Vehicle starting intention recognition method and device, storage medium and computer equipment

Abstract

According to the vehicle starting intention recognition method and device, the storage medium and the computer equipment, in the driving process of the automatic driving vehicle, the vehicle parked on the roadside in the area in front of the automatic driving vehicle can be locked as the target vehicle, the motion information of the target vehicle is detected in real time, and the starting characteristics of the vehicle are obtained from dynamic behaviors; identifying vehicle lamp information of the target vehicle, and capturing an active operation intention of the driver from behavior interaction; heading information is determined according to the parking position of the target vehicle and comprises heading differences with surrounding vehicles and lanes, and the heading information is used for considering the starting possibility. Then, the motion information, the vehicle lamp information and the course information are input into a pre-trained starting intention recognition model, the starting probability output by the starting intention recognition model is obtained, graded response is conducted on the starting intention according to the starting probability, and the starting intention of a target vehicle can be found before the target vehicle actually moves; therefore, sudden brake caused by misjudgment of the automatic driving vehicle is avoided.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a method, apparatus, storage medium, and computer equipment for identifying vehicle starting intention. Background Art

[0002] With the widespread adoption of autonomous driving technology, real-time perception and decision-making capabilities for complex road conditions are crucial for safe driving. Parked vehicles can be haphazard and unpredictable, potentially starting and cutting into the vehicle's lane. Without foreseeable action, collisions can easily occur. Therefore, recognizing the starting intentions of parked vehicles and responding accordingly is crucial for autonomous driving.

[0003] Because the position and speed of roadside vehicles moving forward are subtle, identification presents numerous challenges. Existing methods typically use cameras to continuously capture two frames of images of parked vehicles, tracking the coordinate changes of the target vehicle within the images and calculating its actual displacement to determine whether the vehicle has started moving. However, since vehicle movement is subtle during starting, relying solely on vehicle motion information cannot detect a vehicle's intention to start moving before it actually moves. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defect in the prior art that the movement of the vehicle is not obvious when starting, and only relies on the vehicle's movement information, and cannot detect the vehicle's starting intention before the vehicle actually moves.

[0005] The present application provides a method for identifying vehicle start intention, the method comprising:

[0006] During the driving process of the autonomous driving vehicle, determining a target vehicle in an area ahead of the autonomous driving vehicle; the target vehicle is a vehicle parked on the roadside in the area ahead;

[0007] Detecting the motion information of the target vehicle in real time, identifying the headlight information of the target vehicle, and determining the heading information of the target vehicle according to the parking position of the target vehicle;

[0008] Determining a start intention recognition model; the start intention recognition model is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle start status annotated in the sample vehicle data as sample labels;

[0009] The motion information, the vehicle light information, and the heading information are input into the start intention recognition model to obtain a start probability output by the start intention recognition model, and the start intention of the target vehicle is determined based on the start probability.

[0010] Optionally, the real-time detection of the motion information of the target vehicle includes:

[0011] Collect the current position and heading information of the target vehicle according to a preset sampling frequency;

[0012] Determine the position change information and heading change information of the target vehicle at the current moment based on the position information and heading information at the current moment and the previous moment;

[0013] Determining the speed information of the target vehicle at a current moment based on the position change information and the heading change information;

[0014] The motion information of the target vehicle is generated according to the position change information, the heading change information and the speed information.

[0015] Optionally, the identifying the headlight information of the target vehicle includes:

[0016] capturing a headlight image of the target vehicle at the current moment according to a preset sampling frequency, and identifying the brake light status and turn signal status of the target vehicle at the current moment from the vehicle image;

[0017] Determining the brake light change state and turn light change state of the target vehicle at the current moment according to the brake light state and turn light state at the current moment and the previous moment;

[0018] The headlight information of the target vehicle is generated according to the change state of the brake light and the change state of the turn signal.

[0019] Optionally, determining the heading information of the target vehicle according to the parking position of the target vehicle includes:

[0020] determining a centerline of a parking position of the target vehicle, and determining a vehicle heading difference and a vehicle offset distance between the target vehicle and the parking position based on the centerline;

[0021] Identifying a target vehicle preceding the target vehicle as the nearest preceding vehicle, and identifying a target vehicle following the target vehicle as the nearest following vehicle;

[0022] Heading difference data among the target vehicle, the nearest preceding vehicle, and the nearest following vehicle are determined, and heading information of the target vehicle is generated based on the vehicle heading differences, the vehicle offset distance, and the heading difference data.

[0023] Optionally, determining the heading difference data among the target vehicle, the nearest preceding vehicle, and the nearest following vehicle includes:

[0024] determining a first heading difference between the target vehicle and the nearest preceding vehicle, determining a second heading difference between the target vehicle and the nearest following vehicle, and determining a third heading difference between the nearest preceding vehicle and the nearest following vehicle;

[0025] Heading difference data is generated based on the first heading difference, the second heading difference, and the third heading difference.

