Intelligent vehicle door self-adaptive opening and closing control method and system based on environment perception

Abstract

The invention discloses an intelligent vehicle door self-adaptive opening and closing control method and system based on environment perception, and relates to the technical field of intelligent vehicle door control. The method comprises the following steps: identifying and classifying objects by using an obstacle analysis model to obtain the types and corresponding priorities of the objects; performing collision prediction on an object and a vehicle door by using the collision prediction model to generate a preliminary vehicle door control strategy; performing vehicle door opening and closing simulation on the three-dimensional model of the surrounding environment by using the preliminary vehicle door control strategy, and adjusting parameters of the collision prediction model according to a simulation result to further obtain a final vehicle door control strategy; and performing opening and closing control on the vehicle door by using the final vehicle door control strategy. According to the invention, the surrounding environment of the vehicle door is identified, the speed and position relation analysis is carried out on a plurality of surrounding obstacles, and the self-adaptive active adjustment of the opening degree and the speed of the vehicle door is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent door control, and in particular to an intelligent door adaptive opening and closing control method and system based on environment perception. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] When opening traditional car doors in narrow parking spaces or complex environments, they are prone to collision with surrounding obstacles, causing damage to the door or safety hazards. Even for electric doors with radar, existing anti-collision solutions mostly rely on simple distance sensor warnings and hovering, lack the ability to actively adjust the opening and closing angles of the door, and cannot dynamically adapt to different environments.

[0004] In addition, the current method of controlling the opening and closing of car doors only relies on analyzing the dynamic and static states and sizes of surrounding obstacles, and cannot adjust the opening and closing of the car doors according to the moving speed of the obstacles. In a complex environment with multiple obstacles, the priority analysis of the obstacles is also unclear, and a reasonable door control solution cannot be obtained.

[0005] Therefore, how to adaptively adjust the door opening and closing control strategy according to the situation of multiple obstacles in a complex environment has become a technical problem that needs to be solved urgently in the existing technology. Summary of the invention

[0006] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an intelligent door adaptive opening and closing control method and system based on environmental perception, which can identify the surrounding environment of the door and analyze the speed and position relationship of multiple obstacles around it, so as to realize adaptive and active adjustment of the opening and speed of the door.

[0007] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0008] A first aspect of the present invention provides an intelligent door adaptive opening and closing control method based on environment perception, comprising the following steps:

[0009] Obtaining surrounding environment information, identifying objects that are obstacles in the surrounding environment information, and constructing a three-dimensional model of the surrounding environment;

[0010] Use the obstacle analysis model to identify and classify objects to obtain the type of objects and their corresponding priorities;

[0011] Use a collision prediction model to predict collisions between objects and vehicle doors, and generate a preliminary vehicle door control strategy. Specifically, re-adjust the priorities according to the moving speed, moving direction, and distance of the objects, predict the collision situation between each object and the vehicle door, and generate a preliminary vehicle door control strategy based on the adjusted priority order combined with all vehicle door collision prediction results;

[0012] Use the preliminary vehicle door control strategy to simulate the opening and closing of the vehicle door in a three-dimensional model of the surrounding environment, adjust the parameters of the collision prediction model according to the simulation results, and further obtain the final vehicle door control strategy;

[0013] Use the final vehicle door control strategy to control the opening and closing of the vehicle door.

[0014] The second aspect of the present invention provides an intelligent vehicle door adaptive opening and closing control system based on environmental perception, including:

[0015] A data acquisition module, configured to acquire surrounding environment information, determine objects serving as obstacles in the environment information, and construct a three-dimensional model of the surrounding environment;

[0016] An obstacle analysis module, configured to identify and classify objects using an obstacle analysis model to obtain the types of objects and corresponding priorities;

[0017] A collision prediction module, configured to use a collision prediction model to predict collisions between objects and vehicle doors, and generate a preliminary vehicle door control strategy. Specifically, re-adjust the priorities according to the moving speed, moving direction, and distance of the objects, predict the collision situation between each object and the vehicle door, and generate a preliminary vehicle door control strategy based on the adjusted priority order combined with all vehicle door collision prediction results;

[0018] An opening and closing simulation module, configured to use the preliminary vehicle door control strategy to simulate the opening and closing of the vehicle door in a three-dimensional model of the surrounding environment, adjust the parameters of the collision prediction model according to the simulation results, and further obtain the final vehicle door control strategy;

[0019] An opening and closing control module, configured to use the final vehicle door control strategy to control the opening and closing of the vehicle door.

