Parking function adaptive calling method and device, equipment and medium

Abstract

The application relates to a parking function adaptive calling method, device, equipment and medium. The method comprises the following steps: obtaining multi-modal scene information containing driver state information, outputting an intention probability vector of each parking function at the current moment by using a time sequence intention prediction model, determining and activating a first parking function, and saving a current vehicle pose and a remaining planning path and transferring the execution state to a second parking function to continue completing a remaining parking task if it is detected that the user intention is switched to the second parking function in the execution process. Thus, time sequence prediction of implicit parking intention of a user is realized, parking functions can be actively activated or switched in appropriate scenes without manual operation of the user, task relay between different parking functions is realized, the problem that a narrow parking space and the like cannot be continuously completed by segmented operation is solved, intelligent, non-sensing and continuous calling of the parking function is realized, and user experience is improved.

Description

Technical Field

[0001] This application relates to the field of vehicle driver assistance technology, and in particular to a method, device, equipment and medium for adaptive invocation of parking function. Background Technology

[0002] With the development of autonomous driving technology, more and more vehicles are equipped with various parking functions, including Automatic Parking Assist (APA), Remote Parking Assist (RPA), and Autonomous Valet Parking (AVP). These functions provide users with diverse parking options to meet parking needs in different scenarios. However, with multiple parking functions coexisting, how to intelligently invoke and switch parking functions according to the user's actual needs still faces many technical challenges.

[0003] In existing technologies, parking function invocation mainly relies on explicit user commands or simple rule judgments, which is a passive, responsive interaction. Users need to actively think and choose which function to use before parking, resulting in a heavy operational burden. While some solutions achieve automatic invocation of parking functions based on vehicle status information, they depend on explicit rule triggering mechanisms and can only respond to the vehicle's current state, failing to proactively predict the user's deeper parking intentions. Furthermore, most solutions only utilize vehicle signals (such as gear position and speed) and environmental perception information (such as parking space detection results) for decision-making, neglecting the crucial value of driver status information in intention judgment. More importantly, in existing technologies, each parking function mode is independent and lacks collaborative relay capabilities, making it impossible to achieve seamless switching and task continuation in continuous operation scenarios such as "parking from inside the car first, then remote control from outside the car." Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method, apparatus, device, and medium for adaptive invocation of parking functions.

[0005] Firstly, this application provides a method for adaptively invoking parking functions, including: The pre-trained temporal intent prediction model performs intent prediction processing on the acquired multimodal scene information to obtain the intent probability vectors corresponding to various parking functions at the current time. The multimodal scene information includes driver state information, vehicle state information, environmental perception information, and external interaction information. The first parking function is determined and activated based on the intent probability vector to execute the corresponding parking task; During the execution of the first parking function, when the user intent to switch to the second parking function is detected based on the continuously acquired multimodal scene information and the second parking function is available, the current execution state of the first parking function is saved. The current execution state includes at least the current vehicle pose and the remaining planned path. Invoke the second parking function and pass the current execution status to the second parking function to continue executing the remaining parking tasks.

[0006] Secondly, this application provides a parking function adaptive recall device, comprising: The first processing module is used to perform intent prediction processing on the acquired multimodal scene information based on a pre-trained temporal intent prediction model to obtain the intent probability vectors corresponding to various parking functions at the current time. The multimodal scene information includes driver state information, vehicle state information, environmental perception information, and external interaction information. The first execution module is used to determine and activate the first parking function based on the intent probability vector in order to execute the corresponding parking task. The second processing module is used to save the current execution state of the first parking function when, during the execution of the first parking function, it is detected that the user intends to switch to the second parking function and the second parking function is available based on the continuously acquired multimodal scene information. The current execution state includes at least the current vehicle pose and the remaining planned path. The second execution module is used to call the second parking function and pass the current execution status to the second parking function to continue executing the remaining parking tasks.

[0007] Thirdly, this application provides a parking function adaptive recall device, comprising: processor; Memory, used to store executable instructions; The processor is used to read executable instructions from memory and execute the executable instructions to implement the adaptive invocation method for the parking function in the first aspect.

[0008] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the parking function adaptive invocation method of the first aspect.

[0009] The technical solution provided in this application has the following advantages compared with the prior art: The adaptive parking function invocation method of this application acquires multimodal scene information, including driver state information, and uses a temporal intent prediction model to output the intent probability vectors of various parking functions at the current moment. Based on this, the first parking function is determined and activated. If the user's intent to switch to the second parking function is detected during execution, the current vehicle pose and remaining planned path are saved, and the execution state is passed to the second parking function to continue completing the remaining parking task. Thus, by integrating in-vehicle multimodal information such as driver's gaze, head posture, and hand position, temporal prediction of the user's implicit parking intent is achieved. The parking function can be actively activated or switched in appropriate scenarios without manual operation by the user. At the same time, through the state saving and transmission mechanism, different parking functions can relay tasks, solving the problem of segmented operation that cannot be completed continuously in scenarios such as narrow parking spaces. This achieves intelligent, seamless, and continuous invocation of parking functions, improving the user experience. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0011] Figure 1 A flowchart illustrating an adaptive parking function invocation method provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a time-series intent prediction model provided in an embodiment of this application; Figure 3 A flowchart illustrating a relay parking method provided in an embodiment of this application; Figure 4 A flowchart illustrating another adaptive parking function invocation method provided in this application embodiment; Figure 5 A schematic diagram of a parking function adaptive calling device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a parking function adaptive calling device provided in an embodiment of this application. Detailed Implementation

[0012] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0013] It should be understood that the various steps described in the method implementation of this application may be performed in different orders and / or in parallel. Furthermore, the method implementation may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0014] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0015] It should be noted that the concepts of "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0016] It should be noted that the terms "a" and "a plurality of" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0017] The names of messages or information exchanged between multiple devices of the implementers of this application are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0018] To address the aforementioned problems, embodiments of this application provide a method, apparatus, device, and medium for adaptive invocation of parking functions. The following, in conjunction with… Figure 1-4 The adaptive invocation method for parking function provided in the embodiments of this application will be described in detail.

[0019] Figure 1 The diagram shows a flowchart of an adaptive parking function invocation method provided in an embodiment of this application.

[0020] In this embodiment, the adaptive parking function invocation method can be executed by an electronic device. This electronic device may include, but is not limited to, devices such as in-vehicle infotainment systems, computer devices, cloud servers, or cloud server clusters.

[0021] like Figure 1 As shown, the adaptive invocation method for the parking function may include the following steps.

[0022] S110. The pre-trained temporal intent prediction model performs intent prediction processing on the acquired multimodal scene information to obtain the intent probability vectors corresponding to various parking functions at the current time.

