Road ridge identification method for automatic parking and related equipment

Abstract

The invention discloses a road sill recognition method for automatic parking and related equipment, relates to the technical field of vehicle automatic control and intelligent sensing, and mainly aims to solve the problem that an existing road sill recognition method for automatic parking is low in accuracy. The method comprises the steps that under the condition that the confidence coefficient of a first road sill height estimation value is lower than a preset confidence coefficient, a physical verification decision process is triggered, the first road sill height estimation value is obtained based on ultrasonic waves and / or a visual sensor, and the physical verification decision process is used for driving wheels to make physical contact with a road sill; calculating a second road sill height estimation value and the confidence coefficient of the second road sill height estimation value based on physical characteristics obtained after the wheels make physical contact with the road sill; and based on the confidence coefficient of the first road sill height estimation value and the confidence coefficient of the second road sill height estimation value, respectively carrying out weighted calculation on the first road sill height estimation value and the second road sill height estimation value to obtain a target road sill height estimation value.

Description

Technical Field

[0001] This invention relates to the field of vehicle automatic control and intelligent sensing technology, and in particular to a method and related equipment for automatic parking curb recognition. Background Technology

[0002] In the "releasing parking space" scenario of automated parking, the vehicle needs to autonomously traverse curbs. The core prerequisite is accurate identification of the curb height to determine whether passage is possible and to plan the control strategy. Existing automated parking curb recognition methods mainly rely on non-contact sensors such as ultrasonic radar and visual cameras for indirect measurement and fusion estimation. However, sound waves cause severe scattering, and visual sensors are highly susceptible to lighting conditions. These technologies have inherent and insurmountable drawbacks, resulting in low overall recognition accuracy. Although the industry has compensated for and optimized this through algorithms such as multi-sensor data fusion and complex geometric model calculations, the confidence level of the output still drops significantly under the aforementioned unfavorable scenarios, failing to provide a reliable basis for vehicle control. Therefore, it is impossible to stably provide high-precision, high-confidence curb height information in all weather conditions and all scenarios, thus limiting the reliability and safety of automated parking functions in complex real-world environments. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method and related equipment for automatic parking curb recognition, the main purpose of which is to solve the problem of low accuracy of existing automatic parking curb recognition methods.

[0004] To solve at least one of the above-mentioned technical problems, in a first aspect, the present invention provides a method for identifying curbs in automatic parking, the method comprising: If the confidence level of the first curb height estimate is lower than the preset confidence level, a physical verification decision process is triggered, wherein the first curb height estimate is obtained based on ultrasonic and / or visual sensors, and the physical verification decision process is used to drive the wheels to make physical contact with the curb. Based on the physical characteristics obtained after the wheel comes into physical contact with the curb, calculate the estimated height of the second curb and the confidence level of the estimated height of the second curb; Based on the confidence levels of the first and second road embankment height estimates, the first and second road embankment height estimates are weighted and calculated to obtain the target road embankment height estimate.

[0005] Optionally, if the confidence level of the first embankment height estimate is lower than a preset confidence level, the physical verification decision process is triggered, including: The height estimate of the first curb and the confidence level of the first curb height estimate are obtained based on ultrasonic and / or visual sensors; If the confidence level of the first road embankment height estimate is lower than the preset confidence level, the physical verification decision process is triggered. When the physical verification decision process is triggered, the vehicle is controlled to move toward the curb at a preset speed so that physical contact occurs between the wheels and the curb.

[0006] Optionally, the step of calculating the estimated height of the second curb and the confidence level of the estimated height of the second curb based on the physical characteristics obtained after the wheel makes physical contact with the curb includes: When the physical verification decision process is triggered, the vehicle sensors are activated to collect the raw time-domain signal of the vehicle moving at a preset speed to approach the curb. The time-domain features and frequency-domain features of the original time-domain signal are extracted to form a physical feature vector group, wherein the time-domain features are used to reflect the physical impact intensity and motion state, and the frequency-domain features are used to characterize different physical modes; The physical feature vector set is input into the prediction model to output the estimated height of the second road embankment and the confidence level of the estimated height of the second road embankment.

[0007] Optionally, the original time-domain signal includes vehicle body inertial measurement signal, wheel speed pulse signal, and drive system signal. The vehicle body inertial measurement signal is used to reflect the linear acceleration and angular velocity of the vehicle in three-dimensional space. The wheel speed pulse signal is used to reflect changes in the wheel's rotational speed. The drive system signal is used to reflect the output status of the drive system.

[0008] Optionally, the step of extracting the time-domain and frequency-domain features of the original time-domain signal to form a physical feature vector set includes: The time-domain features of the original time-domain signal are extracted to determine the peak values ​​of accelerations along each axis, the angular velocities along each axis, and the vehicle dynamic parameters, wherein... The peak values ​​of each axial acceleration reflect the impact intensity in each direction at the moment of physical contact between the wheel and the curb. The angular velocities of each axis reflect the changes in vehicle body posture caused by the physical contact between the wheels and the road embankment. The vehicle dynamic parameters reflect the initial motion conditions of the wheels making physical contact with the curb, including the vehicle approach speed and the vehicle-curb angle. Frequency domain features of the original time-domain signal are extracted to determine the vertical vibration energy characteristics and velocity fluctuation energy characteristics, wherein, The vertical vibration energy characteristics reflect the overall lifting motion of the vehicle body caused by the physical contact between the wheel and the curb, the transient impact between the tire and the edge of the curb, and the high-frequency vibration of the suspension system. The speed fluctuation energy characteristics reflect the overall change trend of vehicle speed before and after the wheel makes physical contact with the curb, as well as the slippage or torque fluctuation at the moment the wheel makes physical contact with the curb.