[0020] The third aspect of the present invention provides a medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in the method for intelligent vehicle door adaptive opening and closing control based on environmental perception as described in the first aspect of the present invention.

[0021] The fourth aspect of the present invention provides a device, including a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for intelligent vehicle door adaptive opening and closing control based on environmental perception as described in the first aspect of the present invention.

[0022] The fifth aspect of the present invention provides a computer program product, including a computer program, which when executed by a processor implements the steps in the method for adaptively controlling the opening and closing of an intelligent vehicle door based on environmental perception as described in the first aspect of the present invention.

[0023] The above one or more technical solutions have the following beneficial effects:

[0024] The present invention discloses a method and a system for adaptively controlling the opening and closing of an intelligent vehicle door based on environmental perception. By arranging objects according to their shape, size, static and dynamic states, and moving speed, a control strategy for opening and closing the vehicle door is generated according to the priority order based on the collision prediction results of each object. The vehicle door opening and closing control strategy of the present invention is more reasonable, can adapt to complex scenarios with multiple objects, and will not cause conflicting control strategies due to the large number of objects.

[0025] The present invention uses an LSTM combined with a random forest algorithm to construct a collision prediction model. By introducing a dynamic weight mechanism to balance the parameters of the two models, the moving intention of dynamic objects is analyzed from the perspective of the influence relationship between static objects and dynamic objects, so as to obtain more accurate collision prediction results.

[0026] The advantages of the additional aspects of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0028] Figure 1 It is a flowchart of the method for adaptively controlling the opening and closing of an intelligent vehicle door based on environmental perception in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0030] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof;

[0031] Embodiment 1:

[0032] Embodiment 1 of the present invention provides an intelligent door adaptive opening and closing control method based on environmental perception, as Figure 1 shown, including the following steps:

[0033] Step 1: Obtain the surrounding environment information, determine the objects as obstacles in the environment information, and construct a three-dimensional model of the surrounding environment.

[0034] In a specific implementation manner, data acquisition devices are installed on the vehicle body. In this embodiment, the data acquisition devices include a door radar array, a camera, and ultrasonic sensors. The millimeter-wave radar is used to collect point cloud data, the camera is used to collect image data, and the ultrasonic sensors are used to calibrate the millimeter-wave radar.

[0035] The door radar array is composed of millimeter-wave radars, arranged on the inner side of the door, close to the door edge, and the recognition range includes 180° in the door opening direction.

[0036] The camera uses an existing electronic rearview mirror or arranges the camera below the outer rearview mirror of the door, taking into account both aesthetics and vision.

[0037] The ultrasonic sensors are arranged at the bottom of the door, and two ultrasonic sensors are arranged at the bottom of each door. It is mainly used to calibrate the distance data recognized by the millimeter-wave radar. Specifically, the millimeter-wave radar array recognizes the surrounding objects and calculates the azimuth and real-time distance changes of the objects located at the door in real time; the ultrasonic radar and the millimeter-wave radar work together, but the ultrasonic radar only measures the object distance, and the ultrasonic radar performs object distance calibration to detect whether the distances calculated by the millimeter-wave radar array and the ultrasonic radar are consistent; after calibration, a door coordinate system is established, and the data collected by the data acquisition devices is processed and unified into the door coordinate system using the coordinate system transformation method.

[0038] In this embodiment, the objects that can affect the opening and closing of the door are regarded as obstacles.

[0039] Step 2: Use the obstacle analysis model to identify and classify the objects to obtain the types of the objects and the corresponding priorities.