[0023] In this embodiment of the application, the electronic device can acquire multimodal scene information and perform intent prediction processing on the multimodal scene information based on a pre-trained temporal intent prediction model to obtain the intent probability vectors corresponding to various parking functions at the current time.

[0024] Optionally, the multimodal scene information includes driver status information, vehicle status information, environmental perception information, and external interaction information. Driver status information is collected via the Driver Monitoring System (DMS) camera, including but not limited to the driver's gaze focus, head posture, hand position, and seatbelt status. Vehicle status information is collected via the Controller Area Network (CAN) bus, including but not limited to vehicle speed, gear position, accelerator / brake pedal opening, steering wheel angle, and steering wheel angular velocity. Environmental perception information is collected via ultrasonic radar and / or surround-view cameras, including but not limited to parking space type, parking space size, obstacle distribution, and Automated Valet Parking (AVP) area positioning. External interaction information is collected via Bluetooth module, including but not limited to Bluetooth Received Signal Strength Indication (RSSI) values ​​and mobile application connection status.

[0025] Specifically, the pre-trained temporal intent prediction model employs a three-layer Transformer encoder structure, such as a Transformer encoder network or a Hidden Markov Model (HMM). The electronic device extracts multimodal scene information collected within a preset duration (e.g., 3 seconds) using the current time as the endpoint, constructing a multimodal feature sequence within a continuous time window. The feature vector Ft at each sampling time t has a dimension of M=64, including 12 dimensions of vehicle signal features (vehicle speed, gear encoding, pedal opening, etc.), 24 dimensions of environmental perception features (parking parameters, obstacle distance matrix, AVP sign, etc.), 20 dimensions of driver state features (eye gaze coordinates, head posture angle, hand position encoding, etc.), and 8 dimensions of external interaction features (Bluetooth RSSI, application status, etc.).

[0026] The multimodal feature sequence is input into the temporal intent prediction model (e.g., the feature sequence input into the model is: {Ft} 29, Ft 28 ... Ft}, with a shape of 30×64, after model processing, the intent probability vectors corresponding to various parking functions at the current time are output through an activation function (such as the Softmax activation function). The intent probability vectors include at least two of the following: Automated Parking Assist (APA) intent probability vector, Remote Parking Assist (RPA) intent probability vector, Autonomous Valet Parking (AVP) intent probability vector, and manual parking intent probability vector.

[0027] S120. Determine and activate the first parking function based on the intent probability vector to execute the corresponding parking task.

[0028] In this embodiment of the application, the electronic device can determine and activate the first parking function based on the intent probability vector to perform the corresponding parking task.

[0029] Specifically, the electronic device determines a first parking function from a variety of parking functions based on the intent probability vector, and activates that function to perform the corresponding parking task. The variety of parking functions includes at least two of automatic parking assistance, remote parking assistance, and autonomous valet parking.

[0030] S130. During the execution of the first parking function, when the user intends to switch to the second parking function and the second parking function is available based on the continuously acquired multimodal scene information, the current execution state of the first parking function is saved.

[0031] In this embodiment of the application, during the execution of the first parking function, when the user intends to switch to the second parking function and the second parking function is available based on the continuously acquired multimodal scene information, the electronic device can save the current execution state of the first parking function.

[0032] Specifically, during the execution of the first parking function, the electronic device continuously acquires multimodal scene information and periodically updates the intent probability vector. When it detects, based on the continuously acquired multimodal scene information, that the user intent is to switch from the first parking function to the second parking function, and the second parking function is available, a relay process is triggered, i.e., switching from the first parking function to the second parking function. The electronic device can save the current execution state of the first parking function, which includes at least the current vehicle pose and the remaining planned path.

[0033] S140. Invoke the second parking function and pass the current execution status to the second parking function to continue executing the remaining parking tasks.

[0034] In this embodiment of the application, the electronic device can invoke the second parking function to pass the current execution state to the second parking function so as to continue to execute the remaining parking task.

[0035] Specifically, after determining to proceed with the relay process, the electronic device invokes the second parking function and transmits the current execution status to the second parking function. The second parking function continues to execute the remaining parking task from the received current execution status, i.e., the vehicle's position and remaining planned path when the first parking function stopped, without needing to rescan the parking space or replan the complete path.

[0036] Therefore, by acquiring multimodal scene information, including driver state information, and using a temporal intent prediction model to output the intent probability vectors of various parking functions at the current moment, the first parking function is determined and activated. If the user's intent to switch to the second parking function is detected during execution, the current vehicle pose and remaining planned path are saved, and the execution state is passed to the second parking function to continue completing the remaining parking task. Thus, through multimodal scene information fusion, real-time perception of the driver's state is achieved, providing a richer information foundation for parking intent prediction. The temporal intent prediction model can predict the user's parking needs before they make an operation, enabling seamless adaptive invocation of parking functions. The state saving and passing mechanism enables task relay between different parking functions, solving the technical shortcomings of existing technologies where functions are independent and cannot be operated continuously. In complex scenarios such as narrow parking spaces, users can first use APA to park inside the car, and then get out of the car to use RPA to complete the remaining parking actions. The entire process is seamlessly connected, significantly improving the user experience.

[0037] Optionally, prior to S110, the adaptive parking function invocation method may further include: synchronously collecting the driver state information, the vehicle state information, the environmental perception information, and the external interaction information at a preset sampling period to obtain the multimodal scene information; and aligning the timestamp error of each multimodal scene information with time using a precise time protocol so that the timestamp error of each multimodal scene information is less than a preset timestamp threshold.

[0038] In this embodiment of the application, the electronic device can synchronously collect the driver status information, the vehicle status information, the environmental perception information, and the external interaction information at a preset sampling period to obtain the multimodal scene information.

[0039] Specifically, the electronic device collects driver status information through the driver monitoring system camera at a sampling rate of 30 frames per second, collects vehicle status information through the controller area network bus at a period of 100 milliseconds, collects environmental perception information through ultrasonic radar and surround view cameras, and collects external interaction information through the Bluetooth module. Each sensor operates independently according to its own physical sampling frequency, and the electronic device reads the latest data from each sensor at a preset system decision period (e.g., 100 milliseconds). Each data acquisition channel operates independently to ensure that all modal information is collected with a unified time reference.

[0040] Furthermore, the electronic device can time-align the multimodal scene information using a precise time protocol, so that the timestamp error of each multimodal scene information is less than a preset timestamp threshold.

[0041] Specifically, the multimodal scene information is time-aligned using a Precision Time Protocol (PTP). For example, the system clock is used as the master clock, and each sensor and acquisition module acts as a slave clock, with clock synchronization achieved through the PTP. After time alignment, the timestamp error of each modal information is less than 10 milliseconds. The time-aligned multimodal scene information is then input into the time-series intent prediction model, ensuring that the modal features input to the model are consistent in time.

