Method, system and computer program product for parking assistance

An intelligent parking assistance system using machine learning and rule-based algorithms addresses the challenges of urban parking by adapting to user preferences, enhancing efficiency and safety through personalized space recommendations.

WO2026233677A1PCT designated stage Publication Date: 2026-11-12MERCEDES BENZ GROUP AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2026/061268
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-09
Filing Date
2026-04-20
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Traditional parking methods in urban environments are challenging due to limited space and varying user preferences, often resulting in suboptimal outcomes and vehicle damage, and existing parking assistance technologies lack adaptability and personalization.

Method used

An intelligent parking assistance system using machine learning and rule-based algorithms to identify and recommend parking spaces, adapting to individual user habits by continuously learning from parking choices and providing personalized recommendations.

Benefits of technology

Enhances parking efficiency by accurately identifying optimal spaces, reducing the risk of collisions, and improving user satisfaction through personalized and adaptive parking guidance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000035_0000
    Figure 00000035_0000
  • Figure 00000036_0000
    Figure 00000036_0000
  • Figure 00000037_0000
    Figure 00000037_0000
Patent Text Reader

Abstract

In some implementations, a device may include determining, by a processor based on a current location of a vehicle, that the vehicle has entered a predefined geofenced area; determining, by the processor using a trained prediction model, a plurality of parking space candidates within the predefined geofenced area. In addition, the device may include determining, by the processor using the trained prediction model and based on the plurality of parking space candidates, a selected parking space within the predefined geofenced area, the selected parking space being selected based on an occupancy status of the plurality of parking space and a type of the predefined geofenced area. The device may include determining, by the processor, a first binary representation for the plurality of parking space candidates indicating the selected parking space. Mercedes-Benz Group AG
Need to check novelty before this filing date? Find Prior Art

Description

Mercedes-Benz Group AGMETHOD, SYSTEM AND COMPUTER PROGRAM PRODUCT FOR PARKING ASSISTANCETECHNICAL FIELD

[0001] Embodiments of the present disclosure relate generally to vehicle navigation systems. More particularly, embodiments of the disclosure relate to an intelligent parking assistance system designed to optimize parking location prompts and parking spot recommendations using machine learning and rule-based algorithms.BACKGROUND

[0002] Parking in urban environments can be challenging due to limited space and varying user preferences. Traditional parking methods rely heavily on driver skill and spatial awareness, often resulting in suboptimal parking outcomes and vehicle damage. In addition, traditional parking methods lack adaptability to individual parking habits and fail to provide personalized parking spot recommendations. While existing parking assistance technologies have made some progress in addressing these challenges, they frequently suffer from limitations that impact their effectiveness and user adoption.BRIEF DESCRIPTION OF THE DRAWINGS

[0003] Embodiments of the disclosure are illustrated by way of example and not limitation in the figures of the accompanying drawings in which like references indicate similar elements.

[0004] FIG. 1 is a block diagram illustrating a computing system according to one embodiment.

[0005] FIG. 2 is a block diagram illustrating an example of an autonomous vehicle according to one embodiment.

[0006] FIG. 3 is a block diagram illustrating an example of a parking space recommendation system according to one embodiment.

[0007] FIG. 4 is a block diagram illustrating an example of a feature processor according to one embodiment.

[0008] FIG. 5 is a flowchart depicting feature extraction, training, prediction for determining parking spot recommendations.

[0009] FIG. 6 is a diagram 600 of a user journey in conjunction with the intuitive parking system according to one embodiment.

[0010] FIG. 7 is an exemplary recommendation of the highest-ranking parking space selected based on the machine learning model according to one embodiment.

[0011] FIG. 8 depicts an example of determination of known parking areas according to one embodiment.

[0012] FIG. 9 depicts an example of detection of unknown parking areas according to one embodiment.