[0009] Optional, The input layer of the prediction model is used to input the physical feature vector set. The structural layer includes an LSTM layer and an FCN layer. The LSTM layer is used to model the temporal features of the physical feature vector group within the time window before and after the physical contact event. The FCN layer is used to fuse the temporal features output by the LSTM layer. The output layer is used to output the estimated height of the second curb and the confidence level of the estimated height of the second curb.

[0010] Optionally, the step of weighting the first road curb height estimate and the second road curb height estimate based on the confidence levels of the first and second road curb height estimates to obtain the target road curb height estimate includes: The first weight is determined based on the square of the confidence level of the first road embankment height estimate; The second weight is determined based on the squared value of the confidence level of the second road embankment height estimate; The first weighting value is determined based on the product of the first weight and the estimated height of the first curb. The second weighting value is determined based on the product of the second weight and the estimated height of the second road embankment. Obtain the sum of the weighted values ​​of the first weighted value and the second weighted value; Obtain the sum of the weights obtained by adding the first weight and the second weight; The estimated height of the target curb is determined based on the ratio of the sum of the weighted values ​​to the sum of the weights.

[0011] Secondly, embodiments of the present invention also provide a curb recognition device for automatic parking, comprising: A triggering unit is used to trigger a physical verification decision process when the confidence level of the first curb height estimate is lower than a preset confidence level, wherein the first curb height estimate is obtained based on ultrasonic and / or visual sensors, and the physical verification decision process is used to drive the wheels to make physical contact with the curb; The calculation unit is used to calculate the estimated height of the second curb and the confidence level of the estimated height of the second curb based on the physical characteristics obtained after the wheel makes physical contact with the curb. The acquisition unit is used to perform weighted calculations on the first road embankment height estimate and the second road embankment height estimate based on the confidence levels of the first road embankment height estimate and the second road embankment height estimate, respectively, to obtain the target road embankment height estimate.

[0012] To achieve the above objectives, according to a third aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium comprising a stored program, wherein, when the program is executed by a processor, the steps of the above-described automatic parking curb recognition method are implemented.

[0013] To achieve the above objectives, according to a fourth aspect of the present invention, an electronic device is provided, comprising at least one processor and at least one memory connected to the processor; wherein the processor is configured to invoke program instructions in the memory to execute the steps of the above-described automatic parking curb recognition method.

[0014] By employing the above technical solution, the automatic parking curb recognition method and related equipment provided by this invention address the problem of low accuracy in existing automatic parking curb recognition methods. This invention triggers a physical verification decision process when the confidence level of a first curb height estimate is lower than a preset confidence level. The first curb height estimate is obtained based on ultrasonic and / or visual sensors. The physical verification decision process is used to drive the wheels to make physical contact with the curb. Based on the physical characteristics obtained after the wheels make physical contact with the curb, a second curb height estimate and its confidence level are calculated. Based on the confidence levels of the first and second curb height estimates, a weighted calculation is performed on both estimates to obtain a target curb height estimate.

[0015] In the above scheme, when the confidence level of the height estimate output by the ultrasonic or visual sensor is insufficient due to its own physical limitations or environmental interference, the optimization does not continue within the framework of processing easily interfered far-field wave signals (sound, light). Instead, a fundamentally different perception process is initiated, namely, driving the wheels to make controlled physical contact with the curb. The source of perception is actively switched from indirect wave signal measurement to direct mechanical interaction response through the actual contact between the wheels and the curb. Physical features generated during this interaction are collected and calculated. Since the signal originates from an actual mechanical action process unaffected by light and weather, a more direct and stable correlation is established with the true height, thereby reducing the interference of environmental variables on the recognition process from a perception perspective. Finally, a fusion step based on the weighted calculation of the confidence levels of the two sources is used to construct an integration mechanism guided by real-time reliability assessment. The contribution weights of the initial perception and physical verification results regarding their respective reliability are determined based on their quantitative assessments. This allows the final output to adaptively favor the more reliable source or combination of sources in the current specific scenario, thereby reducing the vulnerability to the failure of any single sensing mode in a particular scenario and improving the robustness of the overall output.

[0016] Correspondingly, the automatic parking curb recognition device, equipment, and computer-readable storage medium provided in the embodiments of the present invention also have the above-mentioned technical effects.

[0017] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a method for identifying curbs in automatic parking according to an embodiment of the present invention is shown. Figure 2 This diagram illustrates the composition of a curb recognition device for automatic parking provided in an embodiment of the present invention. Figure 3 This diagram illustrates the composition of an electronic device for automatic parking curb recognition provided in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0020] To address the issue of low accuracy in existing automatic parking curb recognition methods, this invention provides a curb recognition method for automatic parking, such as... Figure 1 As shown, the method includes: S101. If the confidence level of the first curb height estimate is lower than the preset confidence level, a physical verification decision process is triggered, wherein the first curb height estimate is obtained based on ultrasonic and / or visual sensors, and the physical verification decision process is used to drive the wheels to make physical contact with the curb. In one embodiment, the step of triggering a physical verification decision process when the confidence level of the first road embankment height estimate is lower than a preset confidence level includes: The height estimate of the first curb and the confidence level of the first curb height estimate are obtained based on ultrasonic and / or visual sensors; If the confidence level of the first road embankment height estimate is lower than the preset confidence level, the physical verification decision process is triggered. When the physical verification decision process is triggered, the vehicle is controlled to move toward the curb at a preset speed so that physical contact occurs between the wheels and the curb.