[0042] Therefore, synchronous acquisition with a preset sampling period ensures the temporal correspondence of information from each modality, providing a reliable data foundation for subsequent time-series modeling. Time alignment using a precise time protocol ensures accurate temporal correspondence among features in the multimodal feature sequence, avoiding intentional prediction deviations caused by time misalignment. Timestamp errors are controlled within 10 milliseconds, meeting the stringent real-time requirements of dynamic vehicle scenarios.

[0043] Optionally, the adaptive invocation method for the parking function may further include: obtaining the maximum intent probability vector among the intent probability vectors; if the maximum intent probability vector is less than a preset first threshold, then keeping the current parking function state unchanged; if there is a unique candidate parking function whose intent probability vector is greater than the preset trigger threshold corresponding to the candidate parking function, then determining the candidate parking function as the user intent function; if the intent probability vectors corresponding to multiple candidate parking functions are all greater than their respective preset trigger thresholds, then determining the candidate parking function with the largest intent probability vector among the multiple candidate parking functions as the user intent function.

[0044] In this embodiment of the application, the electronic device can obtain the maximum intent probability vector among the intent probability vectors.

[0045] Specifically, electronic devices can obtain the maximum intent probability vector max(P) from the intent probability vector P = [pAPA,pRPA, pAVP, pManual, pNone] output by the temporal intent prediction model, where each component satisfies Σpi = 1.

[0046] Furthermore, if the maximum intent probability vector is less than a preset first threshold, the current parking function state remains unchanged.

[0047] Specifically, the electronic device can determine whether the maximum intent probability vector is less than a preset first threshold, where the default value of the preset first threshold θmin is 0.5. If the maximum intent probability vector is less than the preset first threshold (i.e., max(P) < 0.5), the user's intent is determined to be unclear, and the electronic device maintains the current parking function state unchanged. If the user is currently performing a parking function, it continues to perform the function; if no parking function is currently activated, no new function is activated.

[0048] Furthermore, if there exists a unique candidate parking function whose intent probability vector is greater than the preset trigger threshold corresponding to the candidate parking function, then the candidate parking function is determined as the user intent function.

[0049] Specifically, if the maximum intent probability vector is greater than or equal to a preset first threshold, the electronic device further determines whether there exists a unique candidate parking function whose intent probability vector is greater than the preset trigger threshold corresponding to that candidate parking function. The preset trigger thresholds θ for each parking function are: Automatic Parking Assist (APA) function (θ=0.60, adjustable range 0.50-0.75), Remote Parking Assist (RPA) function (θ=0.70, adjustable range 0.60-0.85), Autonomous Valet Parking (AVP) function (θ=0.70, adjustable range 0.60-0.85), and Manual Parking (θ=0.60, adjustable range 0.50-0.80).

[0050] If there exists a unique candidate parking function whose intent probability vector P is greater than the preset trigger threshold θ, meaning only one function's intent probability vector exceeds its corresponding trigger threshold, then the candidate parking function is determined as the user's intent function. For example, if the intent probability vector P corresponding to the Automatic Parking Assist (APA) function is greater than θ (0.60), and the intent probability vectors of other functions do not exceed their respective trigger thresholds, then the electronic device determines APA as the user's intent function.

[0051] Furthermore, if the intent probability vectors corresponding to multiple candidate parking functions are all greater than their respective preset trigger thresholds, that is, the intent probability vectors of multiple functions simultaneously exceed their corresponding trigger thresholds, then the candidate parking function with the largest intent probability vector value among the multiple candidate parking functions is determined as the user intent function.

[0052] For example, the intention probability vector P (0.65) corresponding to the Automatic Parking Assist (APA) function is greater than θ (0.60), and the intention probability vector P (0.75) corresponding to the Remote Parking Assist (RPA) function is greater than θ (0.70). Since the intention probability vectors of both functions exceed their respective trigger thresholds, the electronic device compares the probability values ​​of the two functions and determines the RPA function with the larger probability value as the user's intention function.

[0053] Therefore, by setting a first threshold, low-confidence classification noise is effectively filtered out. Subsequent decisions are only made when the system is sufficiently confident in the user's intent, thus avoiding false triggers. Different trigger thresholds are set for each parking function to consider the different confidence requirements of different functions. When multiple intents exceed the threshold simultaneously, the principle of maximizing probability is used for selection, ensuring that the most likely user intent is given priority.

[0054] Optionally, S120 may specifically include: calculating the decision score of each parking function based on the intent probability vector corresponding to the user intent function, the availability conditions corresponding to each parking function, and the preset function priority coefficient; and determining the parking function with the highest decision score as the first parking function and activating it.

[0055] In this embodiment, the electronic device can calculate the decision score of each parking function based on the intent probability vector corresponding to the user intent function, the availability conditions corresponding to each parking function, and the preset function priority coefficient.

[0056] Specifically, the electronic device can obtain the intent probability vector corresponding to the user's intent function. For example, if the user's intent function is RPA, then the corresponding P value is obtained. At the same time, the electronic device determines the availability flag Available(f) of each parking function according to the availability conditions corresponding to each parking function.

[0057] For example, the availability conditions for each parking function are as follows.

[0058] The availability conditions for the Automatic Parking Assist (APA) function include: a valid parking space has been detected (the parking space type is not empty, and the parking space width is ≥ 1.2 times the vehicle width); the path between the vehicle and the valid parking space is reachable (without static obstacles); and the system has no fault codes (DTC, Diagnostic Trouble Code). When all three conditions are met, Available_APA = 1 (available); otherwise, Available_APA = 0 (unavailable).

[0059] The availability conditions for Remote Parking Assist (RPA) include: a safe environment around the vehicle (obstacles in all directions are more than 0.3 meters away); a normal Bluetooth connection (Bluetooth RSSI value greater than -70dBm and connection maintained); and an authorized and responsive mobile application. If all three conditions are met, Available_RPA = 1 (available); otherwise, Available_RPA = 0 (unavailable).

[0060] The availability conditions for the Autonomous Valet Parking (AVP) function include: the vehicle is located within the autonomous valet parking geofence; the autonomous valet parking map data is valid and loaded; and the system has no fault codes. When all three conditions are met, Available_AVP = 1 (available); otherwise, Available_AVP = 0 (unavailable).

[0061] The electronic device calculates the decision score Score(f) for each parking function based on the intent probability vector P(f) corresponding to each parking function, the available flag Available(f), and the preset function priority coefficient Priority(f), using the following formula: Score(f) = P(f) × Available(f) × Priority(f). The default configuration of the function priority coefficient Priority(f) is as follows: Autonomous Valet Parking (AVP): Priority(AVP) = 1.2 (Valet parking has the highest priority); Remote Parking Assist (RPA): Priority(RPA) = 1.0; Automatic Parking Assist (APA): Priority(APA) = 0.9; Manual Parking: Priority(Manual) = 0.8 (Manual parking has the lowest priority).

