plans to switch from the first road to the second road according to a preset driving route.

[0040] Exemplarily, the first road can be a main road, and the second road can be a branch road or a branch road branching out from the main road, in particular a one-way road. The first road can also be an expressway, and the second road can be a corresponding exit ramp. In addition, the first road and the second road can also be urban secondary roads of the same level, internal roads of a park, or branch roads connected through an intersection.

[0041] The environmental sensor mentioned above includes, for example, a front camera, a millimeter wave radar, a laser radar and an ultrasonic sensor of the vehicle, for acquiring multi-modal information such as images, point clouds and distances of the road environment.

[0042] “Real-time acquisition” refers to data continuously collected by the sensor within a preset time period retroactively from the current time, which can be fixed at tens of seconds (e.g. 30 seconds), or can be dynamically adjusted according to the real-time speed of the vehicle. As the vehicle moves forward, the data window input into the machine learning model is also updated synchronously.

[0043] The “preset driving route” is usually automatically planned by the vehicle-mounted navigation system after acquiring the destination input by the user and taking the current position of the vehicle (or the departure point specified by the user) as the starting point, and comprehensively considering real-time traffic, road level, historical congestion probability and driving preferences. The route can be dynamically updated according to the latest road conditions, temporary restrictions or manual changes by the user during driving, and serves as a reference for subsequent congestion identification and path re-planning.

[0044] In step S2, based on the road environment data, a road congestion event located on the second road is identified by means of a trained machine learning model, the road congestion event causing the vehicle to be unable to continue driving along the second road.​​

[0045] In particular, the recognition process of the machine learning model can be either continuously ongoing or triggered according to certain conditions to optimize the computing power. For example, when the vehicle is stably driving on a multi-lane highway and there is no branch or exit in the short term, the machine learning model can be temporarily not invoked. In contrast, if the vehicle is about to enter an exit ramp or a service road in about 100 meters, the recognition operation of the machine learning model is triggered to evaluate the traffic conditions of the second road in advance. In addition, the recognition process of the machine learning model can also be triggered only when the second road is labeled as a single lane or a physical single lane.

[0046] The road blockage event may, for example, include: complete or partial lane closure caused by vehicle accidents; temporary parking of trucks or express delivery vehicles in the center of the road to load and unload goods; road blockage caused by road construction barriers, cones, signs and working machinery; on-site traffic control by traffic police; traffic control signs; natural disasters such as road collapse; and situations caused by multiple vehicle intersections or failed lane changes. The above situations may all cause the vehicle to be unable to continue driving along the intended route.

[0047] The machine learning model can use a convolutional neural network (CNN), a recurrent neural network (RNN), a graph neural network (GNN), or a Transformer architecture. The model can support single-modal input (such as images) or can fuse multi-modal data (such as images, point clouds, and ultrasonic waves).

[0048] The machine learning model has been trained, for example, to predict whether the second road is blocked with high confidence based on road environment data collected by the vehicle's existing environmental sensors at the current location, even if the vehicle is still driving on the first road and has not yet entered the second road, at least for most cases. This makes up for the fixed limitations of the vehicle's environmental sensors in terms of distance and viewing angle, ensuring that road blockage events can be identified in advance. The specific training process of the model will be further described below. Figure 3

[0049] In one specific embodiment, the trained machine learning model can perform multi-dimensional feature extraction and comprehensive evaluation on the road environment data and output the recognition result of the road blockage event when the evaluation result meets the preset conditions.

[0050] Such multi-dimensional feature extraction may, for example, include spatial feature extraction and semantic feature extraction. Spatial features may, for example, relate to the contour shape, spatial size parameters, and distribution position information of stationary obstacles. Based on these spatial feature parameters, the machine learning model can calculate the remaining passable space on the road. When the remaining passable space is below a vehicle safety passing threshold, or the proportion of obstacles continuously occupying the lane width reaches or exceeds W% (e.g., W≥60) and the duration reaches or exceeds T seconds (e.g., T≥30), the model determines that a road blockage event is recognized.​

[0051] Semantic feature extraction, for example, involves parsing text and symbol information in the image. For example, when a road construction sign or an accident warning sign of a preset category is detected, the model determines that a road blockage event is recognized.

[0052] In step S3, an alternative driving route is planned for the vehicle to avoid the road blockage event in the case where the road blockage event is recognized.