[0021] For example, the preliminary judgment of the curb height obtained is called the first curb height estimate, and the reliability assessment of this preliminary judgment is called its confidence level. The pre-set confidence level is a pre-set reliability threshold used to determine whether the preliminary perception result is sufficiently credible to support subsequent decisions.

[0022] In this application, the ultrasonic sensor array deployed on the vehicle is first fused with the surround-view vision system to obtain environmental information and calculate the estimated height of the first curb and its corresponding confidence level. The system's built-in decision logic continuously evaluates this confidence level. When the value is found to be lower than a preset reliability threshold, the current non-contact perception result is determined to be unreliable. For example, in situations such as dim lighting at dusk leading to blurred visual images, or when the curb surface is made of rough gravel causing severe ultrasonic wave scattering, the confidence level given by the sensor fusion algorithm will significantly decrease. Once this judgment is triggered, the physical verification decision process is initiated. The core action is to control the vehicle's power and braking system to move the vehicle smoothly towards the curb at a low and safe preset speed until the front or rear wheels of the vehicle make precise and controlled physical contact with the physical edge of the curb, thereby completing the paradigm shift from "telemetry" to "touch-sensing".

[0023] By employing the aforementioned technical solution, this step establishes an intelligent decision-making front-end for the entire ledge identification process, based on the inherent reliability of the perception results. It departs from the traditional approach of solely relying on and attempting to correct potentially flawed remote signals. When environmental interference renders conventional sensors inaccurate or unclear, it can autonomously initiate a fundamentally different verification path that relies on entity interaction. This is equivalent to adding a reliable degradation operation mode to the perception system, fundamentally reducing reliance on the inherent limitations of a single non-contact perception mode in adverse environments. It ensures that even in complex scenarios with high initial perception uncertainty, the system still has a way to proactively acquire more direct and robust decision-making evidence, laying the necessary foundation for obtaining accurate and robust final ledge height estimates.

[0024] S102. Based on the physical characteristics obtained after the wheel and the curb make physical contact, calculate the estimated height of the second curb and the confidence level of the estimated height of the second curb. In one embodiment, calculating the second curb height estimate and the confidence level of the second curb height estimate based on the physical characteristics obtained after the wheel makes physical contact with the curb includes: When the physical verification decision process is triggered, the vehicle sensors are activated to collect the raw time-domain signal of the vehicle moving at a preset speed to approach the curb. The time-domain features and frequency-domain features of the original time-domain signal are extracted to form a physical feature vector group, wherein the time-domain features are used to reflect the physical impact intensity and motion state, and the frequency-domain features are used to characterize different physical modes; The physical feature vector set is input into the prediction model to output the estimated height of the second road embankment and the confidence level of the estimated height of the second road embankment.

[0025] For example, the physical features in the above steps are the processed representations of the raw dynamic signals collected by various sensors on the vehicle body during the physical contact between the vehicle and the curb at low speed. These features are systematically organized into a set of physical feature vectors. The second curb height estimate is a new and independent curb height judgment calculated by the system based on this set of physical feature vectors, and the confidence level of this estimate is a probabilistic assessment of the reliability of the prediction model's current calculation result.

[0026] In this embodiment, after the system triggers the physical verification decision process, it immediately activates various sensors deployed on the vehicle body. These sensors continuously collect raw time-domain signals reflecting the vehicle's motion state as the vehicle smoothly approaches and contacts the curb at a low and constant preset speed. These signals include vehicle body inertial measurement signals reflecting the vehicle's three-dimensional linear acceleration and angular velocity, wheel speed pulse signals reflecting instantaneous changes in wheel rotation speed, and drive system signals reflecting the real-time power output of the drive system. The system's built-in processor then processes these raw, time-sequential signal streams. On one hand, it directly extracts time-domain features such as peak accelerations and angular velocities along each axis to characterize the impact intensity and vehicle body attitude changes at the moment of contact. On the other hand, it converts the time-domain signals to the frequency domain for analysis, extracting frequency-domain features such as vibration energy in specific frequency bands to distinguish and quantify different physical response modes, such as overall vehicle body lifting and local high-frequency vibrations of the suspension. All these time-domain and frequency-domain features are collectively constructed into a fixed-dimensional, information-rich physical feature vector set. Ultimately, this feature vector set, containing the complete dynamic fingerprint of this physical interaction, is fed into a prediction model that has been pre-trained and optimized with a large amount of data. This model, through its complex internal nonlinear mapping capabilities, interprets and calculates this feature vector, simultaneously outputting two key results: a more accurate estimate of the second-path threshold height based on the inversion of the current physical interaction, and a confidence level of the second-path threshold height estimate that characterizes the model's grasp of the calculation results.