[0040] Step 2.1: Preprocess the data collected after calibrating the camera and the millimeter-wave radar.

[0041] In a specific implementation manner, for the image data collected by the camera, operations such as denoising, color correction, and scale adjustment are performed on the image to improve the image quality. For the data collected by the millimeter-wave radar, the radar data is parsed and screened, noise points are removed, and target information such as effective distance and speed is extracted.

[0042] Step 2.2: Construct an obstacle analysis model based on the object detection network, and use the obstacle analysis model to perform fusion analysis and judgment on the preprocessed data to obtain the recognition result of the object type with distance markings.

[0043] Step 2.2.1: Use the YOLOv5 network as the framework to construct an obstacle analysis model.

[0044] Step 2.2.2: Use the obstacle analysis model to perform object detection on the image data to identify the category and position of the objects in the image.

[0045] For the targets detected by the camera, the monocular ranging algorithm (such as the geometric-based method or the machine learning-based method) can be used to calculate the distance between the target and the camera according to information such as the size, shape, and parallax of the target in the image.

[0046] The above object detection and distance estimation algorithms are all existing algorithms in this field and will not be elaborated here.

[0047] Step 2.2.3: Align the image data and the radar data in time and space to ensure that the two data are synchronized in time and can correspond in space.

[0048] Step 2.2.4: Perform information fusion and matching on the radar data and the image data. Based on the door coordinate system, perform position fusion on the radar information and the image information, and fuse the target information detected by the camera and the target information detected by the millimeter-wave radar. For example, map the radar point cloud information onto the image as a new channel, and perform operations such as dilation on the radar point cloud to enhance its feature expression. Match the targets detected by the camera and the millimeter-wave radar, and fuse information such as the position, speed, and category of the targets according to the matching results to obtain a more accurate target description.

[0049] The above fusion process uses the existing information fusion algorithms in this field and will not be elaborated here.

[0050] Step 2.3: According to the recognition result of the object type, perform priority classification in combination with the corresponding distance markings.

[0051] In this embodiment, the priority is: human > other moving organisms > vehicle > other non-living objects, and moving objects > stationary objects.

[0052] Among them, for safety considerations, the vehicle is only a car, and humans include pedestrians and cyclists. For other non-living objects, including walls, curbs, and thrown objects, etc.

[0053] Step 3: Use the collision prediction model to perform collision prediction on the object and the door, and generate a preliminary door control strategy.

[0054] Among them, the priority is readjusted according to the moving speed, moving direction and distance of the object, and the collision situation between each object and the car door is predicted. Based on the adjusted priority order and combining all the car door collision prediction results, a preliminary car door control strategy is generated.

[0055] Step 3.1: Obtain the recognition results of the car door parameters and object types, and readjust the priority according to the moving speed, moving direction and distance of the object. Among them, the object with a faster approaching speed has a higher priority than the object with a slower approaching speed.

[0056] In this embodiment, the recognition results of the car door parameters and object types include: data such as the size, moving speed, direction and intention of the moving object, the distance from the car door, the size of the car door, the opening and closing angle of the car door and the range of the car door edge during the opening and closing process, the position and size of the static object, etc.

[0057] Specifically, the moving direction of the object is considered first, followed by the distance between the object and the car door, and finally the moving speed of the object. More specifically, the priority: the moving object with a fast moving speed, a close direction and a short distance > the moving object with a slow moving speed, a close direction and a short distance > the moving object with a fast moving speed, a close direction and a long distance > the object with a fast moving speed, a far direction and a short distance > the moving object with a slow moving speed, a close direction and a long distance.

[0058] Among them, the moving distance between the object and the car door refers to the distance from the closest point of the object to the car door to the farthest point of the car door from the vehicle body when the car door is opened to the maximum. The moving direction of the object is set as the approaching direction when the distance between the object and the car door shortens within a period of time, and the moving direction is set as the far direction when the distance between the object and the car door extends within a period of time.

[0059] Step 3.2: Use the collision prediction model to predict the collision between the object and the car door.