[0026] Optionally, determining the start intention recognition model includes:

[0027] Acquire sample vehicle data; the sample vehicle data includes parking feature information of multiple target vehicles and actual vehicle starting status, the parking feature information including motion information, vehicle light information, and heading information;

[0028] Inputting sample vehicle data into a preset machine learning model to obtain a predicted vehicle starting state output by the machine learning model;

[0029] Training the machine learning model with the goal of making the predicted vehicle starting state approach the actual vehicle starting state of the sample vehicle data;

[0030] When the machine learning model meets the preset training conditions, the trained machine learning model is used as the starting intent recognition model.

[0031] Optionally, determining the starting intention of the target vehicle according to the starting probability includes:

[0032] If the starting probability exceeds a first preset threshold, determining that the starting intention of the target vehicle is in a starting state;

[0033] If the starting probability exceeds a second preset threshold and does not exceed the first preset threshold, determining that the starting intention of the target vehicle is in a potential starting state;

[0034] If the starting probability does not exceed the second preset threshold, it is determined that the starting intention of the target vehicle is in a non-starting state.

[0035] The present application also provides a vehicle start intention recognition device, comprising:

[0036] A vehicle identification module is used to identify a target vehicle in an area ahead of the autonomous driving vehicle during its travel; the target vehicle is a vehicle parked on the roadside in the area ahead;

[0037] An information acquisition module is used to detect the motion information of the target vehicle in real time, identify the headlight information of the target vehicle, and determine the heading information of the target vehicle according to the parking position of the target vehicle;

[0038] a model determination module for determining a start intention recognition model; the start intention recognition model is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle start states annotated in the sample vehicle data as sample labels;

[0039] The intention recognition module is used to input the motion information, the vehicle light information and the heading information into the start intention recognition model, obtain the start probability output by the start intention recognition model, and determine the start intention of the target vehicle based on the start probability.

[0040] The present application also provides a storage medium, which stores computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the vehicle starting intention recognition method as described in any of the above embodiments.

[0041] The present application also provides a computer device, comprising: one or more processors, and a memory;

[0042] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the vehicle starting intention recognition method as described in any one of the above embodiments are performed.

[0043] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0044] The present application provides a roadside vehicle start intention recognition method, apparatus, storage medium, and computer device. During the driving process of an autonomous vehicle, a vehicle parked on the roadside in the area ahead of the autonomous vehicle can be identified as a target vehicle and its start intention can be recognized. During this process, the autonomous vehicle can detect the target vehicle's motion information in real time to obtain the vehicle's start characteristics based on its dynamic behavior. It can also identify the target vehicle's headlight information to capture the driver's active operational intention based on behavioral interaction. Simultaneously, it determines heading information based on the target vehicle's parking position, including the heading difference from surrounding vehicles and lanes, to consider the likelihood of starting. Subsequently, a start intention recognition model trained using sample vehicle data labeled with real vehicle start states can be determined. This model can effectively handle nonlinear feature relationships between multiple pieces of information. Therefore, after inputting motion information, headlight information, and heading information into the start intention recognition model, a start probability output by the start intention recognition model can be obtained. This start probability is used to provide a graded response to the start intention, enabling detection of the target vehicle's start intention before it actually moves, thereby avoiding sudden braking due to misjudgment by the autonomous vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0046] Figure 1 A flow chart of a method for identifying vehicle start intention provided in an embodiment of the present application;

[0047] Figure 2 One of the planar schematic diagrams of a vehicle heading deviation and a vehicle offset distance provided in an embodiment of the present application;

[0048] Figure 3 A second planar schematic diagram of a vehicle heading deviation and a vehicle offset distance provided in an embodiment of the present application;

[0049] Figure 4 A schematic diagram of the structure of a vehicle start intention recognition device provided in an embodiment of the present application;

[0050] Figure 5 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] Because the position and speed of roadside vehicles moving forward are subtle, identification presents numerous challenges. Existing methods typically use cameras to continuously capture two frames of images of parked vehicles, tracking the coordinate changes of the target vehicle within the images and calculating its actual displacement to determine whether the vehicle has started moving. However, since vehicle movement is subtle during starting, relying solely on vehicle motion information cannot detect a vehicle's intention to start moving before it actually moves.

[0053] Based on this, this application proposes the following technical solutions, please refer to the following for details:

[0054] In one embodiment, Figure 1 As shown, Figure 1 A flowchart of a vehicle start intention recognition method provided in an embodiment of the present application; the present application provides a vehicle start intention recognition method, which specifically includes the following:

[0055] S110: During the driving of the autonomous driving vehicle, a target vehicle in the area ahead of the autonomous driving vehicle is determined; the target vehicle is a vehicle parked on the roadside in the area ahead.

[0056] In this step, while the autonomous vehicle is driving, the autonomous vehicle can lock onto a vehicle parked on the roadside in the area in front of the autonomous vehicle as a target vehicle through the built-in vehicle-mounted control system to identify its starting intention and prevent the vehicle from suddenly starting and cutting into its own lane.

[0057] Specifically, the autonomous vehicle can continuously monitor the area ahead through its built-in on-board control system, and treat vehicles parked on the roadside in the area ahead as target vehicles. It can then use multi-source sensors such as cameras, lidar, and millimeter-wave radar to simultaneously analyze the status of each target vehicle, so as to identify whether the target vehicle has the intention to start and enter its own lane, so that the on-board control system can dynamically adjust the driving strategy of the autonomous vehicle based on the recognition results.