[0027] By employing the aforementioned technical solution, this step achieves a fundamental perception transformation: converting a specific, controlled mechanical contact event into a series of precisely quantifiable and analyzable digital features, ultimately interpreting the target's geometric information through an intelligent model. Since the data source for the entire calculation process is the vehicle's response signal generated by the unavoidable mechanical interaction between the wheel and the curb, these signals have a direct and deterministic physical causal relationship with the curb height, and their acquisition process is almost unaffected by ambient light, weather conditions, or the optical and acoustic properties of the curb surface. Therefore, the height estimate calculated using this approach fundamentally reduces its dependence on distant and volatile external perception environments in terms of accuracy and stability. Simultaneously, the confidence level output by the predictive model provides an honest assessment of the inherent reliability of the measurement results for this specific interaction, enabling the system to obtain height information that is not only more accurate but also inherently quality-assessed, providing high-quality and reliable input for subsequent decision fusion steps.

[0028] In one embodiment, the original time-domain signal includes vehicle body inertial measurement signal, wheel speed pulse signal, and drive system signal. The vehicle body inertial measurement signal is used to reflect the linear acceleration and angular velocity of the vehicle in three-dimensional space. The wheel speed pulse signal is used to reflect changes in the wheel's rotational speed. The drive system signal is used to reflect the output status of the drive system.

[0029] For example, the aforementioned raw time-domain signal is the underlying data stream collected in real time by the corresponding sensors activated on the vehicle during the execution of the physical verification decision process and the control of the vehicle to move towards the curb at a preset speed. The vehicle body inertial measurement signal specifically refers to the raw data generated by the inertial measurement unit, reflecting the linear acceleration and angular velocity of the vehicle's rotation along each axis in three-dimensional space. The wheel speed pulse signal specifically refers to the raw pulse sequence generated by the wheel speed sensor, reflecting the instantaneous changes in the rotational speed of each wheel. The drive system signal specifically refers to the raw data provided by the vehicle drive motor controller or related electronic control unit, reflecting the current actual output state of the drive system (such as torque or current).

[0030] In this embodiment, when the system triggers the physical verification decision process and controls the vehicle to approach the curb at a low speed and smoothly, the three types of sensors are simultaneously activated to enter a high-frequency data acquisition state. During the brief period of approach and even slight contact, inertial measurement units at various points on the vehicle body continuously measure and output the acceleration and angular velocity changes of the vehicle body, forming the vehicle body inertial measurement signal; magnetoelectric or Hall effect sensors at each wheel bearing continuously capture the pulses passing through wheel teeth or specific markers, forming wheel speed pulse signals reflecting the actual rotational behavior of the tires; simultaneously, the vehicle's overall controller or motor controller acquires and outputs the current motor torque command or phase current feedback in real time through the internal bus, forming the drive system signal. These signals are recorded synchronously with high fidelity, collectively forming a multi-dimensional and original temporal data foundation describing this specific physical interaction event. For example, when a wheel touches the edge of a curb at a constant low speed, the inertial measurement unit immediately captures the vertical impact acceleration of the vehicle body, the wheel speed sensor keenly records the instantaneous speed fluctuation of the drive wheel due to the obstruction, and the drive system signal reflects the instantaneous change in torque required to maintain the preset approach speed.

[0031] Using the aforementioned technical solution, this step constructs a comprehensive and reliable underlying data foundation for subsequent feature extraction and model calculation. By simultaneously acquiring raw signals from three core dimensions—vehicle motion attitude, tire rotational dynamics, and drive execution state—a complete dynamic profile describing the transient event of wheel-road contact is obtained. Vehicle inertial measurement signals directly record the rigid body motion and vibration of the vehicle body excited by the contact impact; wheel speed pulse signals reflect the most direct interaction between the tire and road embankment, the rolling pair; and drive system signals reveal the power system's response and resistance to this interaction. This simultaneous acquisition of multi-source heterogeneous data allows subsequent processing to comprehensively interpret the information contained in this contact event from different perspectives and physical levels, thus providing in-depth, mutually corroborating data support for accurately inverting the road embankment height and reducing the risk of interpretation bias or information loss that may be introduced by relying on signals from a single type of sensor.

[0032] In one embodiment, the step of extracting the time-domain and frequency-domain features of the original time-domain signal to form a physical feature vector set includes: The time-domain features of the original time-domain signal are extracted to determine the peak values ​​of accelerations along each axis, the angular velocities along each axis, and the vehicle dynamic parameters, wherein... The peak values ​​of each axial acceleration reflect the impact intensity in each direction at the moment of physical contact between the wheel and the curb. The angular velocities of each axis reflect the changes in vehicle body posture caused by the physical contact between the wheels and the road embankment. The vehicle dynamic parameters reflect the initial motion conditions of the wheels making physical contact with the curb, including the vehicle approach speed and the vehicle-curb angle. Frequency domain features of the original time-domain signal are extracted to determine the vertical vibration energy characteristics and velocity fluctuation energy characteristics, wherein, The vertical vibration energy characteristics reflect the overall lifting motion of the vehicle body caused by the physical contact between the wheel and the curb, the transient impact between the tire and the edge of the curb, and the high-frequency vibration of the suspension system. The speed fluctuation energy characteristics reflect the overall change trend of vehicle speed before and after the wheel makes physical contact with the curb, as well as the slippage or torque fluctuation at the moment the wheel makes physical contact with the curb.