[0060] In this embodiment, an LSTM (Long Short-Term Memory Network) model and a random forest model are combined to form a collision prediction model. The LSTM model is used to capture the temporal relationship between dynamic features, and the random forest model is used to analyze static features to judge the collision risk between the object and the car door.

[0061] Step 3.2.1: Use the LSTM model to process the parameters of the moving object. Specifically, the LSTM is used to process the temporal data to capture the dynamic changes of the moving object and predict the future trajectory and position of the moving object.

[0062] Step 3.2.2: Use the random forest model to extract the static features of the static object and the car door, and perform collision recognition to obtain the collision risk between the static obstacle and the car door.

[0063] Step 3.2.3: Use the Kalman filter algorithm to predict the intention of the moving object. Specifically, based on data such as the positions of surrounding stationary objects, the speed and direction of the moving object, and in combination with the trajectory of the moving object, predict the intention of the moving object, such as whether to change lanes, decelerate, etc.

[0064] In this embodiment, the Kalman filter algorithm is used to further analyze and verify the intention of the moving object. The Kalman filter is a recursive algorithm used to estimate the state of a linear dynamic system. It recursively updates the state estimate by combining the prior knowledge of the system (system model) and the observed data, thereby minimizing the estimation error. Its core idea is to utilize the dynamic model and observation model of the system to obtain the optimal state estimate of the system through the optimal estimation theory.

[0065] Step 3.2.4: Fuse the outputs of the LSTM model and the random forest model, and combine with the intention prediction result to comprehensively judge whether a collision will occur and generate a preliminary door control strategy.

[0066] Step 3.2.4.1: First, according to the performance of the LSTM model and the random forest model, assign weights to them. The weights can be determined based on indicators such as the historical accuracy of the model and the performance on the validation set. However, conventional static indicators cannot comprehensively consider the complex relationship between the obstacle and the door. Therefore, in this embodiment, the cuckoo search algorithm is used to dynamically adjust the weights assigned to the LSTM model and the random forest model. The cuckoo search algorithm is an optimization algorithm based on natural selection and can be used to optimize model parameters. The specific steps for using the cuckoo search algorithm to optimize the weights of the LSTM model and the random forest model are as follows:

[0067] 1. Initialization.

[0068] Determine the objective function. Define a fitness function to evaluate the performance of the model combination. For example, the prediction error (such as the mean square error MSE) of the model combination can be used as the fitness function. The smaller the error, the higher the fitness.

[0069] Initialize the population. Randomly generate a set of initial weight combinations as bird nests. Each weight combination includes the weights of the LSTM model and the random forest model, and the sum of the weights is 1. For example, generate 10 different weight combinations, and each combination is a two-dimensional vector, such as [w LSTM , w RF .

[0070] Set the algorithm parameters. Determine parameters such as the population size N (i.e., the number of bird nests), the discovery probability P a (the probability that the bird nest owner discovers the foreign bird eggs, usually taken as about 0.25), and the maximum number of iterations T, etc.

[0071] 2. Iterative optimization.

[0072] Position update. For each weight combination, a new weight combination is generated according to the Lévy flight mechanism. Lévy flight is a random walk process, and its step size satisfies a heavy-tailed distribution, which helps to explore globally.

[0073] The specific update formula is:

[0074] new_nest = current_nest + step × Lévy(λ)

[0075] where new_nest is the new nest, current_nest is the current nest, step is the step size factor, and Lévy(λ) is a random number of the Lévy distribution.

[0076] Fitness evaluation. The fitness value of each new weight combination is calculated using a fitness function. For each weight combination, the outputs of the LSTM model and the random forest model are weighted and fused according to the weight combination, and then the prediction error of the fused result is calculated.

[0077] Select the optimal: Among all the current weight combinations, select the weight combination with the best fitness value and retain it for the next generation. If the fitness of the newly generated weight combination is better than the current combination, replace the current combination.