[0053] The alternative driving route, for example, includes guiding the vehicle to continue driving along the first road without entering the second road. In addition, the alternative driving route can also include guiding the vehicle to enter the second road and then switch to driving on the third road upstream of the location where the road blockage event occurs. In addition, other alternative routes can also be flexibly planned according to the timing of recognizing the road blockage event, real-time traffic conditions, detour distance, and user preferences.

[0054] In one embodiment, after the alternative driving route is generated, the user can be notified of the route change information through voice prompts or a car interface, and options for confirming or selecting the alternative route are provided.

[0055] In another embodiment, the user can be explicitly outputted the reason for changing the route, i.e., the recognition of the road blockage event on the preset driving route, so that the user can clearly understand the basis for the change.

[0056] In another embodiment, the type of road blockage event can also be identified by means of a machine learning model, and the time required for the blockage to be removed is estimated. Then, the time is compared with the detour time of the alternative driving route or the waiting time that may be caused by congestion on the detour section, so as to determine whether to wait for the blockage to be removed and then continue driving on the original route, or to directly select the alternative driving route.

[0057] Figure 3 A flowchart showing Figure 2 A flowchart showing two additional steps of the driving assistance method. In this embodiment, Figure 2 The driving assistance method shown also includes two additional steps S01 and S02, which are mainly used to describe the training process of the machine learning model. These two steps can be performed before step S2, or at least partially before step S2.

[0058] In step S01, the machine learning model is pre-trained, which is usually done offline. The pre-training process is mainly based on an initial training dataset to train the model from scratch. The initial training dataset, for example, covers historical road environment data under various road types (including urban trunk roads, highways, rural roads, etc.), different weather conditions, and various traffic conditions (such as peak congestion, off-peak smoothness, accident sites, etc.). These data are collected by roadside sensors or environmental sensors of sampling vehicles, including images, point clouds, and radar data. Each type of data is finely labeled to indicate whether there is a road congestion event, and optionally further indicate the specific type of the event (e.g., vehicle accident, road construction, temporary stop, etc.). The pre-training process will continue until a preset termination condition is reached. The preset termination condition can be that the accuracy of the machine learning model on the validation set reaches a preset threshold, or reaches a preset number of iterations. Usually, the pre-training is mainly completed in the test phase before the vehicle is shipped, to ensure that the model has good initial performance.

[0059] In step S02, the machine learning model is re-trained based on the pre-training, which is usually done online. Specifically, during the actual driving of the vehicle, the real-time acquired road environment data is input into the pre-trained machine learning model to identify the road congestion event located on the second road. For the identified road congestion event, additional road environment data of the vehicle when driving on the first road and before entering the second road is extracted. These additional road environment data, for example, can be obtained from the data recorder or the memory of the environmental sensors of the vehicle, such as image sequences (the number n is selected according to experience or test calibration), point cloud data, ultrasonic data, etc. Then, the additional road environment data is used to construct an incremental training dataset to re-train the pre-trained model online.

[0060] In one embodiment, the vehicle user only identifies the road congestion event in front of the vehicle by means of the pre-trained machine learning model after switching from the main road to the auxiliary road during daily commuting. However, due to the proximity to the congestion point, it is not possible to temporarily plan a new alternative driving route, and only parking and waiting for the congestion to be removed. For this situation, the system automatically extracts the images recorded by the camera when the vehicle is driving on the main road and before turning to the auxiliary road. In these images taken at earlier times, the road congestion event on the auxiliary road in front is far away from the shooting position and the image is very small, but it can be seen vaguely. Then, these images labeled as "blocked auxiliary road" are added to the existing training data as incremental training data to re-train the machine learning model. After re-training, when the vehicle returns to this point on the main road again, the navigation system indicates that the vehicle should enter the auxiliary road, and if there is still a road congestion event on the auxiliary road at this time, this blocked auxiliary road situation will be identified earlier, thereby leaving more time for the vehicle to plan an alternative driving route.

[0061] The trigger condition for retraining can be flexibly set to optimize the model performance and reduce unnecessary computational overhead. For example, when the vehicle fails to re-plan an alternative route due to a too-late identification of a road congestion event during driving (i.e. invalid identification), the retraining of the model can be automatically triggered. In addition, the retraining of the model can also be triggered only for the commuter route or high-frequency route (e.g. daily commuter route) of the vehicle. In this way, the retraining can be more focused on specific scenarios, avoiding negative impact on the model performance, while ensuring the optimal performance of the model on the key routes.

[0062] Through the training process of steps S01 and S02, the Figure 2 However, step S02 does not have to be terminated only after the initial completion, but after the model is put into use, the model can still be updated online through the retraining process to further optimize its performance.