[0058] It is understood that the area in front of an autonomous vehicle refers to the area within a certain range along its forward path, based on the autonomous vehicle's current direction of travel. For example, based on the right-hand driving rule, the area in front of the autonomous vehicle in this application may be a rectangular area 60 meters in front of the autonomous vehicle and 10 meters to the right of the autonomous vehicle.

[0059] S120: Detecting the motion information of the target vehicle in real time, identifying the headlight information of the target vehicle, and determining the heading information of the target vehicle according to the parking position of the target vehicle.

[0060] In this step, after the target vehicle is determined through step S110, the vehicle-mounted control system can detect the motion information of the target vehicle in real time to obtain the vehicle's starting characteristics from dynamic behavior; it can also identify the target vehicle's headlight information to capture the driver's active operation intention from behavioral interaction; at the same time, it can determine the heading information based on the target vehicle's parking position, including the heading difference with surrounding vehicles and lanes, to consider the possibility of starting.

[0061] Among them, motion information refers to the physical motion parameters such as position, speed, acceleration, displacement trajectory, heading angle and its rate of change that change with time in space, which can be used to describe the dynamic behavior of the target vehicle; vehicle light information refers to the status information related to the driver's operating intention obtained by sensing the vehicle's external light signals, which can be used to capture the vehicle's interactive behavior characteristics; heading information refers to the directional relationship parameter between the target vehicle's current direction and the parking space, road or lane geometry, which can be used to determine its potential movement direction after starting.

[0062] Specifically, during the operation of an autonomous vehicle, the on-board control system can use multi-sensor fusion technology to collect and analyze the dynamic motion information of the target vehicle in real time. For example, it can use lidar and vision systems to detect the position and posture changes of the target vehicle over a continuous period of time, and then calculate its displacement trend and slight deviations in the heading angle. These subtle signs of movement can serve as important dynamic behavior characteristics of the target vehicle about to start. At the same time, the on-board control system can also use cameras to perceive the light signal status of the target vehicle, identifying key visual signals such as whether its brake lights are off and whether the turn signals are activated, thereby determining whether the driver is performing active driving operations. In addition, the on-board control system uses high-precision maps combined with the actual parking position of the target vehicle to measure its orientation, assess whether its current posture meets the conditions for departure, and further infer the possibility of its starting based on the current road structure.

[0063] Therefore, through multi-dimensional analysis of motion information, headlight information and heading information, the on-board control system can identify potential starting behaviors earlier, and accurately predict whether its path will intersect with the autonomous driving vehicle after starting, so as to adjust its own driving strategy in advance and avoid the risk of collision caused by sudden cutting in.

[0064] S130: Determine a starting intention recognition model; the starting intention recognition model is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle starting status marked in the sample vehicle data as sample labels.

[0065] In this step, after obtaining the relevant information of the target vehicle in step S120, the vehicle control system can also obtain a pre-trained start intention recognition model to perform intention recognition on it, thereby obtaining a high-precision recognition result.

[0066] It should be understood that the start intention recognition model here is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle start states annotated in the sample vehicle data as sample labels. The machine learning model here can adopt GBDT, LightGBM, XGBoost, AdaBoost, etc., without limitation.

[0067] Specifically, during training, the machine learning model learns the features of numerous vehicle start and non-start behavior samples from historical data, enabling it to model response patterns for different feature combinations. This allows it to maintain high fitting and generalization performance when handling nonlinear feature relationships and interactions. Consequently, at runtime, the trained start intention recognition model can comprehensively determine whether the current target vehicle has a start intention and output a corresponding classification result or probability score.

[0068] S140: Inputting the motion information, the vehicle light information, and the heading information into a start intention recognition model to obtain a start probability output by the start intention recognition model, and determining the start intention of the target vehicle based on the start probability.

[0069] In this step, after the starting intention recognition model is determined in step S130, the vehicle control system can input the motion information, headlight information and heading information into the starting intention recognition model, and then obtain the starting probability output by the starting intention recognition model. Here, the starting intention is graded according to the starting probability, and the starting intention of the target vehicle can be discovered before the target vehicle actually moves, thereby avoiding sudden braking of the autonomous driving vehicle due to misjudgment.

[0070] Specifically, before model prediction, the vehicle control system extracts and converts multiple sensory data, such as the target vehicle's current motion, headlights, and heading, into its feature data. This data is then fed into the start intention recognition model. Because the model is trained using a machine learning algorithm, it can integrate the changing trends of various features and output a probability value representing the likelihood of a start, known as the start probability.

[0071] It is understandable that the starting probability here can be in the range of [0,1]. The smaller the starting probability, the smaller the starting intention of the target vehicle. At this time, the on-board control system can maintain the normal monitoring state. On the contrary, it means that the starting intention of the target vehicle is greater. At this time, the on-board control system can enter the early warning mode and adjust the speed, following distance or path planning of the autonomous driving vehicle in advance to improve the ability to respond to emergencies. In this way, the on-board control system can achieve a transition from "passive response" to "active prediction", and can accurately identify the behavioral intention of the target vehicle before it actually starts, thereby avoiding the path interference or collision risk that it may cause to the autonomous driving vehicle in advance.