[0078] Parasitic nest update: For each weight combination, a random number r (r ∈ [0, 1]) is generated and compared with the discovery probability P a If r > P a , then randomly change the weight combination, that is, regenerate a new weight combination.

[0079] 3. Judgment of end condition.

[0080] Judge whether the end condition is met: If the maximum number of iterations T is reached or the fitness value no longer improves significantly, stop the iteration. Output the final optimal weight combination, that is, the best weight allocation of the LSTM model and the random forest model

[0081] This embodiment uses the cuckoo search algorithm to introduce long-distance jumps through the Lévy flight mechanism, which can effectively explore the global solution space and avoid falling into local optima. In the prediction of car door collisions, this means that the algorithm can more comprehensively consider the impact of different weight combinations on the collision prediction performance and find a better weight allocation scheme.

[0082] The cuckoo search algorithm maintains a good balance between global search and local search. By controlling parameters such as the discovery probability, the algorithm can flexibly switch between global exploration and local search. This enables the algorithm to quickly find potential optimal solutions when optimizing weight allocation and further optimize through local search, allowing the contribution degrees of moving objects and static objects in the vehicle door collision prediction process to be dynamically weighted according to the actual environment.

[0083] Step 3.2.4.2: Secondly, further correct the collision risk prediction value according to the intention prediction result.

[0084] Assume that the outputs of the LSTM model and the random forest model are P LSTM and P RF , and the weights are w LSTM and w RF , then the fused collision risk prediction value P fused can be expressed as:

[0085] P fused = w LsTM ·P LSTM + w RF ·P RF .

[0086] Where w LSTM + w RF = 1.

[0087] Secondly, use the intention prediction result as an additional adjustment factor to correct the fused collision risk prediction value. The adjustment factor is expressed as P intent , indicating the impact of the intention of the moving object (such as lane change, deceleration, etc.) on the collision risk.

[0088] After that, adjust the fused collision risk prediction value P final according to the intention prediction result. For example, if the intention prediction module determines that the moving object has a high intention to change lanes and the lane change direction may collide with the vehicle door, the weight of the collision risk can be increased; conversely, if the intention prediction module determines that the moving object has a deceleration intention, the weight of the collision risk can be decreased.

[0089] The specific adjustment formula can be expressed as:

[0090] P final = P fused ·(1 + α·P intent ).

[0091] Where α is an adjustment coefficient used to control the influence degree of the intention prediction result on the fusion result. If P intent represents a high collision intention, then α can take a positive value; conversely, if Pintent Indicates a lower collision intention, then α can take a negative value.

[0092] Finally, according to the adjusted collision risk prediction value P final , a threshold value (such as 0.5) is set to determine whether a collision will occur. If P final is greater than the threshold value, it is predicted that a collision will occur; otherwise, it is predicted that no collision will occur. And the angles and opening / closing speeds at which the car door will not collide are given to generate the corresponding preliminary car door control strategy.

[0093] The hybrid model structure of LSTM and random forest in this embodiment combines the advantages of deep learning and traditional machine learning. Through the combination of the cuckoo search algorithm and the intention prediction algorithm, the prediction results of the collision relationship between the car door and the object are double-corrected, and the complex relationship between the obstacle and the car door can be considered more comprehensively. For example, when the vehicle surrounding environment is a narrow space, the static object collision factor is considered to be larger; when the vehicle surrounding environment is a wide space, the moving object collision factor is considered to be larger; when the vehicle surrounding environment is a place with dense traffic or a large number of people, the result of intention prediction is mainly considered. The collision relationship prediction of the above-mentioned collision prediction model in this embodiment can adapt to a variety of complex scenarios and improve the accuracy of collision prediction. At the same time, through the intention prediction module, the behavior of the obstacle can be better understood, further improving the reliability of the prediction.

[0094] Step 4: Use the preliminary car door control strategy to simulate the opening and closing of the car door in the three-dimensional model of the surrounding environment, and adjust the parameters of the collision prediction model according to the simulation results to further obtain the final car door control strategy.