[0062] Furthermore, the electronic device compares the decision scores (Score(f)) of each parking function, identifies the parking function with the highest decision score as the first parking function, and activates it.

[0063] Specifically, if the decision score (Score(f)) of any parking function f is greater than the preset decision threshold (Score_threshold, default value is 0.5), then the parking function with the highest decision score is selected as the first parking function and activated. If the decision score (Score(f)) of all parking functions is less than or equal to the decision threshold (Score_threshold), then a silent mode is entered, and the electronic device does not activate any automatic parking functions, only providing parking image assistance (such as 360° surround view image and parking radar prompts).

[0064] Therefore, a three-dimensional weighted decision-making mechanism based on intent probability vectors, availability flags, and priority coefficients was established, enabling intelligent function arbitration when multiple parking modes coexist. When a user's intended function becomes unavailable due to objective conditions, the system can automatically downgrade to other available functions, ensuring that users can still obtain parking assistance services in most cases. Through decision thresholds and a silent mode, the safety risks associated with forcibly activating parking functions when conditions are insufficient are avoided.

[0065] Optionally, S130 may specifically include: during the execution of the first parking function, continuously acquiring the multimodal scene information and periodically updating the intent probability vector; when the intent probability vector corresponding to the second parking function exceeds a preset trigger threshold, and the difference between the intent probability vector corresponding to the second parking function and the intent probability vector corresponding to the first parking function is greater than a preset backlash coefficient, determining that the user intent is to switch from the first parking function to the second parking function.

[0066] In this embodiment of the application, during the execution of the first parking function, the electronic device can continuously acquire the multimodal scene information and periodically update the intent probability vector.

[0067] Specifically, the electronic device acquires new multimodal scene information in each sampling period (e.g., 100 milliseconds) and inputs the continuously acquired multimodal scene information into the temporal intent prediction model, periodically updating the intent probability vector at the current moment and tracking changes in user intent in real time.

[0068] Furthermore, the electronic device determines whether the intent probability vector corresponding to the second parking function exceeds its corresponding preset trigger threshold. For example, if the second parking function is RPA, it determines whether the intent probability vector P is greater than θ (0.70). If the intent probability vector corresponding to the second parking function exceeds the preset trigger threshold, the electronic device further determines whether the difference between the intent probability vector corresponding to the second parking function and the intent probability vector corresponding to the first parking function is greater than a preset hysteresis coefficient (default value is 0.15).

[0069] When both of the above conditions are met simultaneously, that is, when the probability vector of the intent corresponding to the second parking function exceeds the preset trigger threshold, and the difference between the probability vector and the probability vector of the intent corresponding to the first parking function is greater than the preset backlash coefficient, the electronic device determines that the user's intent is to switch from the first parking function to the second parking function.

[0070] For example, the vehicle is currently performing APA (First Parking) function, with the current intent probability vector being P = [P_APA=0.70, P_RPA=0.30 ...]. During APA execution, the user starts looking for their phone to get out of the car. The DMS camera detects the user repeatedly looking towards the phone and outside the car, and the Bluetooth RSSI value begins to decay. The electronic device continuously updates the intent probability vector, resulting in P = [P_APA=0.55, P_RPA=0.75 ...]. At this point, P_RPA=0.75>θ_RPA=0.70, and P_RPA - P_APA = 0.75 - 0.55 = 0.20>0.15 (preset hysteresis coefficient). When both conditions are met, the electronic device determines that the user's intent is to switch from APA to RPA.

[0071] Therefore, by periodically updating the intent probability vector, real-time tracking of changes in user intent is achieved, enabling timely capture of user intent to switch from one parking mode to another. By setting a preset trigger threshold, it is ensured that switching is only triggered when the user's intent for the second function is sufficiently clear, avoiding frequent false triggers caused by probability fluctuations.

[0072] Optionally, S130 may specifically include: storing the current vehicle pose and the remaining planned path in a shared memory area, wherein the current vehicle pose information includes the vehicle's x-coordinate, y-coordinate and heading angle in the global coordinate system, and the remaining planned path includes the remaining path point sequence and the current path point index; and recording the timestamp of the relay moment so that the second parking function can confirm the timeliness of the current execution state based on the timestamp.

[0073] In this embodiment of the application, the electronic device can store the current vehicle pose and the remaining planned path in a shared memory area.

[0074] Optionally, the current vehicle pose information includes the vehicle's x-coordinate, y-coordinate, and heading angle in the global coordinate system, and the remaining planned path includes a sequence of remaining path points and a current path point index. For example, the current vehicle pose information includes: the vehicle's x-coordinate (in meters) in the global coordinate system; the vehicle's y-coordinate (in meters) in the global coordinate system; and the vehicle's yaw angle (in radians). The remaining planned path includes: a sequence of remaining path points path = [point_1, point_2 ... point_n], where each path point contains coordinates and desired speed information; and a current path point index current_index, indicating the path point position reached by the first parking function.

[0075] Furthermore, the electronic device can record a timestamp (timestamp_ms) of the relay moment and store this timestamp along with the vehicle's pose and the remaining planned path. When the second parking function is invoked, it uses the timestamp to determine the validity of the current execution state. If the timestamp exceeds a preset validity period (e.g., 5 seconds), the second parking function determines that the relay state has expired and re-performs environmental perception and path planning; if it is within the validity period, it directly uses the stored state to perform the relay.

[0076] For example, in the scenario of APA relaying to RPA, the vehicle is performing APA automatic parking, and the planning module calculates that the remaining space is extremely small, requiring multiple maneuvers to continue parking from inside the vehicle. At this time, the electronic equipment detects the user's intention to switch from APA to RPA. The electronic equipment performs the following operations: The APA controller saves the current vehicle position and the remaining planned path sequence to the relay buffer. The APA controller issues a command to keep the vehicle stationary, activating the automatic parking function. A pop-up window on the central control screen prompts the user that "the parking space is narrow, it is recommended to use remote parking," with two options: "Switch to remote parking" and "Continue manual parking," automatically selecting the former after a 5-second countdown. The vehicle's infotainment system sends an RPA wake-up command to the mobile app via Bluetooth, and the mobile app automatically switches to the RPA control interface. The system uses seat sensors, door status sensors, and Bluetooth RSSI signals to determine that the driver has exited the vehicle. After the driver confirms on the mobile app, the RPA controller reads the relay status from the relay buffer and executes the remaining parking path at a speed of v_max ≤ 1.5 km / h to complete the parking maneuver.