[0013] FIG. 10 is a flow diagram illustrating a process of machine learning model-based parking space determination according to one embodiment.FIG. 11 is a block diagram illustrating an example of a data processing system which may be used with one embodiment of the disclosure.DETAILED DESCRIPTION

[0014] Various embodiments and aspects of the disclosures will be described with reference to details discussed below, and the accompanying drawings will illustrate the various embodiments. The following description and drawings are illustrative of the disclosure and are not to be construed as limiting the disclosure. Numerous specific details are described to provide a thorough understanding of various embodiments of the present disclosure. However, in certain instances, well-known or conventional details are not described in order to provide a concise discussion of embodiments of the present disclosures.

[0015] According to some embodiments, methods include determining, by a processor based on a current location of a vehicle, that the vehicle has entered a predefined geofenced area; determining, by the processor using a trained prediction model, a plurality of parking space candidates within the predefined geofenced area. In addition, the method may include determining, by the processor using the trained prediction model and based on the plurality of parking space candidates, a selected parking space within the predefined geofenced area, the selected parking space being selected based on an occupancy status of the plurality of parking space and a type of the predefined geofenced area. The method may include determining, by the processor, a first binary representation for the plurality of parking space candidates indicating the selected parking space. Moreover, the method may include determining, by the processor, a second binary representation for the plurality of parking space candidates indicating one or more unselected parking spaces, where the first binary representation includes a value of 1, and where the second binary representation includes a value of 0. Also, the method may include in response to determining the selected parking space within the predefined geofenced area, activating, by the processor, a notification. Further, the method may include receiving, by the processor, a user response to the notification. The method may includenavigating, by the processor, the vehicle to the selected parking space within the predefined geofenced area based on the user response.

[0016] The system of the present disclosure detects when the vehicle enters a predefined parking area using its current location; uses a trained prediction model to identify several possible parking spaces within this area. The system selects the best parking space from the identified candidates using the prediction model. The system also assigns a binary value to each parking space candidate: T for the selected space and ‘0’ for unselected spaces. Once a parking space is selected, the system sends a notification to the user. The system receives and processes the user's response to the notification. Based on the user's response, the system navigates the vehicle to the chosen parking space. This method aims to simplify the parking process by automatically identifying and guiding the vehicle to the optimal parking space.

[0017] Embodiments of the present disclosure provides methods that continuously learn from each parking choice made by the user of the vehicle, adapting to changing preferences over time. This dynamic learning capability ensures that the methods remain responsive to the users’ evolving habits and preferences.

[0018] The methods can handle both new users and experienced users. For new users, it employs a rule-based approach to provide initial parking suggestions. As the user gains experience, the methods transition to a machine learning-based approach, leveraging historical data to make more accurate predictions.

[0019] By combining intuitive rules with machine learning algorithms, the methods provide a balanced approach to parking assistance. This integration allows the methods to provide reliable recommendations while continuously improving its accuracy through machine learning.

[0020] Additionally, the methods maintain running statistics for proper feature scaling. This ensures that numerical features are appropriately scaled, enhancing the model’s ability to understand relative differences and make precise predictions.

[0021] Referring now to FIG. 1 , an example diagram of a computing system 100 is shown, according to one embodiment of the disclosure. As shown, computing system 100 includes an intuitive parking system 102. The intuitive parking system 102 may be included in a vehicle driven by the user and recommending the best parking slot to the user via a prompt. The intuitive parking system 102 may be communicatively coupled to one or more servers 104 and servers 106 over a network 108.

[0022] In one embodiment, intuitive parking system 102 includes, but is not limited to, sensor system 110, control unit system 112, integration system 114, interface system 116, and operating function system 118. The intuitive parking system 102 may further include certain common components included in ordinary parking assist system.

[0023] The sensor system 110 interfaces with a control unit system 112 through a dedicated high-speed CAN bus network, enabling real-time data processing and integration with the vehicle’s steering and braking systems. The control unit system 112 processes the sensor data and accurate parking assistance.

[0024] Integration system 114 enables seamless communication and coordination between parking assistance components and various vehicle subsystems.

[0025] The integration system 114 employs a gateway module that manages: (a) Message routing between different network segments; (b) Protocol conversion and message prioritization; (c) Network load balancing and traffic management; (d) Error detection and recovery procedures.