[0095] Step 4.1: Based on the constructed three-dimensional model of the surrounding environment, import the geometric parameters of the car door and the initial state of the vehicle into the simulation environment. Among them, the geometric parameters of the car door include the size of the car door, the opening / closing angle range, the maximum extension range of the car door edge, etc. The initial state of the vehicle includes the initial position of the vehicle and the initial angle of the car door. At the same time, the initial positions, speeds, directions and other parameters of the moving objects and static objects identified in Step 3 are also input into the three-dimensional model.

[0096] Step 4.2: According to the preliminary car door control strategy generated in Step 3, set the opening / closing angle and speed of the car door, and gradually simulate the opening and closing process of the car door. During the simulation process, the position and angle of the car door are updated in real time, and at the same time, according to the moving speed and direction of the object, the position of the object is dynamically updated. The simulation step size can be set according to actual needs. For example, the simulation is carried out with a time step of 0.1 second.

[0097] Step 4.3: In each time step of the simulation, detect whether there is a collision between the car door and surrounding objects. If a collision occurs, record the time, location of the collision, type of the colliding object, and related parameters (such as the speed and direction of the colliding object). Meanwhile, record key parameters such as the opening / closing angle and speed of the car door during the simulation.

[0098] And count the number of collisions that occur during the simulation, the types of colliding objects, and the state of the car door at the time of collision (such as the opening / closing angle and speed of the car door). Analyze which types of objects the collisions mainly concentrate on, and in which opening / closing states of the car door collisions are more likely to occur. For simulation scenarios without collisions, analyze whether the opening / closing strategy of the car door is too conservative, and whether there is room to appropriately adjust the opening / closing angle or speed of the car door to improve the convenience of using the car door.

[0099] Step 4.4: Adjust the parameters of the collision prediction model.

[0100] Step 4.4.1: Adjust the collision risk threshold. According to the predicted collision risk value at the time of collision in the simulation results, adjust the collision risk threshold. If it is found in the simulation that an object with a low predicted collision risk value also has a collision, it means that the current collision risk threshold is too high and needs to be appropriately lowered; conversely, if an object with a high predicted collision risk value does not have a collision in the simulation and the opening / closing strategy of the car door is too conservative, the collision risk threshold can be appropriately increased.

[0101] Step 4.4.2: Adjust the model weights. If the simulation results show that the collision prediction accuracy of certain types of objects (such as moving objects or stationary objects) is low, the weights of the LSTM model and the random forest model can be readjusted. For example, if it is found in the simulation that the collision prediction error of stationary objects is large, the weight of the random forest model can be appropriately increased; if the collision prediction error of moving objects is large, the weight of the LSTM model can be increased. The adjustment of the weights can refer to the optimization process of the cuckoo search algorithm to improve the prediction performance of the model by dynamically adjusting the weights.

[0102] Step 4.4.3: Adjust the intention prediction parameters. If the simulation results indicate that the correction effect of intention prediction on the collision risk is not ideal, the parameters of the intention prediction module can be adjusted. For example, adjust the value of the adjustment coefficient α to more accurately reflect the influence degree of the intention prediction result on the collision risk. If it is found in the simulation that the correction of the intention prediction result on the collision risk is too sensitive or not sensitive enough, the value range of α or other parameters in the formula can be adjusted according to the actual situation.

[0103] Step 4.4.4: Adjust the input features of the model. According to the simulation results, analyze whether the input features of the current model can fully reflect the collision relationship between the car door and the object. If it is found that some important features (such as the acceleration of the object, the acceleration of the car door, etc.) are not considered, these features can be added to the input of the model to improve the prediction accuracy of the model.

[0104] Step 4.5: After adjusting the parameters of the collision prediction model, use the new model parameters to re - conduct the simulation of the car door opening and closing. Compare the adjusted simulation results with the previous simulation results to verify whether the adjusted model can more accurately predict the collision relationship between the car door and the object, and whether it can generate a more reasonable car door control strategy. If the verification results show that the model performance has been improved, apply the adjusted model parameters to the final car door control strategy generation process; if the model performance does not meet the expected effect, continue to adjust and optimize the model parameters until satisfactory prediction accuracy and control effect are achieved.