[0077] When performing relay parking from APA to RPA, the following relay trigger conditions must be met: the APA function is in progress and the vehicle is stationary (|v|<0.1 km / h for ≥1 second); the intention prediction module outputs p_RPA>θ_RPA (default θ_RPA = 0.70); the RPA function availability flag Available_RPA = TRUE; and the APA planning module estimates the remaining path with at least two parking maneuvers.

[0078] Therefore, by sharing memory to store the relay status, high-speed data exchange between different parking function modules is achieved, with a relay latency of less than 50 milliseconds, ensuring smooth switching. By storing the current vehicle position and remaining planned path, the second parking function does not need to rescan parking spaces or replan the complete path; it can directly continue execution from the stopping position of the first parking function, achieving true task continuation rather than restarting. By recording the relay timestamp, the second parking function can determine the timeliness of the relay status and re-perform environmental perception and path planning when the status expires, ensuring safety. This solves the technical defects of existing solutions where functions are independent and cannot operate continuously, achieving a complete closed loop for parking tasks in complex scenarios such as narrow parking spaces and underground parking garage turnstiles.

[0079] Optionally, the adaptive invocation method for the parking function may further include: during the execution of the first parking function, when the target parking space is detected to be occupied or the planned path is blocked by an obstacle based on the environmental perception information, pausing the current parking task, outputting a prompt message to the user, and keeping the parking image assistance function active.

[0080] In this embodiment, during the execution of the first parking function, the electronic system can continuously monitor the surrounding environment through environmental perception information. When one of the following abnormal situations is detected based on the environmental perception information, a degradation switching process is triggered: the target parking space is occupied, and when the vehicle arrives near the target parking space, the perception module detects that there are other vehicles or obstacles in the target parking space; the planned path is blocked by obstacles, and during the vehicle's journey along the planned path, the perception module detects that the path is blocked by temporary obstacles and the detour path is unavailable; or an AVP-related sensor malfunction is detected.

[0081] Furthermore, the electronic device can pause the current parking task, output prompts to the user, and keep the parking image assistance function active.

[0082] Specifically, the electronic system controls the vehicle to stop safely at its current location, activating the automatic parking function. It then displays prompts to the user via the central control screen or instrument panel, such as "Valet parking path obstructed, target parking space occupied, switched to parking assist mode." The system maintains the 360° surround view camera and parking radar assist functions, providing the driver with real-time images of the surrounding environment and obstacle distance prompts. Finally, it returns vehicle control to the driver for manual intervention or to search for a new parking space.

[0083] For example, a vehicle is using the AVP valet parking function and approaches a target parking space. At this point, the perception module detects that the target parking space is occupied by a temporarily parked vehicle. The electronic device performs the following actions: the AVP controller issues a safe parking command, the vehicle smoothly stops at the current location, and the automatic parking function is activated. The central control screen displays the message "Valet parking path obstructed, target parking space occupied, switched to parking assist mode." The electronic device keeps the 360° surround view and parking radar assist functions active, providing the driver with real-time images of the surrounding environment. The electronic device returns vehicle control to the driver, who can then manually intervene. Optionally, the electronic device recalculates the intent probability vector based on the current environment; if the APA conditions are met, it proactively recommends the APA function.

[0084] Therefore, when AVP cannot complete the task due to an occupied parking space or obstructed path, the system can safely disengage, avoiding safety hazards caused by prolonged waiting or attempts to detour. By promptly providing users with prompts, the system ensures they understand the current situation, facilitating correct subsequent operational decisions. Maintaining the parking image assist function provides ample environmental awareness support for manual operation after user takeover. Compared to direct emergency braking, this application's downgrade transition process is smoother, enhancing user safety and overall experience.

[0085] Therefore, this application models implicit user behavior using a temporal intent prediction model. Without requiring users to actively click buttons or input commands, it proactively recommends or activates the corresponding parking function in appropriate scenarios, significantly reducing the user's operational burden. A three-dimensional weighted function arbitration mechanism based on intent probability, availability flags, and priority coefficients is established, enabling intelligent decision-making across multiple modes such as APA, RPA, AVP, and manual driving, ensuring the correct function is used at the right time. The proposed state saving and transmission mechanism enables task relay between APA and RPA, achieving a complete closed loop of parking inside the vehicle first, followed by remote control from outside, in narrow parking space scenarios. Driver state information (eye focus, head posture, hand position, and seatbelt status acquired by the DMS camera) is deeply integrated with vehicle signals, environmental perception, and external interaction information to construct a multi-dimensional basis for user intent judgment. Frequent function switching oscillations are prevented through hysteresis coefficients, accidental function triggering in unsafe scenarios is avoided through decision thresholds and silent modes, and safety is ensured through relay state timeliness checks.

[0086] Figure 2 A schematic diagram of the structure of a timing intent prediction model provided in an embodiment of this application is shown.

[0087] like Figure 2 As shown, the input layer takes a multimodal feature sequence within a continuous time window as input. Specifically, with the current time t as the endpoint, it extracts multimodal scene information collected within a preset duration (e.g., 3 seconds, corresponding to 30 sampling points, with a sampling period of 100 milliseconds) to construct a multimodal feature sequence {Ft}. 29, Ft The input feature sequence is 28 ... Ft}, with a shape of 30×64. Positional encoding is added to the input feature sequence to preserve the chronological order of each feature vector in the time series. Positional encoding is calculated using sine and cosine functions, with the specific formula as follows: .

[0088] .

[0089] Where pos represents the position of the feature vector in the sequence (0 to 29), and i represents the dimension index (0 to d / 2-1, d=64). and Let represent the position encoding values ​​of the 2i-th and 2i+1-th dimensions of the feature vector at position pos, respectively. Using the above formula, each position is encoded as a unique, periodic position vector, which is added to the corresponding multimodal feature vector and then input to the encoder layer, enabling the model to utilize the sequential information of features at different times in the sequence.

[0090] The location-encoded feature sequence is processed sequentially through three Transformer encoder layers. Each Transformer encoder layer includes the following sub-layer structure: a multi-head self-attention mechanism with 8 attention heads, each with a hidden layer dimension of 128. A feedforward neural network with a hidden layer dimension of 512, employing a two-layer fully connected structure, with the intermediate layer using the ReLU activation function for non-linear transformation. Residual connections and layer normalization are also included. After processing by the three Transformer encoder layers, the output sequence is processed by a Global Average Pooling layer. Global Average Pooling averages the sequence over time, aggregating the feature vectors from 30 time steps into a single 128-dimensional vector representation. This vector incorporates the temporal features of all multimodal scene information within a continuous time window. The 128-dimensional feature vector output from the Global Average Pooling layer is input to a Fully Connected Layer, which maps it to a 5-dimensional output vector. The 5-dimensional vector output from the Fully Connected Layer is processed by the Softmax activation function to obtain the intent probability vectors corresponding to various parking functions at the current time step. The final output intent probability vector P = [pAPA, pRPA, pAVP, pManual, pNone], where each component satisfies Σpi = 1. Each component represents the probability that the user selects the Automatic Parking Assist (APA), Remote Parking Assist (RPA), Autonomous Valet Parking (AVP), Manual Parking, or no parking intent at the current moment.