[0072] In the above embodiment, while the autonomous vehicle is driving, a vehicle parked on the roadside in the area ahead of the autonomous vehicle can be identified as a target vehicle and its start intention can be identified. During this process, the autonomous vehicle can detect the target vehicle's motion information in real time to obtain the vehicle's start characteristics from its dynamic behavior. It can also identify the target vehicle's headlight information to capture the driver's active operational intention from behavioral interaction. Simultaneously, based on the target vehicle's parked position, it can determine its heading information, including the heading difference from surrounding vehicles and lanes, to consider the possibility of starting. Subsequently, a start intention recognition model trained using a machine learning model such as GBDT, LightGBM, XGBoost, or AdaBoost can be determined. This model can effectively handle nonlinear feature relationships between multiple pieces of information. Therefore, after inputting the motion information, headlight information, and heading information into the start intention recognition model, the start intention recognition model outputs a start probability. Here, the start probability is used to provide a graded response to the start intention, allowing the target vehicle's start intention to be detected before it actually moves, thereby avoiding sudden braking due to misjudgment by the autonomous vehicle.

[0073] In one embodiment, the process of detecting the target vehicle's motion information in real time in step S120 may include:

[0074] S1211: Collect the target vehicle's current position information and heading information according to a preset sampling frequency.

[0075] S1212: Determine the position change information and heading change information of the target vehicle at the current moment based on the position information and heading information at the current moment and the previous moment.

[0076] S1213: Determine the speed information of the target vehicle at the current moment based on the position change information and the heading change information.

[0077] S1214: Generate motion information of the target vehicle based on the position change information, the heading change information, and the speed information.

[0078] In this embodiment, when detecting the motion information of the target vehicle, the on-board control system can collect the position information and heading information of the target vehicle at the current moment according to a preset sampling frequency and record the data, and then obtain the position information and heading information at the current moment and the previous moment from the recorded data, so as to calculate the position change information and heading change information at the current moment, and then determine the speed information of the target vehicle at the current moment based on the position change information and heading change information. Finally, the on-board control system can generate the motion information of the target vehicle based on the position change information, heading change information and speed information.

[0079] It is understandable that during the driving process of an autonomous vehicle, the on-board control system can start a periodic data collection program based on a set time interval, and then collect relevant information of the target vehicle according to a preset sampling frequency, and then mark the collected data with a corresponding timestamp and store it for subsequent analysis of the relevant information of the target vehicle.

[0080] Specifically, the vehicle control system can identify the target vehicle through a perception module such as a visual sensor or lidar, and then extract the target vehicle's current position and heading information. Simultaneously, the computer device can obtain the position and heading information collected at the previous moment and calculate the information change between the two moments, including the position change information and the heading change information. The vehicle control system can then use the time interval between the two moments and combine these two change information to calculate the target vehicle's speed information, thereby detecting the target vehicle's motion information at the current moment.

[0081] It can be understood that the position change information can be used to reflect whether the target vehicle has undergone obvious spatial displacement between two consecutive moments, and the heading change information can be used to reflect the degree of change in the direction of the target vehicle's head between two moments; the speed information represents the measurement of the position change rate between two moments, which is the core physical quantity for determining whether the vehicle is in motion.

[0082] In one embodiment, the process of identifying the headlight information of the target vehicle in step S120 may include:

[0083] S1221: Capture a headlight image of the target vehicle at the current moment according to a preset sampling frequency, and identify the brake light status and turn signal status of the target vehicle at the current moment from the vehicle image.

[0084] S1222: Determine the brake light change state and turn light change state of the target vehicle at the current moment based on the brake light state and turn light state at the current moment and the previous moment.

[0085] S1223: Generate headlight information of the target vehicle according to the change state of the brake light and the change state of the turn signal.

[0086] In this embodiment, when identifying the headlight information of the target vehicle, the on-board control system can capture the headlight image of the target vehicle at the current moment according to a preset sampling frequency, and identify the brake light status and turn light status of the target vehicle at the current moment from the vehicle image, and record the data. Then, the brake light status and turn light status at the current moment and the previous moment are obtained from the recorded data, so that the brake light change status and turn light change status at the current moment can be calculated; finally, the on-board control system can generate the headlight information of the target vehicle based on the brake light change status and turn light change status.

[0087] Specifically, the vehicle control system can capture images of the target vehicle through a forward-facing camera or a high-resolution visual sensor, obtain an image of its headlights at the current moment, and crop the captured image to the target area and correct the lighting conditions to ensure that the lighting features are clearly identifiable in the image. Then, the vehicle control system can analyze the key areas in the image through a deep learning visual recognition model or a traditional image processing algorithm to accurately extract the brake light status and turn signal status of the target vehicle at the current moment. At the same time, the computer device can obtain the brake light status and turn signal status captured at the previous moment, and then calculate the current light change status, including the brake light change status and turn signal change status, by comparing the brake light and turn signal status at these two moments, thereby forming the headlight information.