[0105] In this embodiment, the preliminary car door control strategy is used to simulate the opening and closing of the car door in the three - dimensional model of the surrounding environment. The opening and closing process of the car door can be fully simulated and verified in the virtual environment. The collision prediction model is optimized and adjusted according to the simulation results, thereby further improving the accuracy and reliability of the car door control strategy and providing more powerful guarantee for the safe operation of the vehicle.

[0106] Step 5: Use the final car door control strategy to control the opening and closing of the car door.

[0107] In this embodiment, according to the car door control strategy, for an object approaching rapidly, if the distance is relatively close, it will prohibit opening the door and prompt the user that there is an object approaching rapidly. For an object with a slower speed, it will calculate the optimal opening angle and open the door at a certain speed. For a stationary object, it will open the door normally, and the opening speed will decrease before approaching the object to ensure stopping in front of the object. The specific opening strategy is based on the prediction results of the collision prediction model and is given based on the adjusted priority order, giving priority to the collision prediction results of objects with a higher priority, and considering all objects with collision risks. Finally, the most reasonable car door control strategy is obtained through comprehensive consideration.

[0108] Embodiment Two:

[0109] Embodiment Two of the present invention provides an intelligent car door adaptive opening and closing control system based on environmental perception, including:

[0110] A data acquisition module, configured to acquire surrounding environment information, determine the objects in the environment information that are obstacles, and construct a three - dimensional model of the surrounding environment;

[0111] An obstacle analysis module, configured to identify and classify an object using an obstacle analysis model to obtain the type of the object and the corresponding priority;

[0112] A collision prediction module, configured to perform a collision prediction on the object and the vehicle door using a collision prediction model to generate a preliminary vehicle door control strategy. Among them, the priority is readjusted according to the moving speed, moving direction, and distance of the object, and the collision situation between each object and the vehicle door is predicted. Based on the adjusted priority order and combined with all vehicle door collision prediction results, a preliminary vehicle door control strategy is generated;

[0113] An opening and closing simulation module, configured to perform a vehicle door opening and closing simulation in a three-dimensional model of the surrounding environment using the preliminary vehicle door control strategy, and adjust the parameters of the collision prediction model according to the simulation results to further obtain the final vehicle door control strategy;

[0114] An opening and closing control module, configured to perform an opening and closing control on the vehicle door using the final vehicle door control strategy.

[0115] Embodiment III:

[0116] Embodiment III of the present invention provides a medium, on which a program is stored. When the program is executed by a processor, it implements the steps in the environment perception-based intelligent vehicle door adaptive opening and closing control method described in Embodiment I of the present invention.

[0117] Embodiment IV:

[0118] Embodiment IV of the present invention provides a device, including a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the environment perception-based intelligent vehicle door adaptive opening and closing control method described in Embodiment I of the present invention.

[0119] Embodiment V:

[0120] Embodiment V of the present invention provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the steps in the environment perception-based intelligent vehicle door adaptive opening and closing control method described in Embodiment I of the present invention.

[0121] The steps involved in Embodiments II, III, IV, and V above correspond to those in Method Embodiment I. For specific implementation manners, reference may be made to the relevant description part of Embodiment I.

[0122] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device. Thus, they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0123] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications or deformations that can be made without creative efforts by those skilled in the art on the basis of the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. An intelligent door adaptive opening and closing control method based on environment perception, characterized in that: The following steps are involved: Obtaining surrounding environment information, identifying objects that are obstacles in the surrounding environment information, and constructing a three-dimensional model of the surrounding environment; Use the obstacle analysis model to identify and classify objects to obtain the type of objects and their corresponding priorities; Using the collision prediction model to predict the collision between the object and the door, a preliminary door control strategy is generated, wherein the priority is readjusted according to the object's moving speed, moving direction and distance, and the collision between each object and the door is predicted, and the preliminary door control strategy is generated based on the adjusted priority order and all the door collision prediction results; Using the preliminary door control strategy to simulate the door opening and closing in the 3D model of the surrounding environment, the collision prediction model parameters are adjusted according to the simulation results to further obtain the final door control strategy; The final door control strategy is used to control the opening and closing of the door.