[0026] Real-time data exchange includes: (a) Vehicle speed and acceleration data; (b) Steering angle and torque information; (c) Wheel speed and brake pressure values; (d) Gear selection and transmission state; (e) Engine torque and throttle position.

[0027] The interface system 116 includes multiple interface components that enable driver interaction and system control. A primary human-machine interface (HMI) is implementedthrough the central multimedia display unit positioned in the vehicle's center console. This display provides visual feedback including parking space detection status, steering angle guidance, and distance warnings through a graphical user interface (GUI).

[0028] Visual feedback is supplemented by audible warnings through the vehicle’s audio system. The interface system 116 generates distinct audio tones that vary in frequency and repetition rate based on obstacle proximity.

[0029] The electronic control unit (ECU) processes sensor data and driver inputs to generate parking guidance.

[0030] The HMI display includes color-coded distance indicators and dynamic parking path projections overlaid on camera views from multiple vehicle-mounted cameras. The system automatically switches between relevant camera views based on vehicle position and parking maneuver phase.

[0031] The operating functions system 118 includes multiple integrated subsystems working in coordinated sequence to execute automated parking maneuvers.

[0032] The maneuver planning subsystem employs a path-generation algorithm that calculates optimal parking trajectories. The algorithm considers multiple variables including steering angle limitations, minimum turning radius, vehicle dimensions, and obstacle positions to generate a multi-point maneuvering path that minimizes the required number of direction changes.

[0033] Dynamic position tracking maintains continuous real-time monitoring of vehicle position relative to the target parking space through sensor fusion algorithms.

[0034] The system maintains operational logs recording key parameters including steering angles, speed profiles, and sensor readings during each parking maneuver. This data is utilized by the system's adaptive learning algorithms to optimize performance for frequently encountered parking scenarios.

[0035] FIG. 2 is a block diagram illustrating an example of an autonomous vehicle according to one embodiment.

[0036] The sensor system 110 may include one or more ultrasonic sensors 222 positioned around the vehicle to detect obstacles and measure distances. The one or more ultrasonic sensors 222 can provide real-time data to the system about the surroundings.

[0037] In addition to distance measurement, the ultrasonic sensors 222 are equipped with signal processing algorithms to filter out noise and enhance detection reliability. This ensures that the intuitive parking system 102 can accurately identify obstacles even in challenging environments, such as in the presence of multiple reflective surfaces or varying ambient conditions.

[0038] The integration of ultrasonic sensors 222 with the intuitive parking system 102 enables automated steering and braking adjustments during parking maneuvers. The ultrasonic sensors 222 continuously monitors the vehicle’s surroundings, providing feedback to the central control unit 232, which in turn, guides the vehicle into the parking space with minimal driver intervention. This technology significantly reduces the risk of collisions and enhances overall parking efficiency.

[0039] The control unit system 112 includes a central control unit 232, a steering control module 234, and a brake control module 236. Central control unit 232 can process multiple sensor inputs and coordinate automated parking operations in real-time.

[0040] The central control unit 232 incorporates multiple communication interfaces.

[0041] The central control unit 232 implements a three-layer software architecture: (a) A hardware abstraction layer managing direct communication with sensors and actuators; (b) A middleware layer handling data processing and parking space calculations; (c) An application layer executing parking algorithms and managing user interface interactions.

[0042] A decision-making algorithm evaluates parking space suitability based on multiple parameters including: (a) Vehicle dimensions and turning radius; (b) Required maneuvering space; (c) Obstacle positions and movements; (d) Surface gradient and texture analysis.

[0043] A steering control module 234 includes an electronic control unit (ECU) dedicated to managing the electric power steering system during automated parking operations.

[0044] The steering control module 234 implements a cascade control architecture with: (a) An outer loop controller managing steering angle positioning; (b) An inner loop controller regulating motor torque; (c) A feedforward compensation system for dynamic response optimization.

[0045] The brake control module 236 implements a multi-mode control strategy: (a) Normal operation mode with smooth brake modulation; (b) Emergency stop mode with maximum deceleration capability; (c) Hold mode for maintaining vehicle position on inclines.