[0033] For example, the peak values ​​of each axial acceleration in the above steps refer to the maximum linear acceleration values ​​achieved by the vehicle in the forward, lateral, and vertical axes at the instant the wheels contact the curb, extracted from the vehicle body inertial measurement signals. These values ​​are used to quantify the impact force caused by the contact in various dimensions. The angular velocities of each axis refer to the angular velocities of the vehicle's rotation around its forward, lateral, and vertical axes, extracted from the same signal source. These angular velocities describe the rate of change in the vehicle's pitch, yaw, and roll caused by the contact. Vehicle dynamic parameters specifically refer to the vehicle's motion state immediately before the physical contact, mainly including the instantaneous velocity of the vehicle approaching the curb and the angle formed between the vehicle's direction of travel and the curb's extension direction. Vertical vibration energy characteristics refer to the signal energy calculated in specific low-frequency and high-frequency ranges after frequency domain analysis of the vertical acceleration signal. These characteristics characterize the motion modes triggered by the contact, primarily characterized by the slow overall lifting of the vehicle body and the high-frequency vibration modes primarily characterized by rapid local vibrations of the tires and suspension. Speed ​​fluctuation energy characteristics refer to the signal energy in the lower and mid-frequency ranges calculated by performing frequency domain analysis on vehicle speed or wheel speed signals. This is used to characterize the macroscopic change trend of vehicle speed before and after contact, as well as the slight slippage or torque fluctuation that may occur in the drive wheels at the moment of contact.

[0034] In implementing this step, this embodiment performs two aspects of feature calculation in parallel during the processing of the acquired raw time-domain signals. Firstly, it directly searches and calculates the waveform data arranged in chronological order to identify the values ​​of the vehicle's three-axis acceleration at various time points, capturing the maximum value within the entire contact event time window as the peak value of each axial acceleration. Simultaneously, it directly reads the angular velocity readings at the same moment as the angular velocities of each axis. Furthermore, it calculates the stable vehicle speed at the last moment before contact from the wheel speed pulse signals as the vehicle approach speed, and combines this with orientation information from the visual or inertial measurement unit to calculate the vehicle-road curb angle. On the other hand, the system employs a fast signal transformation algorithm to convert a time-domain signal containing the vertical acceleration at the moment of contact to the frequency domain, and then integrates and calculates its energy in the 20-50 Hz and 50-100 Hz frequency bands respectively, thus obtaining the vertical vibration energy characteristics. Similarly, it converts a time-domain signal containing the vehicle speed at the moment of contact to the frequency domain, and calculates its energy in the 0-5 Hz and 5-20 Hz frequency bands respectively, thus obtaining the velocity fluctuation energy characteristics. Finally, all these calculated values ​​are arranged in a predetermined order and dimension to form a highly condensed physical feature vector set containing time-domain intensity, time-domain attitude, initial conditions, frequency-domain vibration mode, and frequency-domain velocity fluctuation mode.

[0035] It is important to note that the extracted time-domain and frequency-domain features in this application are organized into a twelve-dimensional physical feature vector with fixed dimensions and a clear order, according to their physical meaning and calculation sequence. The construction of this vector follows a clear correspondence: the first dimension corresponds to the vehicle approach speed, and the second dimension corresponds to the vehicle-road curb angle; these two dimensions together constitute the vehicle dynamic parameters describing the initial contact conditions. The third, fourth, and fifth dimensions correspond to the peak acceleration values ​​of the X, Y, and Z axes, respectively, i.e., the peak acceleration values ​​of each axis. The sixth, seventh, and eighth dimensions correspond to the angular velocities of the X, Y, and Z axes, respectively, i.e., the angular velocities of each axis. The ninth and tenth dimensions correspond to the low-frequency energy of the Z-axis acceleration in the 20-50 Hz frequency band and the high-frequency energy in the 50-100 Hz frequency band, respectively, i.e., the vertical vibration energy characteristics. The eleventh and twelfth dimensions correspond to the low-frequency energy of the velocity signal in the 0-5 Hz frequency band and the high-frequency energy in the 5-20 Hz frequency band, respectively, i.e., the velocity fluctuation energy characteristics. Thus, all twelve specific features are arranged sequentially to form a digital representation that comprehensively describes the wheel-road curb physical contact event, from initial conditions and time-domain impact response to frequency-domain vibration modes. This structured twelve-dimensional physical feature vector serves as the direct input for the subsequent lightweight deep learning model to estimate the height of the second curb and calculate its confidence level.

[0036] By employing the aforementioned technical solution, this step achieves a crucial transformation from raw, high-dimensional, time-flowing sensor data streams to a structured, digitalized feature set rich in physical meaning. It doesn't merely capture the most prominent impact peak of the contact event; rather, through dual analysis in the time and frequency domains, it systematically mines the multi-layered dynamic information inherent in this interaction. Time-domain features directly and faithfully record the "intensity" and "attitude" results of the interaction, while frequency-domain features profoundly reveal the processes and patterns that trigger these results in different physical components, such as the overall vehicle body, local suspension, and tire treads. This feature extraction method allows subsequent predictive models to simultaneously understand the macroscopic mechanical performance and microscopic vibration composition of the contact event, thus enabling a more comprehensive and intrinsic understanding of this specific interaction between the wheel and the curb. This lays a solid and rich information foundation for high-fidelity inference of geometric parameters from the mechanical response, reducing the risk of model misunderstanding or information loss due to incomplete feature representation.