[0091] Figure 3 A flowchart illustrating a relay parking method provided in an embodiment of this application is shown.

[0092] like Figure 3 As shown, taking the relay parking scenario from APA (Automatic Parking Assist) to RPA (Remote Parking Assist) as an example, the relay parking process includes the following six states: Status 1: APA is activated and executed.

[0093] The vehicle is performing the APA (Automatic Parking Assist) function. The APA controller controls the vehicle to reverse into the parking space at a speed of ≤5km / h according to the planned trajectory. It continuously collects user status information through multimodal perception, such as real-time monitoring of the driver's gaze focus and head posture through the DMS (Driver Monitoring System) camera; monitoring of the driver's seat occupancy status through the seat sensor; monitoring of the seat belt locking status through the seat belt sensor; real-time monitoring of the Bluetooth RSSI (Received Signal Strength Indicator) value change trend through the Bluetooth module; and recording the current number of parking maneuvers N (i.e., the number of times the vehicle needs to move forward and backward to adjust its direction in a narrow parking space to complete the parking operation).

[0094] During APA execution, the system continuously determines whether the relay trigger conditions are met, including: the vehicle has been stationary for ≥1 second; the RPA intent probability P_RPA output by the intent prediction model is >0.70 (i.e., the confidence level of the user's intent to use RPA exceeds the preset trigger threshold); the APA planning module estimates that the remaining parking path requires at least 2 swiping operations; in the driver status information, the seat belt status has changed from locked to unlocked, and the door status has changed from closed to open; the Bluetooth RSSI signal strength shows an attenuating trend (indicating that the user is carrying a mobile phone away from the vehicle).

[0095] When all five conditions above are met, the system determines that the user wants to get out of the car and use RPA to complete the remaining parking task, and enters state 2.

[0096] State 2: APA frozen and state saved.

[0097] After the relay is triggered, the system performs APA freeze and state saving operations, specifically including: the APA controller suspends the execution of the parking trajectory, and the vehicle remains stationary; the current APA execution state is saved to the shared memory area, and the saved content includes: the current vehicle's pose information in the global coordinate system (x-coordinate, y-coordinate, yaw angle); the sequence of remaining planned path points and the index of the current path point; the timestamp of the relay moment (used for subsequent state timeliness verification); the automatic parking function (Auto Hold) is activated to ensure that the vehicle remains stationary and safe; a prompt interface pops up on the central control screen to the user, displaying "The parking space is narrow, it is recommended to use remote parking," and providing two options: "Switch to remote parking" and "Continue manual parking." If the user does not make a selection within the preset countdown (e.g., 5 seconds), the system automatically selects the "Switch to remote parking" option and enters the relay process by default.

[0098] State 3: RPA wake-up and mobile push notification.

[0099] After the user confirms the switch to RPA, the system performs an RPA wake-up operation, which includes: the vehicle's infotainment system sending an RPA wake-up command to the user's mobile app via Bluetooth; the user's mobile app automatically switches to the RPA control interface, displaying a top-down view of the vehicle's surroundings (transmitted in real-time by the vehicle's surround-view cameras) and a "Start Parking" button; the user gets out of the vehicle and closes the door, and the system confirms the user's exit through the following methods: the seat occupancy sensor status changes from occupied to unoccupied; the door status changes from open to closed; and the Bluetooth RSSI value further decreases, indicating that the user has left the vehicle with their mobile phone. After confirming the user has exited the vehicle, the system updates the status synchronously on both the mobile app and the vehicle's infotainment system, waiting for the user to click "Start Parking".

[0100] Status 4: RPA takes over execution.

[0101] After the user clicks "Start Parking" on the mobile app, the RPA controller reads the relay status from the shared memory and continues to execute the remaining parking task. Specifically, the RPA controller performs parking at a speed of ≤1.5km / h according to the remaining path point sequence saved in the relay status; during the execution, the distance to obstacles around the vehicle is monitored in real time by ultrasonic radar to ensure a safe distance d>0.15m (i.e., a safety margin of more than 15cm is always maintained between the vehicle body and obstacles); the remaining squeezing operation and precise parking action are completed; the entire process does not require rescanning the parking space or replanning the complete path, and it seamlessly continues from the APA stopping position.

[0102] Status 5: Parking complete.

[0103] When the RPA controller determines that the vehicle has arrived at the target parking space and meets the parking completion conditions: the vehicle automatically engages the parking gear (P); the electronic parking brake (EPB) is engaged; a parking completion notification is sent to the user's mobile APP, and the parking completion status is displayed on the mobile APP interface simultaneously; the vehicle automatically turns on the hazard lights to alert pedestrians and vehicles in the vicinity.

[0104] Status 6: Standby listening.

[0105] After parking is completed, the system enters standby monitoring mode: the vehicle remains parked, and the system reduces power consumption; it continuously monitors the user's next parking or retrieval intention; if the user returns to the vehicle and starts the engine, the system automatically exits standby mode and resumes driving mode.

[0106] Figure 4 A flowchart illustrating another adaptive parking function invocation method provided in an embodiment of this application is shown.

[0107] like Figure 4 As shown, the adaptive invocation method for the parking function may include the following steps.

[0108] S410, System initialization, vehicle power-on, all sensors start.

[0109] Specifically, after the vehicle is powered on, the parking function adaptively invokes the system startup initialization process. This involves loading pre-trained time-series intent prediction model parameters, and each sensor starting up and completing self-tests, including the Driver Monitoring System (DMS) camera, ultrasonic radar, surround-view camera, Bluetooth module, and Controller Area Network (CAN bus) communication interface. The initialization configuration of each module is then completed, including setting the sampling period and configuring Precision Time Protocol (PTP) clock synchronization parameters. Specific implementation details are described above and will not be repeated here.

[0110] S420, multimodal scene information acquisition.

[0111] Specifically, multimodal scene information is synchronously collected at a preset sampling period (e.g., 100 milliseconds). Driver status information, including driver's gaze focus, head posture, hand position, and seat belt status, is collected through the driver monitoring system camera at a sampling rate of 30 frames per second. Vehicle status information, including vehicle speed, gear position, pedal opening, steering wheel angle, and steering wheel angular velocity, is collected through the controller area network bus at a period of 100 milliseconds. Environmental perception information, including parking space type, parking space size (width and length), obstacle distance matrix, and autonomous valet parking area positioning markers, is collected through ultrasonic radar and surround view cameras. External interaction information, including Bluetooth received signal strength indicator (RSSI) value and mobile application connection status, is collected through the Bluetooth module. Specific implementation details are as described above and will not be repeated here.