[0046] The operating functions system 118 includes a parking space detection module 242, automatic functions 244, and a parking space recommendation module 246.

[0047] Some or all of modules such as parking space detection module 242 and automatic functions 244 may be implemented in software, hardware, or a combination thereof. For example, these modules may be installed in persistent storage device 260, loaded into memory 262, and executed by one or more processors (not shown). Note that some or all of these modules may be communicatively coupled to or integrated with some or all modules of control unit system 112. Some of modules such as parking space detection module 242 and automatic functions 244 may be integrated together as an integrated module.

[0048] Parking space detection module 242 employs a dual-mode operational framework based on parking area classification including known parking areas and unknown parking areas (will be described in connection with FIGs. 8 and 9).

[0049] Parking space detection module 242 includes spatial analysis system that continuously evaluates the vehicle's surroundings during low-speed operation. The parking space detectionmodule 242 employs a multi-sensor fusion architecture to identify and measure potential parking spaces with high precision.

[0050] The parking space detection module 242 implements a two-phase scanning protocol. During the initial phase, the parking space detection module 242 performs a rapid preliminary scan to identify potential parking spaces exceeding minimum dimensional thresholds.

[0051] Space classification algorithms categorize detected spaces into parallel, perpendicular, and angled parking configurations. The parking space detection module 242 applies specific measurement criteria for each parking type.

[0052] Real-time environmental mapping creates a dynamic obstacle buffer zone around detected spaces. The parking space detection module 242 continuously updates this buffer zone to account for moving objects and adjusts space availability status accordingly. A predictive algorithm estimates the movement trajectories of nearby vehicles to prevent space detection errors.

[0053] The parking space detection module 242 incorporates a learning algorithm that adapts detection parameters based on successful parking operations. This self-optimization process refines space measurement accuracy and improves detection reliability for challenging parking scenarios including curved curbs and irregular space boundaries.

[0054] Detected parking space data is processed through multiple validation layers before being presented to the driver. These validation checks assess factors including surface grade, curb height, and lateral clearance to ensure the space is suitable for the specific vehicle's dimensions and maneuvering capabilities.

[0055] The parking space detection module 242 maintains a temporary spatial memory buffer that records recently detected parking spaces, enabling quick re-validation when the driver decides to reverse to a previously identified space. This feature reduces parking space detection module 242 response time and improves the overall parking assistance experience.

[0056] The automatic functions 244 includes multiple integrated control subsystems that work in coordinated operation to execute precise parking maneuvers while maintaining safety parameters.

[0057] The collision avoidance subsystem operates continuously during parking maneuvers, maintaining a dynamic safety envelope around the vehicle.

[0058] These automated functions operate under a central control architecture that maintains synchronized operation and priority-based decision making.

[0059] Safety override protocols are maintained across all automated functions, allowing immediate driver intervention through steering input, brake application, or system deactivation button engagement. The system records all automatic function operations in a temporary memory buffer for diagnostic and optimization purposes.

[0060] FIG. 3 is a block diagram 300 illustrating an example of a parking space recommendation module used with the intuitive parking system according to one embodiment.

[0061] As described above, once the parking prompt is displayed and the driver interacts with parking prompt, intuitive parking system transitions to Stage 2, which addresses the question of how to park. This stage involves several considerations, primarily focusing on a machine learning-based approach.

[0062] The parking spot recommendation module 246 learns from users’ past parking habits and predict the most suitable parking spots based on historical parking decisions. The parking spot recommendation module 246 employs a combination of machine learning (specifically, an SGD Classifier) and rule-based approaches to rank available parking spots according to user preferences. Note the terms “parking spot” and “parking space” are generally used interchangeably and both may refer to a designated area where a vehicle can be parked, whether in a parking lot, garage, or on a street.

[0063] The parking spot recommendation module 246 includes data collector 372, feature processor 374, machine learning model trainer 376, predictor 378.

[0064] The prediction process begins with the feature processor 374 extracts all relevant features for each parking spot. When new parking spots become available, the feature processor 374 first extracts all relevant features for each spot. The features used in this model are categorized into numerical and categorical feature groups.