[0037] In one embodiment, The input layer of the prediction model is used to input the physical feature vector set. The structural layer includes an LSTM layer and an FCN layer. The LSTM layer is used to model the temporal features of the physical feature vector group within the time window before and after the physical contact event. The FCN layer is used to fuse the temporal features output by the LSTM layer. The output layer is used to output the estimated height of the second curb and the confidence level of the estimated height of the second curb.

[0038] For example, the prediction model described above is a specially trained lightweight deep learning model. Its input layer receives a set of physical feature vectors, which are structured features containing both temporal and frequency domain information extracted from the physical contact event between the wheel and the curb in the preceding steps. The LSTM layer is a special type of recurrent neural network layer designed to capture and model the dynamic changes and sequential dependencies of the physical feature vector set within a short time window before and after the contact event; these dynamic changes are called temporal features. The FCN layer, or fully connected layer, further integrates and transforms the temporally information-rich features output by the LSTM layer. The second curb height estimate is a specific height value finally calculated by the model, and the confidence level of this estimate is a probability score given by the model regarding the reliability of its calculation result.

[0039] This embodiment employs an offline training and online deployment / inference architecture when constructing and utilizing the prediction model. In the offline phase, the model is trained using data collected from numerous physical contact experiments conducted on road embankments of known heights, thereby learning the complex mapping relationship between physical feature vectors and the actual road embankment height. Specifically, in terms of model structure, when the twelve-dimensional physical feature vector set is input, it first enters the LSTM layer. The memory units within this layer can sequentially process features arranged chronologically or with causal relationships within the vector set, effectively capturing, for example, the evolution pattern of the entire dynamic process of a vehicle body from a smooth approach to a sudden impact and then to gradual stabilization—its temporal characteristics. Subsequently, the features output from the LSTM layer, already containing temporal contextual information, are fed into the FCN layer. This layer, through multiple layers of nonlinear transformations, fuses and refines these features into a higher-level abstract representation. Finally, the output layer decodes this abstract representation into two specific values: one is a second road embankment height estimate inferred based on all current input information, and the other is the confidence level corresponding to this estimate, evaluated by the model through its internal mechanisms (such as softmax probability or specific design). Before deployment, the entire model underwent lightweight optimizations such as pruning and quantization, enabling it to be converted into efficient embedded code, deployed on automotive-grade microcontrollers, and meeting stringent real-time requirements.

[0040] Using the aforementioned technical solution, this step achieves intelligent and high-precision inversion of the height information contained in a transient physical interaction event through a carefully designed neural network structure. Leveraging the ability of the LSTM layer to process temporal features, the model understands the causal relationships of the contact event as a dynamic process, rather than just a few isolated feature points. This allows for a deeper interpretation of the impact's establishment, transmission, and attenuation processes, thus more accurately relating it to the geometric height that triggered the process. The FCN layer further provides powerful feature fusion and abstraction capabilities, ensuring robustness in the mapping from original features to the final height value. This model design, combining temporal modeling and deep feature fusion, enables the system to extract the deepest and most deterministic information from a simple contact, significantly reducing reliance on initial perception results sensitive to environmental interference. It also provides the core computational engine for generating a second-path threshold height estimate that is both highly accurate and reliable. The design of the model's output confidence level provides a crucial quality assessment basis for subsequent decision fusion steps, enabling the entire system to flexibly and reliably integrate information from different sources and with varying degrees of reliability through self-evaluation.

[0041] S103. Based on the confidence levels of the first curb height estimate and the second curb height estimate, the first curb height estimate and the second curb height estimate are weighted and calculated respectively to obtain the target curb height estimate.

[0042] In one embodiment, the step of weighting the first curb height estimate and the second curb height estimate based on the confidence levels of the first and second curb height estimates to obtain the target curb height estimate includes: The first weight is determined based on the square of the confidence level of the first road embankment height estimate; The second weight is determined based on the squared value of the confidence level of the second road embankment height estimate; The first weighting value is determined based on the product of the first weight and the estimated height of the first curb. The second weighting value is determined based on the product of the second weight and the estimated height of the second road embankment. Obtain the sum of the weighted values ​​of the first weighted value and the second weighted value; Obtain the sum of the weights obtained by adding the first weight and the second weight; The estimated height of the target curb is determined based on the ratio of the sum of the weighted values ​​to the sum of the weights.

[0043] In this step of the application, the first step is to determine the currently available combination of input information. If the confidence level of the first road embankment height estimate is already high enough in the previous decision-making process, and the system has not initiated the physical verification process, then the system will directly use the first road embankment height estimate as the final target road embankment height estimate. Conversely, if the system performs physical verification due to insufficient initial perceived confidence, it will simultaneously possess two sets of information: the first road embankment height estimate and its confidence level obtained based on ultrasonic and / or visual sensors, and the second road embankment height estimate and its confidence level calculated after physical contact. In this case, the system performs a weighted fusion calculation based on reliability assessment. Specifically, the system first squares the first confidence level and the second confidence level respectively, obtaining the first weight and the second weight used to measure the proportion or "voice" of their respective height estimates in the fusion. This operation of squaring the confidence level means giving greater trust to high-confidence results and further suppressing the influence of low-confidence results. Next, the system multiplies the first road embankment height estimate by the first weight to obtain the first weighted value, and multiplies the second road embankment height estimate by the second weight to obtain the second weighted value. Then, these two weighted values ​​are added together to obtain the weighted sum, and the first weight and the second weight are added together to obtain the weight sum. Finally, the target road embankment height estimate is determined by dividing this weighted sum by the weight sum. For example, at dusk, the initial visual perception has low confidence due to dim lighting, but the height estimate it provides may still contain some valid information; at the same time, the physical verification process is not affected by lighting and may provide a second estimate with high accuracy and high confidence. Through the weighted calculation in this step, the high-confidence physical verification result will dominate the final result, while the low-confidence initial perception result will only have a slight impact, thus obtaining a target value that is closer to reality.