[0088] It can be understood that the change in brake light status is the physical action of a vehicle preparing to start from a standstill, and can be used to sense when the vehicle is about to start. For example, when the brake lights are constantly on, it means that the vehicle may be parked and braking. If the target vehicle's brake lights turn from on to off, it means that the driver has released the brake pedal and the vehicle may be about to start. The change in turn signal status is a key signal for determining the path direction risk of the target vehicle after it starts. For example, when the turn signal is off, it means that the vehicle may be parked. If the target vehicle's left turn signal turns from off to on, it means that the driver has proactively turned on the signal to pull out of the road and the vehicle is preparing to enter the main lane.

[0089] In one embodiment, the process of determining the heading information of the target vehicle according to the parking position of the target vehicle in step S120 may include:

[0090] S1231: Determine a center line of the parking position of the target vehicle, and determine a vehicle heading difference and a vehicle offset distance between the target vehicle and the parking position based on the center line.

[0091] S1232: Identify the target vehicle preceding the target vehicle as the nearest preceding vehicle, and identify the target vehicle following the target vehicle as the nearest following vehicle.

[0092] S1233: Determine the heading difference data between the target vehicle, the nearest preceding vehicle, and the nearest following vehicle, and generate the heading information of the target vehicle based on the vehicle heading differences, the vehicle offset distance, and the heading difference data.

[0093] In this embodiment, when determining the heading information of the target vehicle, the on-board control system can first determine the center line of the target vehicle's parking position, and determine the vehicle heading difference and vehicle offset distance between the target vehicle and the parking position based on the center line; at the same time, the previous target vehicle and the next target vehicle of the target vehicle can also be identified as the nearest front vehicle and the nearest rear vehicle, respectively, and then the heading difference data between the target vehicle, the nearest front vehicle and the nearest rear vehicle can be determined, and the heading information of the target vehicle can be generated based on the vehicle heading difference, vehicle offset distance and heading difference data.

[0094] Specifically, the vehicle control system uses image recognition or LiDAR scanning to acquire the target vehicle's actual posture data, including body coordinates and vehicle head orientation. It then calculates the angle between the target vehicle's current head orientation and the centerline, serving as a measure of vehicle heading deviation. The system also measures the lateral offset between the target vehicle's leftmost side and the leftmost side of its parking position to indicate whether the vehicle is fully positioned or significantly tilted.

[0095] It is understandable that since the target vehicle may be temporarily parked on the roadside or even illegally parked, its parking position is not necessarily a parking space, but may also be in the right lane. Figure 2 and Figure 3 As shown, Figure 2 One of the planar schematic diagrams of a vehicle heading deviation and a vehicle offset distance provided in an embodiment of the present application; Figure 2 A second planar schematic diagram of a vehicle heading deviation and a vehicle offset distance provided in an embodiment of the present application; Figure 2 When the target vehicle is parked in the right lane, the angle between the target vehicle's head direction and the road centerline is the vehicle heading difference. The lane distance between the leftmost side of the target vehicle and the left boundary of the road is the vehicle offset distance ; Figure 3 When the target vehicle is parked in a parking space, the angle between the target vehicle's head direction and the centerline of the parking space is the vehicle heading difference. The distance between the leftmost side of the target vehicle and the left boundary of the parking space is the vehicle offset distance. .

[0096] Furthermore, to further enhance the spatial rationality of the judgment, the vehicle control system can also expand its perception range in the direction ahead and behind the target vehicle, identifying two parked vehicles adjacent to the target vehicle as the closest preceding and following vehicles, respectively. By comparing the heading angles of these three vehicles, the vehicle control system can calculate the heading difference between the target vehicle and its front and rear neighbors to determine whether it has a tendency to deviate significantly from the surrounding vehicle arrangement.

[0097] In one embodiment, the process of determining the heading difference data between the target vehicle, the nearest preceding vehicle, and the nearest following vehicle in step S1233 may include:

[0098] S2331: Determine a first heading difference between the target vehicle and the nearest preceding vehicle, determine a second heading difference between the target vehicle and the nearest following vehicle, and determine a third heading difference between the nearest preceding vehicle and the nearest following vehicle.

[0099] S2332: Generate heading difference data according to the first heading difference, the second heading difference, and the third heading difference.

[0100] In this embodiment, when calculating the heading difference data between the target vehicle and the adjacent vehicles, the on-board control system can respectively determine the first heading difference between the target vehicle and the nearest preceding vehicle, the second heading difference between the target vehicle and the nearest following vehicle, and the third heading difference between the nearest preceding vehicle and the nearest following vehicle; finally, the on-board control system can generate the heading difference data based on the first heading difference, the second heading difference and the third heading difference.