[0063] Figure 4 A flowchart of a driving assistance method according to another exemplary embodiment of the present application is shown. In this embodiment, Figure 2 The driving assistance method shown also includes additional steps S4, S5, S6 and S7.

[0064] In step S4, when the vehicle is driving on the first road, traffic condition information is obtained through a communication network, and it is checked whether the traffic condition information indicates that there is a road congestion event on the second road. The traffic condition information can be obtained from at least one of the following sources, for example: information pushed in real time by a traffic management center, real-time traffic data displayed by a vehicle-mounted navigation system, road information shared by other vehicles in a vehicle network.

[0065] In step S5, the traffic condition information is verified using the identification result of the machine learning model in step S2, for example, whether the two are consistent.

[0066] In one embodiment, the consistent situation includes that both the machine learning model and the traffic condition information identify that there is a road congestion event on the second road, or both identify that there is no congestion event. In another embodiment, the consistent situation not only involves the existence of the road congestion event, but also can cover the location and type of the event. For example, when the traffic condition information indicates that there is a rear-end accident 50 meters away from the intersection at the exit ramp of the highway, the machine learning model also identifies the rear-end accident at the same location.

[0067] On the contrary, inconsistent situations can include: the machine learning model identifies a road congestion event on the second road, but the traffic condition information shows that the road segment is clear; or the traffic condition information indicates a road congestion event, but the machine learning model does not detect any anomaly. Such differences can be due to the update delay of the traffic condition information. In addition, inconsistencies can also be reflected in the specific location or type of road congestion event. For example, the traffic condition information can report a blockage caused by construction on a road segment, while the machine learning model identifies a traffic jam caused by an accident, or there is a deviation in the specific location of the blockage determined by the two. These inconsistent situations need to be further analyzed and processed to ensure that the vehicle can make reasonable driving decisions based on the most accurate information.

[0068] In Figure 4 In the embodiment shown, if it is determined in step S5 that the recognition result of the machine learning model is consistent with the traffic condition information, no reporting operation is performed in step S6, for example, indicating that the current traffic condition information is up-to-date. On the contrary, if it is found in step S5 that the two are inconsistent, the recognition result of the machine learning model in step S2 is reported (for example, to the relevant traffic information platform) in step S7 to update the traffic condition information.

[0069] Figure 5 A schematic diagram showing the application of the driving assistance method according to the present application in one exemplary scenario is shown. In this exemplary scenario, the vehicle 1 is driving along the auxiliary road 62 according to a preset driving route 101, which indicates that the vehicle 1 should go straight into the roundabout under the viaduct. However, as shown, Figure 5 a rear-end collision accident 50 occurs at the entrance of the roundabout, two vehicles 2, 3 are across the center of the auxiliary road 62, and the owners 51, 52 get out to check, causing the road 62 to be completely impassable.

[0070] In this case, the vehicle 1 obtains road environment data in real time through its environment sensors, which are collected within a predetermined time period (e.g. 30 seconds) before the current time and are input into the pre-trained machine learning model. Since the machine learning model has only been pre-trained, although it can detect the road congestion event 50 in front, it cannot identify this event at an earlier time point. Therefore, when the rear-end collision is identified, the vehicle 1 has already entered the auxiliary road 62.

[0071] When the rear-end collision in front is identified, the vehicle 1 is still a distance away from the accident site, and the system immediately replans an alternative driving route 102 for the vehicle 1 to avoid this congestion point 50. As shown, Figure 5 the alternative driving route 102 guides the vehicle 1 to turn right into a branch road 63 at a junction before reaching the rear-end location, so as to bypass the congestion area and continue to the destination.

[0072] To address the above situation, additional road environment data stored earlier by the vehicle 1's environmental sensors can be extracted. This data was collected in advance before the vehicle 1 entered the auxiliary road 62 from the main road, including images taken within one minute before entering the auxiliary road 62, radar point cloud data, and ultrasonic data. Subsequently, this additional road environment data is used to construct an incremental training dataset for retraining the machine learning model.

[0073] Figure 6 A schematic diagram illustrating the application of the driving assistance method according to this application in another exemplary scenario is shown.

[0074] In this exemplary scenario, vehicle 1 is traveling along a preset route 101 and is about to switch from the main road 61 to the auxiliary road 62. It is worth noting that... Figure 5 The same location shown in Figure 50 experienced another blockage event. However, with Figure 5 The difference is that the machine learning model at this point has been retrained using additional road environment data collected earlier when it passed this location last time, thus learning from the previous experience and making optimizations. With the help of the retrained machine learning model, vehicle 1 can identify the congestion event 50 on the auxiliary road 62 earlier, even before it enters the auxiliary road 62 and is still traveling on the main road 61.