2. The method for adaptively opening and closing intelligent door based on environment perception as claimed in claim 1, characterized in that: Data acquisition equipment is installed on the vehicle body to obtain surrounding environment information. Specifically, the data acquisition equipment includes a door radar array composed of millimeter-wave radar, a camera and an ultrasonic sensor. The millimeter-wave radar is used to collect point cloud data, the camera is used to collect image data, and the ultrasonic sensor is used to calibrate the millimeter-wave radar.

3. The method for adaptively opening and closing intelligent door based on environment perception as claimed in claim 2, characterized in that: The specific steps for identifying and classifying objects using the obstacle analysis model are: Preprocess the data collected after camera and millimeter wave radar calibration; An obstacle analysis model is built based on the target detection network, and the obstacle analysis model is used to perform fusion analysis and judgment on the pre-processed data to obtain the object type recognition result with distance mark; The obstacle analysis model is used to perform fusion analysis and judgment on the preprocessed data.

4. The method for adaptively opening and closing intelligent door based on environment perception as claimed in claim 1, characterized in that: The specific steps of using the obstacle analysis model to perform fusion analysis and judgment on the pre-processed data are as follows: The obstacle analysis model is constructed using the YOLOv5 network as the framework; Use the obstacle analysis model to detect targets on image data and identify the category and location of objects in the image; Align image data and radar data in time and space to ensure that the two data are synchronized in time and correspond in space; Information fusion and matching of radar data and image data.

5. The method for adaptively opening and closing intelligent door based on environment perception as claimed in claim 1, characterized in that: The LSTM model and the random forest model are combined to form a collision prediction model. The LSTM model is used to capture the temporal relationship between dynamic features, and the random forest model is used to analyze static features to determine the collision risk between objects and car doors.

6. The method for adaptively opening and closing intelligent door based on environment perception as claimed in claim 5, characterized in that: The specific steps of using the collision prediction model to predict the collision between an object and a car door are as follows: Use the LSTM model to process the parameters of moving objects; The random forest model is used to extract static features of stationary objects and car doors, and collision recognition is performed to obtain the collision risk between static obstacles and car doors. Use Kalman filter algorithm to predict the intention of moving objects; The outputs of the LSTM model and the random forest model are fused and combined with the intention prediction results to comprehensively judge whether a collision will occur and generate a preliminary door control strategy.

7. An intelligent door adaptive opening and closing control system based on environmental perception, characterized in that: include: A data acquisition module is configured to acquire surrounding environment information, determine objects in the surrounding environment information that are obstacles, and construct a three-dimensional model of the surrounding environment; An obstacle analysis module is configured to identify and classify objects using an obstacle analysis model to obtain the type of the object and the corresponding priority; A collision prediction module is configured to use a collision prediction model to predict collisions between objects and vehicle doors, and generate a preliminary vehicle door control strategy, wherein the priority is readjusted according to the object's moving speed, moving direction, and distance, and a collision situation between each object and the vehicle door is predicted, and the preliminary vehicle door control strategy is generated based on the adjusted priority order combined with all vehicle door collision prediction results; an opening and closing simulation module, configured to use the preliminary door control strategy to perform door opening and closing simulation in a three-dimensional model of the surrounding environment, adjust the collision prediction model parameters according to the simulation results, and further obtain a final door control strategy; The opening and closing control module is configured to use the final door control strategy to control the opening and closing of the door.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for adaptive opening and closing control of an intelligent vehicle door based on environmental perception as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: A plurality of instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executing the environment-aware intelligent vehicle door adaptive opening and closing control method described in any one of claims 1 to 6.

10. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the environmental perception-based intelligent door adaptive opening and closing control method described in any one of claims 1-6.

Images