[0101] Specifically, the on-board control system can calculate the angle difference between the target vehicle and the nearest preceding vehicle based on their actual heading angles to obtain a first heading difference, which is used to indicate the degree of deviation in direction between the target vehicle and the vehicle in front of it. Similarly, the on-board control system can process the heading angle between the target vehicle and the nearest following vehicle to obtain a second heading difference, which is used to indicate the degree of deviation in direction between the target vehicle and the vehicle behind it. Furthermore, the on-board control system can obtain a third heading difference between the nearest preceding vehicle and the nearest following vehicle to determine whether the overall arrangement direction of the adjacent vehicles in front and behind the target vehicle remains consistent, which can be used as a reference for comparison. Through a comprehensive evaluation of these three directional differences, a set of multi-angle heading difference data can be formed, allowing the on-board control system to analyze whether the target vehicle has significantly deviated from the normal arrangement trend in direction, or whether there are signs of adjusting the direction of the vehicle head in preparation for departure.

[0102] In one embodiment, the process of determining the departure intention recognition model in step S130 may include:

[0103] S131: Acquire sample vehicle data; the sample vehicle data includes parking feature information of multiple target vehicles and the actual vehicle starting status, and the parking feature information includes motion information, headlight information, and heading information.

[0104] S132: Input the sample vehicle data into a preset machine learning model to obtain the predicted vehicle starting state output by the machine learning model.

[0105] S133: Training the machine learning model with the goal of predicting a vehicle starting state close to the actual vehicle starting state of the sample vehicle data.

[0106] S134: When the machine learning model meets the preset training conditions, the trained machine learning model is used as the starting intent recognition model.

[0107] In this embodiment, when determining the starting intention recognition model, the present application may first select a corresponding machine learning model for improvement and training, such as GBDT, LightGBM, XGBoost or AdaBoost, etc., and then may perform improvement and training based on the machine learning model. During the training process, sample vehicle data containing parking feature information of multiple target vehicles and the actual vehicle starting state may be obtained first. After inputting the sample vehicle data into the machine learning model, the predicted vehicle starting state output by the machine learning model may be obtained. Then, the present application may train the machine learning model with the goal of predicting the vehicle starting state to be close to the actual vehicle starting state in the sample vehicle data. When the machine learning model meets certain training conditions or parameter convergence conditions, the training is deemed to be completed, and the trained model may be used as the final starting intention recognition model.

[0108] It should be noted that the present application can also store the trained start intention recognition model in the vehicle control system, so that when the vehicle starts to recognize the intention, the vehicle control system can directly call the pre-stored start intention recognition model to recognize the start intention of the target vehicle.

[0109] In one embodiment, the process of determining the target vehicle's starting intention based on the starting probability in step S140 may include:

[0110] S141: If the starting probability exceeds a first preset threshold, determining that the starting intention of the target vehicle is in a starting state.

[0111] S142: If the starting probability exceeds the second preset threshold and does not exceed the first preset threshold, the starting intention of the target vehicle is determined to be in a potential starting state.

[0112] S143: If the starting probability does not exceed the second preset threshold, determining that the starting intention of the target vehicle is in a non-starting state.

[0113] In this embodiment, after the start intention recognition model outputs the target vehicle's start probability, the vehicle control system can classify the target vehicle's start intention based on pre-set first and second thresholds. For example, when the start probability exceeds the first threshold, the target vehicle's start intention is considered to be in a start state; when the start probability exceeds the second threshold but does not exceed the first threshold, the target vehicle's start intention is considered to be in a potential start state; and when the start probability does not exceed the second threshold, the target vehicle's start intention is considered to be in a non-start state.

[0114] It is understood that the first preset threshold can be used to calibrate the lower limit of the probability of obvious starting behavior, while the second preset threshold is used to identify the sensitive range of potential starting signs. Therefore, when the starting probability exceeds the first preset threshold, it indicates that the target vehicle's behavioral characteristics are highly similar to historical starting vehicles. Therefore, the on-board control system can mark it as a starting state and trigger high-priority safety strategies such as active avoidance, deceleration, increasing vehicle distance, or path adjustment to avoid collision. When the starting probability is between the second preset threshold and the first preset threshold, it indicates that the target vehicle has a certain possibility of starting but has not yet reached a strong momentum. Therefore, the on-board control system can mark it as a potential starting state and enter a monitoring and warning mode to focus on it and limit vehicle speed increase to prevent the target vehicle from suddenly merging or cutting into the main lane. When the starting probability is lower than the second preset threshold, it indicates that the probability of the target vehicle starting in the short term is low. Therefore, the on-board control system can mark it as a non-starting state and maintain normal driving status.

[0115] The following describes a vehicle start intention recognition device provided in an embodiment of the present application. The vehicle start intention recognition device described below and the vehicle start intention recognition method described above can be referenced to each other.

[0116] In one embodiment, Figure 4 As shown, Figure 4 This is a schematic diagram of the structure of a vehicle start intention recognition device provided in an embodiment of the present application. The present application also provides a vehicle start intention recognition device, including a vehicle determination module 210, an information acquisition module 220, a model determination module 230, and an intention recognition module 240, specifically including the following:

[0117] The vehicle determination module 210 is used to determine a target vehicle in the area ahead of the autonomous driving vehicle during its driving process; the target vehicle is a vehicle parked on the roadside in the area ahead.

[0118] The information acquisition module 220 is used to detect the motion information of the target vehicle in real time, identify the headlight information of the target vehicle, and determine the heading information of the target vehicle according to the parking position of the target vehicle.