[0112] S430, Construction of multimodal feature sequences.

[0113] Specifically, the electronic device uses a precise time protocol to time-align the modal information, ensuring that the timestamp error of each modal information is less than a preset timestamp threshold (e.g., 10 milliseconds). Using the current time t as the endpoint, multimodal information collected within a preset duration (e.g., 3 seconds, corresponding to 30 sampling points) is extracted to construct a multimodal feature sequence {Ft} within a continuous time window. 29, Ft 28 ... Ft}. Specific implementation methods are described above and will not be repeated here.

[0114] S440, Temporal Intent Prediction Model Inference.

[0115] Specifically, the multimodal feature sequences are input into a pre-trained temporal intent prediction model for intent prediction inference. After model processing, the Softmax activation function outputs the intent probability vectors corresponding to each of the various parking functions at the current time. Specific implementation details are as described above and will not be repeated here.

[0116] S450, Availability Condition Check.

[0117] Specifically, before making a decision, the electronic device performs an availability check on each parking function and determines the availability flag (Available(f)) for each parking function. The specific implementation method is described above and will not be repeated here.

[0118] S460, Decision Score Calculation.

[0119] Specifically, based on the intent probability P(f) corresponding to each parking function, the available flag Available(f), and the preset function priority coefficient Priority(f), the decision score Score(f) for each parking function is calculated. The specific implementation method is as described above and will not be repeated here.

[0120] S470, Decision Score Analysis.

[0121] Specifically, if the decision score (Score(f)) of any parking function f is greater than the preset decision threshold (Score_threshold, default value is 0.5), then the parking function with the highest decision score is selected as the first parking function and activated. If the decision score (Score(f)) of all parking functions is less than or equal to the decision threshold (Score_threshold), then a silent mode is entered, and the electronic device does not activate any automatic parking functions, only providing parking image assistance (such as 360° surround view image and parking radar prompts). Specific implementation methods are as described above and will not be repeated here.

[0122] S480, Implementation and Feedback.

[0123] Specifically, the electronic device performs corresponding operations based on the decision result: if the decision is to activate a parking function, the function is invoked and a control command is sent to the vehicle actuator to perform the parking action, while continuously performing environmental perception and obstacle monitoring until parking is completed, automatically engaging the parking gear and sending a completion notification; if the decision is a silent mode, the parking image assistance function remains activated, no automatic parking function is invoked, and the process returns to step S420 for continuous monitoring; if a function switch or relay is triggered, the current function is safely paused, the current execution state (vehicle position and remaining planned path) is saved, and after user confirmation, the new function is invoked to continue execution from the saved state; if a system failure occurs, a safety downgrade is performed, the vehicle is stopped, the parking function is activated, and the user is prompted to take over.

[0124] Figure 5 A schematic diagram of the structure of a parking function adaptive calling device provided in an embodiment of this application is shown.

[0125] like Figure 5 As shown, the parking function adaptive calling device 500 may include a first processing module 510, a first execution module 520, a second processing module 530, and a second execution module 540.

[0126] The first processing module 510 can be used to perform intent prediction processing on the acquired multimodal scene information based on a pre-trained temporal intent prediction model, and obtain the intent probability vectors corresponding to various parking functions at the current time. The multimodal scene information includes driver state information, vehicle state information, environmental perception information, and external interaction information.

[0127] The first execution module 520 can be used to determine and activate the first parking function based on the intent probability vector to execute the corresponding parking task.

[0128] The second processing module 530 can be used to save the current execution state of the first parking function when, during the execution of the first parking function, it detects that the user intends to switch to the second parking function and the second parking function is available based on the continuously acquired multimodal scene information. The current execution state includes at least the current vehicle pose and the remaining planned path.

[0129] The second execution module 540 can be used to call the second parking function and pass the current execution status to the second parking function to continue executing the remaining parking tasks.

[0130] In some embodiments of this application, the parking function adaptive recall device 500 further includes: The information acquisition module is used to synchronously acquire the driver status information, the vehicle status information, the environmental perception information, and the external interaction information at a preset sampling period to obtain the multimodal scene information; The time alignment module is used to align the timestamps of each of the multimodal scene information using a precise time protocol, so that the timestamp error of each of the multimodal scene information is less than a preset timestamp threshold.

[0131] In some embodiments of this application, the parking function adaptive recall device 500 further includes: The vector acquisition module is used to acquire the maximum intent probability vector among the intent probability vectors; The first determining module is used to maintain the current parking function state unchanged if the maximum intention probability vector is less than a preset first threshold. The second determining module is used to determine the candidate parking function as the user intent function if there is a unique candidate parking function whose intent probability vector is greater than the preset trigger threshold corresponding to the candidate parking function. The third determining module is used to determine the candidate parking function with the largest intent probability vector among the multiple candidate parking functions as the user intent function if the intent probability vectors corresponding to multiple candidate parking functions are all greater than their respective preset trigger thresholds.

[0132] In some embodiments of this application, the first execution module 520 includes: The scoring calculation unit is used to calculate the decision score of each parking function based on the intent probability vector corresponding to the user intent function, the availability conditions corresponding to each parking function, and the preset function priority coefficient. The first determining unit is used to determine the parking function with the highest decision score as the first parking function and activate it.

[0133] In some embodiments of this application, the second processing module 530 includes: The information acquisition unit is used to continuously acquire the multimodal scene information and periodically update the intent probability vector during the execution of the first parking function. The second determining unit is used to determine that the user's intention is to switch from the first parking function to the second parking function when the intention probability vector corresponding to the second parking function exceeds a preset trigger threshold and the difference between the intention probability vector corresponding to the second parking function and the intention probability vector corresponding to the first parking function is greater than a preset backlash coefficient.

[0134] In some embodiments of this application, the second processing module 530 includes: A data storage unit is used to store the current vehicle pose and the remaining planned path to a shared memory area. The current vehicle pose information includes the vehicle's x-coordinate, y-coordinate and heading angle in the global coordinate system. The remaining planned path includes the remaining path point sequence and the current path point index. The third determining unit is used to record the timestamp of the relay moment, so that the second parking function can confirm the timeliness of the current execution state based on the timestamp.

[0135] In some embodiments of this application, the parking function adaptive recall device 500 further includes: The third processing module is used to pause the current parking task, output a prompt message to the user, and keep the parking image assistance function active when the target parking space is occupied or the planned path is blocked by an obstacle based on the environmental perception information during the execution of the first parking function.