[0044] Final target curb height estimate Estimated height of the first embankment Estimated height of the second road embankment Based on their respective confidence levels and The weighted average of the squares is calculated, and its mathematical expression is:

[0045]

[0046]

[0047] in: First curb height estimate. A preliminary curb height estimate obtained based on non-contact sensing methods such as ultrasonic and / or visual sensors.

[0048] : Confidence level of the estimated height of the first embankment. A representation Reliability assessment values ​​typically range from 0 to 1, with higher values ​​indicating greater reliability. The more credible it is.

[0049] The second curb height estimate is calculated using a prediction model based on the physical characteristics obtained after the wheel comes into physical contact with the curb.

[0050] : Confidence level of the second embankment height estimate. A representation The reliability assessment value, also between 0 and 1, is output synchronously by the prediction model; a higher value indicates greater reliability. The more credible it is.

[0051] First weight. Its value is the first confidence level. square ( ), used to assign in fused computing The corresponding proportion.

[0052] Second weight. Its value is the second confidence level. square ( ), used to assign in fused computing The corresponding proportion.

[0053] : Estimated height of the target curb. (Based on...) and Perform weight-based , The weighted average is used to obtain a single, uniform estimate of the curb height for vehicle control decisions.

[0054] By employing the aforementioned technical solution, this step provides an adaptive, information quality-oriented decision endpoint for the entire curb recognition process. It does not mechanically select a particular path, but creatively constructs a dynamic weighted fusion mechanism, enabling the final decision to flexibly reflect the real-time reliability of different information sources in the specific context. By using the square of the confidence level as the weight, the system strengthens the contribution of high-reliability information while naturally weakening the interference of low-reliability information, achieving an intelligent "superior weight" allocation. This ensures that the final target curb height estimate not only integrates the advantages of multi-source information, but also that its formation process itself is inherently sensitive to information quality. This significantly reduces the risk of misjudgment due to accidental failure of initial perception in specific environments or deviations in physical verification caused by extremely rare special circumstances. It improves the robustness and overall reliability of the entire system's output, providing crucial and reliable control basis for subsequent vehicles to smoothly and safely cross curbs.

[0055] Furthermore, as a response to the above Figure 1 In addition to the implementation of the method shown, this embodiment of the invention also provides an automatic parking curb recognition device for the above-mentioned... Figure 1 The method shown is implemented accordingly. This device embodiment corresponds to the foregoing method embodiment. For ease of reading, this device embodiment will not repeat the details of the foregoing method embodiment, but it should be clear that the device in this embodiment can implement all the contents of the foregoing method embodiment. Figure 2 As shown, the device includes: a triggering unit 21, a calculation unit 22, and an acquisition unit 23, wherein... Triggering unit 21 is used to trigger a physical verification decision process when the confidence level of the first curb height estimate is lower than a preset confidence level, wherein the first curb height estimate is obtained based on ultrasonic and / or visual sensors, and the physical verification decision process is used to drive the wheel to make physical contact with the curb; The calculation unit 22 is used to calculate the second road embankment height estimate and the confidence level of the second road embankment height estimate based on the physical characteristics obtained after the wheel and the road embankment make physical contact. The acquisition unit 23 is used to perform weighted calculations on the first road embankment height estimate and the second road embankment height estimate based on the confidence levels of the first road embankment height estimate and the second road embankment height estimate, respectively, to obtain the target road embankment height estimate.

[0056] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and by adjusting kernel parameters, a method for recognizing road ledges in automatic parking can be implemented, addressing the problem of low accuracy in existing automatic parking ledge recognition methods.

[0057] This invention provides a computer-readable storage medium including a stored program that, when executed by a processor, implements the automatic parking curb recognition method.

[0058] This invention provides a processor for running a program, wherein the program executes the automatic parking curb recognition method during runtime.

[0059] This invention provides an electronic device, which includes at least one processor and at least one memory connected to the processor; wherein the processor is used to call program instructions in the memory to execute the automatic parking curb recognition method described above. This invention provides an electronic device 30, such as... Figure 3 As shown, the electronic device includes at least one processor 301, and at least one memory 302 and bus 303 connected to the processor; wherein, the processor 301 and the memory 302 communicate with each other through the bus 303; the processor 301 is used to call program instructions in the memory to execute the above-mentioned automatic parking curb recognition method.

[0060] The smart electronic devices mentioned in this article can be PCs, tablets, mobile phones, etc.

[0061] This application also provides a computer program product that, when executed on a process management electronic device, is suitable for executing a program that initializes the above-described automatic parking curb recognition method steps.

[0062] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0063] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] This application also provides a computer program product, which includes computer software instructions that, when executed on a processing device, cause the processing device to perform actions such as... Figure 1 The control flow of the memory in the corresponding embodiment.