[0075] Accordingly, vehicle 1 immediately replans an alternative route. Since vehicle 1 has not yet entered auxiliary road 62, the system has more time and space to consider different detour options. For example, it can plan a better alternative route 103, instructing vehicle 1 to continue along main road 61 instead of entering auxiliary road 62 as originally planned. This not only avoids unnecessary detours but also reduces the additional time and safety risks that may result from temporary route changes. In this way, the retrained machine learning model significantly improves vehicle 1's adaptability and flexibility in complex traffic environments, providing drivers with a more efficient and safer navigation experience.

[0076] Although specific embodiments of this application are described in detail herein, they are given for illustrative purposes only and should not be construed as limiting the scope of this application. Various substitutions, modifications, and alterations can be conceived without departing from the spirit and scope of this application.

Claims

1. A driving assistance method, the driving assistance method comprising the following steps: Step S1: While the vehicle (1) is traveling on the first road (61), the environmental sensors of the vehicle (1) acquire road environment data in real time. The road environment data includes at least the first road (61) and the second road (62) branching off from the first road (61). The vehicle (1) plans to switch from the first road (61) to the second road (62) according to a preset driving route (101). Step S2: Based on the road environment data, identify a road blockage event (50) on the second road (62) using a trained machine learning model, the road blockage event (50) causing the vehicle (1) to be unable to continue driving along the second road (62); as well as Step S3: If a road congestion event (50) is detected, an alternative route (102, 103) is planned for the vehicle (1) to avoid the road congestion event (50).

2. The driving assistance method according to claim 1, wherein, The first road (61) is a main road, and the second road (62) is an auxiliary road or a branch road; or The first road (61) is a highway, and the second road (62) is an exit ramp.

3. The driving assistance method according to claim 1 or 2, wherein, The driving assistance method also includes pre-training a machine learning model in the following manner: Obtain the initial training dataset, which includes historical road environment data under various road types and traffic conditions and their corresponding road congestion event (50) annotation information; as well as The machine learning model is pre-trained using an initial training dataset until a preset termination condition is met, at which point training stops.

4. The driving assistance method according to any one of claims 1 to 3, wherein, The driving assistance method also includes retraining the machine learning model in the following manner: During the actual driving of the vehicle (1), real-time road environment data is input into a pre-trained machine learning model to identify road congestion events (50) located on the second road (62); For the identified road congestion events (50), additional road environment data obtained by the vehicle (1) with the help of environmental sensors when it is traveling on the first road (61) and has not yet entered the second road (62) is extracted to construct an incremental training dataset; as well as Online retraining of pre-trained machine learning models is achieved using incremental training datasets.

5. The driving assistance method according to claim 3 or 4, wherein, The machine learning model is pre-trained during the testing phase before the vehicle (1) leaves the factory; and / or The retraining of the machine learning model is carried out during the daily commuting of the vehicle (1).

6. The driving assistance method according to any one of claims 1 to 5, wherein, The alternative travel routes (102, 103) include: Guide vehicle (1) to continue along the first road (61) without entering the second road (62); or After guiding vehicle (1) onto the second road (62), switch to the third road (63) upstream of the location where the road congestion event (50) occurred.

7. The driving assistance method according to any one of claims 1 to 6, wherein, The real-time road environment data is collected by the environmental sensors of the vehicle (1) within a preset time period before the current moment. The length of the preset time period is fixed at tens of seconds or dynamically adjusted according to the real-time driving speed of the vehicle (1).

8. The driving assistance method according to any one of claims 1 to 7, wherein, The driving assistance method further includes the following steps: When the vehicle (1) is traveling on the first road (61), it obtains traffic information through the communication network and checks whether the traffic information indicates that there is a road blockage event (50) on the second road (62). The traffic information is verified using the recognition results obtained from the machine learning model in step S2. If the traffic condition information is inconsistent with the identification result of step S2, the identification result of step S2 will be reported to update the traffic condition information.

9. A driving assistance device, the driving assistance device comprising a memory (11) and a processor (12), the memory (11) storing computer program instructions, wherein when the computer program instructions are executed by the processor (12), the processor (12) is capable of performing a driving assistance method according to any one of claims 1 to 8.

10. A computer program product comprising computer program instructions, wherein, When the computer program instructions are executed by one or more processors, the one or more processors are able to perform the driving assistance method according to any one of claims 1 to 8.

Images