[0119] The model determination module 230 is used to determine the starting intention recognition model; the starting intention recognition model is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle starting status marked in the sample vehicle data as sample labels.

[0120] The intention recognition module 240 is used to input the motion information, vehicle light information and heading information into the start intention recognition model, obtain the start probability output by the start intention recognition model, and determine the start intention of the target vehicle based on the start probability.

[0121] In the above embodiment, during the driving process of the autonomous vehicle, a vehicle parked on the roadside in the area ahead of the autonomous vehicle can be identified as a target vehicle and its start intention can be identified. During this process, the autonomous vehicle can detect the target vehicle's motion information in real time to obtain the vehicle's start characteristics from its dynamic behavior. It can also identify the target vehicle's headlight information to capture the driver's active operational intention from behavioral interaction. Simultaneously, based on the target vehicle's parking position, it can determine its heading information, including the heading difference from surrounding vehicles and lanes, to consider the likelihood of start. Subsequently, a start intention recognition model trained with sample vehicle data labeled with real vehicle start states can be determined. This model can effectively handle nonlinear feature relationships between multiple pieces of information. Therefore, after inputting motion information, headlight information, and heading information into the start intention recognition model, the start intention recognition model outputs a start probability. This start probability is used to provide a graded response to the start intention, allowing the target vehicle's start intention to be detected before it actually moves, thereby avoiding sudden braking due to misjudgment by the autonomous vehicle.

[0122] In one embodiment, the information acquisition module 220 may include:

[0123] The information collection submodule is used to collect the target vehicle's current position information and heading information according to a preset sampling frequency.

[0124] The change information determination submodule is used to determine the position change information and heading change information of the target vehicle at the current moment based on the position information and heading information at the current moment and the previous moment.

[0125] The speed calculation submodule is used to determine the speed information of the target vehicle at the current moment based on the position change information and the heading change information.

[0126] The first information aggregation submodule is used to generate motion information of the target vehicle based on the position change information, the heading change information and the speed information.

[0127] In one embodiment, the information acquisition module 220 may further include:

[0128] The state recognition submodule is used to capture the headlight image of the target vehicle at the current moment according to a preset sampling frequency, and identify the brake light state and turn signal state of the target vehicle at the current moment from the vehicle image.

[0129] The change state determination submodule is used to determine the target vehicle's brake light change state and turn light change state at the current moment based on the brake light state and turn light state at the current moment and the previous moment.

[0130] The second information aggregation submodule is used to generate the headlight information of the target vehicle according to the change state of the brake light and the change state of the turn signal.

[0131] In one embodiment, the information acquisition module 220 may further include:

[0132] The difference calculation submodule is used to determine the center line of the target vehicle's parking position and determine the vehicle heading difference and vehicle offset distance between the target vehicle and the parking position based on the center line.

[0133] The vehicle determination submodule is used to identify the target vehicle in front of the target vehicle as the nearest preceding vehicle, and to identify the target vehicle behind the target vehicle as the nearest following vehicle.

[0134] The third information aggregation submodule is used to determine the heading difference data between the target vehicle, the nearest preceding vehicle and the nearest following vehicle, and generate the heading information of the target vehicle based on the vehicle heading difference, the vehicle offset distance and the heading difference data.

[0135] In one embodiment, the third information summary may include:

[0136] The heading difference determining unit is configured to determine a first heading difference between the target vehicle and the nearest preceding vehicle, a second heading difference between the target vehicle and the nearest following vehicle, and a third heading difference between the nearest preceding vehicle and the nearest following vehicle.

[0137] The difference data aggregation unit is configured to generate heading difference data according to the first heading difference, the second heading difference, and the third heading difference.

[0138] In one embodiment, the model determination module 230 may include:

[0139] The model building submodule is used to obtain sample vehicle data; the sample vehicle data includes the parking feature information of multiple target vehicles and the actual vehicle starting status. The parking feature information includes motion information, headlight information and heading information.

[0140] The model prediction submodule is used to input sample vehicle data into a preset machine learning model to obtain the predicted vehicle starting status output by the machine learning model.

[0141] The model training submodule is used to train the machine learning model with the goal of predicting the vehicle starting state to be close to the actual vehicle starting state of the sample vehicle data.

[0142] The model determination submodule is used to use the trained machine learning model as the starting intent recognition model when the machine learning model meets the preset training conditions.

[0143] In one embodiment, the intent recognition module 240 may include:

[0144] The first state determination submodule is configured to determine that the target vehicle's start intention is in the start state if the start probability exceeds a first preset threshold.

[0145] The second state determination submodule is configured to determine that the target vehicle's start intention is in a potential start state if the start probability exceeds a second preset threshold and does not exceed a first preset threshold.

[0146] The third state determination submodule is configured to determine that the starting intention of the target vehicle is a non-starting state if the starting probability does not exceed a second preset threshold.

[0147] In one embodiment, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the vehicle starting intention recognition method as described in any of the above embodiments.

[0148] In one embodiment, the present application also provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the vehicle starting intention recognition method as described in any of the above embodiments.