[0068] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0070] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0071] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for identifying curbs in automatic parking, characterized in that, include: If the confidence level of the first curb height estimate is lower than the preset confidence level, a physical verification decision process is triggered, wherein the first curb height estimate is obtained based on ultrasonic and / or visual sensors, and the physical verification decision process is used to drive the wheels to make physical contact with the curb. Based on the physical characteristics obtained after the wheel comes into physical contact with the curb, calculate the estimated height of the second curb and the confidence level of the estimated height of the second curb; Based on the confidence levels of the first and second road embankment height estimates, the first and second road embankment height estimates are weighted and calculated to obtain the target road embankment height estimate.

2. The method according to claim 1, characterized in that, When the confidence level of the estimated height of the first embankment is lower than the preset confidence level, the physical verification decision process is triggered, including: The height estimate of the first curb and the confidence level of the first curb height estimate are obtained based on ultrasonic and / or visual sensors; If the confidence level of the first road embankment height estimate is lower than the preset confidence level, the physical verification decision process is triggered. When the physical verification decision process is triggered, the vehicle is controlled to move toward the curb at a preset speed so that physical contact occurs between the wheels and the curb.

3. The method according to claim 2, characterized in that, The calculation of the second curb height estimate and its confidence level based on the physical characteristics obtained after the wheel and curb come into physical contact includes: When the physical verification decision process is triggered, the vehicle sensors are activated to collect the raw time-domain signal of the vehicle moving at a preset speed to approach the curb. The time-domain features and frequency-domain features of the original time-domain signal are extracted to form a physical feature vector group, wherein the time-domain features are used to reflect the physical impact intensity and motion state, and the frequency-domain features are used to characterize different physical modes; The physical feature vector set is input into the prediction model to output the estimated height of the second road embankment and the confidence level of the estimated height of the second road embankment.

4. The method according to claim 3, characterized in that, The original time-domain signals include vehicle body inertial measurement signals, wheel speed pulse signals, and drive system signals. The vehicle body inertial measurement signal is used to reflect the linear acceleration and angular velocity of the vehicle in three-dimensional space. The wheel speed pulse signal is used to reflect changes in the wheel's rotational speed. The drive system signal is used to reflect the output status of the drive system.

5. The method according to claim 3, characterized in that, The step of extracting the time-domain and frequency-domain features of the original time-domain signal to form a physical feature vector set includes: The time-domain features of the original time-domain signal are extracted to determine the peak values ​​of accelerations along each axis, the angular velocities along each axis, and the vehicle dynamic parameters, wherein... The peak values ​​of each axial acceleration reflect the impact intensity in each direction at the moment of physical contact between the wheel and the curb. The angular velocities of each axis reflect the changes in vehicle body posture caused by the physical contact between the wheels and the road embankment. The vehicle dynamic parameters reflect the initial motion conditions of the wheels making physical contact with the curb, including the vehicle approach speed and the vehicle-curb angle. Frequency domain features of the original time-domain signal are extracted to determine the vertical vibration energy characteristics and velocity fluctuation energy characteristics, wherein, The vertical vibration energy characteristics reflect the overall lifting motion of the vehicle body caused by the physical contact between the wheel and the curb, the transient impact between the tire and the edge of the curb, and the high-frequency vibration of the suspension system. The speed fluctuation energy characteristics reflect the overall change trend of vehicle speed before and after the wheel makes physical contact with the curb, as well as the slippage or torque fluctuation at the moment the wheel makes physical contact with the curb.

6. The method according to claim 3, characterized in that, The input layer of the prediction model is used to input the physical feature vector set. The structural layer includes an LSTM layer and an FCN layer. The LSTM layer is used to model the temporal features of the physical feature vector group within the time window before and after the physical contact event. The FCN layer is used to fuse the temporal features output by the LSTM layer. The output layer is used to output the estimated height of the second curb and the confidence level of the estimated height of the second curb.

7. The method according to claim 1, characterized in that, The step of weighting the first road curb height estimate and the second road curb height estimate based on the confidence levels of the first and second road curb height estimates to obtain the target road curb height estimate includes: The first weight is determined based on the square of the confidence level of the first road embankment height estimate; The second weight is determined based on the squared value of the confidence level of the second road embankment height estimate; The first weighting value is determined based on the product of the first weight and the estimated height of the first curb. The second weighting value is determined based on the product of the second weight and the estimated height of the second road embankment. Obtain the sum of the weighted values ​​of the first weighted value and the second weighted value; Obtain the sum of the weights obtained by adding the first weight and the second weight; The estimated height of the target curb is determined based on the ratio of the sum of the weighted values ​​to the sum of the weights.

8. A curb recognition device for automatic parking, characterized in that, Also includes: A triggering unit is used to trigger a physical verification decision process when the confidence level of the first curb height estimate is lower than a preset confidence level, wherein the first curb height estimate is obtained based on ultrasonic and / or visual sensors, and the physical verification decision process is used to drive the wheels to make physical contact with the curb; The calculation unit is used to calculate the estimated height of the second curb and the confidence level of the estimated height of the second curb based on the physical characteristics obtained after the wheel makes physical contact with the curb. The acquisition unit is used to perform weighted calculations on the first road embankment height estimate and the second road embankment height estimate based on the confidence levels of the first road embankment height estimate and the second road embankment height estimate, respectively, to obtain the target road embankment height estimate.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed by a processor, it implements the steps of the automatic parking curb recognition method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes at least one processor and at least one memory connected to the processor; wherein the processor is configured to invoke program instructions in the memory to execute the steps of the automatic parking curb recognition method as described in any one of claims 1 to 7.

Images