[0136] It should be noted that, Figure 5 The parking function adaptive calling device 500 shown can perform... Figure 1-4 The various steps in the method embodiment shown are implemented. Figure 1-4 The processes and effects in the method embodiments shown are not described in detail here.

[0137] Figure 6 A schematic diagram of a parking function adaptive calling device provided in an embodiment of this application is shown.

[0138] In some embodiments of this application, Figure 6 The parking function adaptive calling device shown can be an electronic device. Specifically, the electronic device can include, but is not limited to, devices such as computer equipment, cloud servers, or cloud server clusters.

[0139] like Figure 6 As shown, the parking function adaptive calling device may include a processor 601 and a memory 602 storing computer program instructions.

[0140] Specifically, the processor 601 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0141] Memory 602 may include a large-capacity storage for information or instructions. For example, and not limitingly, memory 602 may include a hard disk drive (HDD), a floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 602 may include removable or non-removable (or fixed) media. Where appropriate, memory 602 may be internal or external to the integrated gateway device. In a particular embodiment, memory 602 is a non-volatile solid-state memory. In a particular embodiment, memory 602 includes read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (Electrically Programmable ROM, EPROM), an electrically erasable programmable PROM (EEPROM), an electrically alterable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0142] The processor 601 reads and executes computer program instructions stored in the memory 602 to perform the steps of the parking function adaptive invocation method provided in the embodiments of this application.

[0143] In one example, the parking function adaptive calling device may also include a transceiver 603 and a bus 604. For example, Figure 6 As shown, the processor 601, memory 602 and transceiver 603 are connected via bus 604 and communicate with each other.

[0144] Bus 604 may include hardware, software, or both. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 604 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0145] This application also provides a computer-readable storage medium that can store a computer program. When the computer program is executed by a processor, the processor enables the processor to implement the parking function adaptive invocation method provided in this application.

[0146] The aforementioned storage medium may include, for example, a memory 602 containing computer program instructions, which can be executed by the processor 601 of the parking function adaptive invocation device to complete the parking function adaptive invocation method provided in this embodiment. Optionally, the storage medium may be a non-transitory computer-readable storage medium, such as a ROM, random access memory (RAM), compact disc ROM (CD-ROM), magnetic tape, floppy disk, and optical data storage device.

[0147] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0148] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for adaptively invoking parking functions, characterized in that, include: The pre-trained temporal intent prediction model performs intent prediction processing on the acquired multimodal scene information to obtain the intent probability vectors corresponding to various parking functions at the current time. The multimodal scene information includes driver state information, vehicle state information, environmental perception information, and external interaction information. The first parking function is determined and activated based on the intent probability vector to execute the corresponding parking task; During the execution of the first parking function, when the user intent to switch to the second parking function is detected based on the continuously acquired multimodal scene information and the second parking function is available, the current execution state of the first parking function is saved. The current execution state includes at least the current vehicle pose and the remaining planned path. Invoke the second parking function and pass the current execution status to the second parking function to continue executing the remaining parking tasks.

2. The adaptive parking function invocation method according to claim 1, characterized in that, Before the pre-trained temporal intent prediction model performs intent prediction processing on the acquired multimodal scene information, the method further includes: The driver status information, the vehicle status information, the environmental perception information, and the external interaction information are collected synchronously at a preset sampling period to obtain the multimodal scene information; The timestamp error of each of the multimodal scene information is less than a preset timestamp threshold by using a precise time protocol.

3. The adaptive parking function invocation method according to claim 1, characterized in that, The method further includes: Obtain the maximum intent probability vector from the intent probability vectors; If the maximum intent probability vector is less than a preset first threshold, then the current parking function state remains unchanged. If there exists a unique candidate parking function whose intent probability vector is greater than the preset trigger threshold corresponding to the candidate parking function, then the candidate parking function is determined as the user intent function. If the intent probability vectors corresponding to multiple candidate parking functions are all greater than their respective preset trigger thresholds, then the candidate parking function with the largest intent probability vector among the multiple candidate parking functions is determined as the user intent function.

4. The adaptive parking function invocation method according to claim 3, characterized in that, The step of determining and activating the first parking function based on the intent probability vector includes: Based on the intent probability vector corresponding to the user intent function, the availability conditions corresponding to each parking function, and the preset function priority coefficient, the decision score of each parking function is calculated. The parking function with the highest decision score is identified as the first parking function and activated.

5. The adaptive parking function invocation method according to claim 1, characterized in that, The step of detecting the user's intention to switch to the second parking function based on the continuously acquired multimodal scene information includes: During the execution of the first parking function, the multimodal scene information is continuously acquired and the intent probability vector is periodically updated; When the intent probability vector corresponding to the second parking function exceeds a preset trigger threshold, and the difference between the intent probability vector corresponding to the second parking function and the intent probability vector corresponding to the first parking function is greater than a preset backlash coefficient, it is determined that the user intent is to switch from the first parking function to the second parking function.

6. The adaptive parking function invocation method according to claim 1, characterized in that, Saving the current execution state of the first parking function includes: The current vehicle pose and the remaining planned path are stored in a shared memory area. The current vehicle pose information includes the vehicle's x-coordinate, y-coordinate and heading angle in the global coordinate system. The remaining planned path includes the sequence of remaining path points and the index of the current path point. Record the timestamp of the relay moment so that the second parking function can confirm the timeliness of the current execution status based on the timestamp.

7. The adaptive parking function invocation method according to claim 1, characterized in that, The method further includes: During the execution of the first parking function, when the target parking space is detected to be occupied or the planned path is blocked by an obstacle based on the environmental perception information, the current parking task is paused, a prompt message is output to the user, and the parking image assistance function remains active.

8. A parking function adaptive recall device, characterized in that, include: The first processing module is used to perform intent prediction processing on the acquired multimodal scene information based on a pre-trained temporal intent prediction model to obtain the intent probability vectors corresponding to various parking functions at the current time. The multimodal scene information includes driver state information, vehicle state information, environmental perception information, and external interaction information. The first execution module is used to determine and activate the first parking function based on the intent probability vector in order to execute the corresponding parking task. The second processing module is used to save the current execution state of the first parking function when, during the execution of the first parking function, it is detected that the user intends to switch to the second parking function and the second parking function is available based on the continuously acquired multimodal scene information. The current execution state includes at least the current vehicle pose and the remaining planned path. The second execution module is used to call the second parking function and pass the current execution status to the second parking function to continue executing the remaining parking tasks.

9. A parking function adaptive recall device, characterized in that, include: processor; Memory, used to store executable instructions; The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the parking function adaptive invocation method according to any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, causes the processor to implement the parking function adaptive invocation method as described in any one of claims 1-7.

Images