[0149] Schematically, as Figure 5 As shown, Figure 5 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 can be provided as a server. Figure 5Computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions executable by processing component 302, such as application programs. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, processing component 302 is configured to execute the instructions to perform the vehicle launch intention recognition method according to any of the above-described embodiments.

[0150] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0151] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0152] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0153] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0154] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying vehicle starting intention, characterized in that: The method comprises: During the driving process of the autonomous driving vehicle, determining a target vehicle in an area ahead of the autonomous driving vehicle; the target vehicle is a vehicle parked on the roadside in the area ahead; Detecting the motion information of the target vehicle in real time, identifying the headlight information of the target vehicle, and determining the heading information of the target vehicle according to the parking position of the target vehicle; Determining a start intention recognition model; the start intention recognition model is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle start status annotated in the sample vehicle data as sample labels; The motion information, the vehicle light information, and the heading information are input into the start intention recognition model to obtain a start probability output by the start intention recognition model, and the start intention of the target vehicle is determined based on the start probability.

2. The vehicle start intention recognition method according to claim 1, characterized in that: The real-time detection of the motion information of the target vehicle includes: Collect the current position and heading information of the target vehicle according to a preset sampling frequency; Determine the position change information and heading change information of the target vehicle at the current moment based on the position information and heading information at the current moment and the previous moment; Determining the speed information of the target vehicle at a current moment based on the position change information and the heading change information; The motion information of the target vehicle is generated according to the position change information, the heading change information and the speed information.

3. The vehicle start intention recognition method according to claim 1, characterized in that: The identifying the headlight information of the target vehicle includes: capturing a headlight image of the target vehicle at the current moment according to a preset sampling frequency, and identifying the brake light status and turn signal status of the target vehicle at the current moment from the vehicle image; Determining the brake light change state and turn light change state of the target vehicle at the current moment according to the brake light state and turn light state at the current moment and the previous moment; The headlight information of the target vehicle is generated according to the change state of the brake light and the change state of the turn signal.

4. The vehicle start intention recognition method according to claim 1, characterized in that: Determining the heading information of the target vehicle according to the parking position of the target vehicle includes: determining a centerline of a parking position of the target vehicle, and determining a vehicle heading difference and a vehicle offset distance between the target vehicle and the parking position based on the centerline; Identifying a target vehicle preceding the target vehicle as the nearest preceding vehicle, and identifying a target vehicle following the target vehicle as the nearest following vehicle; Heading difference data among the target vehicle, the nearest preceding vehicle, and the nearest following vehicle are determined, and heading information of the target vehicle is generated based on the vehicle heading differences, the vehicle offset distance, and the heading difference data.

5. The vehicle start intention recognition method according to claim 4, characterized in that: The determining of the heading difference data among the target vehicle, the nearest preceding vehicle, and the nearest following vehicle includes: determining a first heading difference between the target vehicle and the nearest preceding vehicle, determining a second heading difference between the target vehicle and the nearest following vehicle, and determining a third heading difference between the nearest preceding vehicle and the nearest following vehicle; Heading difference data is generated based on the first heading difference, the second heading difference, and the third heading difference.

6. The vehicle start intention recognition method according to claim 1, characterized in that: Determining the start intention recognition model includes: Acquire sample vehicle data; the sample vehicle data includes parking feature information of multiple target vehicles and actual vehicle starting status, the parking feature information including motion information, vehicle light information, and heading information; Inputting sample vehicle data into a preset machine learning model to obtain a predicted vehicle starting state output by the machine learning model; Training the machine learning model with the goal of making the predicted vehicle starting state approach the actual vehicle starting state of the sample vehicle data; When the machine learning model meets the preset training conditions, the trained machine learning model is used as the starting intent recognition model.

7. The vehicle start intention recognition method according to claim 1, characterized in that: The determining the starting intention of the target vehicle according to the starting probability includes: If the starting probability exceeds a first preset threshold, determining that the starting intention of the target vehicle is in a starting state; If the starting probability exceeds a second preset threshold and does not exceed the first preset threshold, determining that the starting intention of the target vehicle is in a potential starting state; If the starting probability does not exceed the second preset threshold, it is determined that the starting intention of the target vehicle is in a non-starting state.

8. A vehicle starting intention recognition device, characterized in that: include: a vehicle determination module, configured to determine a target vehicle in an area ahead of the autonomous vehicle during its travel; The target vehicle is a vehicle parked on the roadside in the front area; An information acquisition module is used to detect the motion information of the target vehicle in real time, identify the headlight information of the target vehicle, and determine the heading information of the target vehicle according to the parking position of the target vehicle; a model determination module for determining a start intention recognition model; the start intention recognition model is obtained by training a preset machine learning model using sample vehicle data as training samples and the actual vehicle start states annotated in the sample vehicle data as sample labels; The intention recognition module is used to input the motion information, the vehicle light information and the heading information into the start intention recognition model, obtain the start probability output by the start intention recognition model, and determine the start intention of the target vehicle based on the start probability.

9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, enable the one or more processors to perform the steps of the vehicle starting intention recognition method according to any one of claims 1 to 7.

10. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the vehicle starting intention recognition method according to any one of claims 1 to 7 are performed.

Images