[0065] FIG. 4 is a block diagram illustrating an example of a feature processor according to one embodiment.

[0066] In some embodiments, feature processor 374 may determine a binary representation for a second type feature. Referring to FIG. 4, for example, one-hot encoders 480 encode categorical features.

[0067] One-hot encoders 480 include parking slot type encoder 481 , maneuver direction encoder 482, compass direction encoder 483, and parking slot attribute encoder 484.

[0068] Categorical features include parking slot type feature, maneuver direction feature, compass direction feature, and parking slot attribute feature.

[0069] Parking slot type feature describes the type of parking spot available. Examples include right box parking, left box parking, parallel parking, and more. Each type has specific characteristics that may influence the driver's preference based on ease of maneuvering and personal comfort.

[0070] Maneuver direction feature indicates whether parking requires driving forward or backward into the spot. Users often have a strong preference for one direction based on their comfort level, which significantly influences their parking decisions.

[0071] Compass direction feature may divide the 360° compass degree into five meaningful categories: front, back, left, right, and other. This categorization helps capture user preferencesfor parking spots in specific orientations, enhancing the accuracy of parking spot recommendations.

[0072] Parking slot attribute feature differentiates between regular spots and special spots, such as disabled parking. Parking slot feature can also indicate that a parking spot is equipped with facilities for charging electric vehicles. It ensures that the system provides appropriate spot recommendations based on user eligibility, thereby enhancing the overall user experience.

[0073] In some embodiments, feature processor 374 may determine a scaled cumulative value for a first type feature based on new input data. For example, feature processor 374 relies on the new data points to normalize numerical features for an update. Referring to FIG. 4, for example, cumulative scalers 490 scale numerical features using cumulative scaling factors.

[0074] Still referring to FIG. 4, cumulative scalers include free space width scaler 491 , distance from vehicles scaler 492, number of passengers scaler 493, number of strokes scaler 494.

[0075] Cumulative scalers scale numerical features including free space width feature, distance from vehicle, number of passengers feature, number of strokes feature.

[0076] Free space width feature measures the extra space between the car and the spot boundaries, calculated as (parking spot width - vehicle width). Larger values indicate more comfortable parking conditions, as some users consistently choose spots with more free space.

[0077] Distance from vehicle feature measures how far the parking spot is from the current vehicle position, in centimeters. Users often prefer closer spots, especially under certain conditions.

[0078] Number of passengers feature records the number of occupants in the car during parking. It is relevant because it might influence spot choice, with larger spots usually being selected when there are more passengers.

[0079] Number of strokes feature indicates the number of steering maneuvers needed to park, which reflects parking difficulty. Lower numbers mean easier parking.

[0080] Cumulative scalers 490 function like a running average calculator for each numerical feature. During each new parking event, the cumulative scalers 490 take the new values (such as free space feature and distance from vehicle feature) and updates its understanding of what is “normal” for these values, adjusting its scaling accordingly.

[0081] For example, if most parking spots are 300-400 cm away, a spot that is 1000 cm away may be considered “far.” Conversely, if most spots are 800-1000 cm away, a 400 cm spot is considered “close.” This dynamic scaling enables the model to comprehend relative differences rather than absolute values.

[0082] One-hot Encoding can convert categorical feature such as parking slot type feature, maneuver direction feature, compass direction feature, parking slot attribute feature into a format that can be provided to machine learning algorithms to improve predictions. For example, parking slot type encoder encodes right box parking as [1 , 0, 0], left box parking as [0, 1, 0], and parallel parking as [0, 0, 1], In another example, Maneuver direction encoder 482 encodes forward as [1 , 0] and backward as [0, 1], In a further example, Compass direction encoder encodes front as [1 , 0, 0, 0, 0], back as [0, 1, 0, 0, 0], left as [0, 0, 1, 0, 0], right as [0, 0, 0, 1, 0], other as [0, 0, 0, 0, 1], In another example, parking slot encoder 484 encodes regular as [1 , 0] and disabled as [0, 1], In each of these examples, one-hot encoders 480 transform each categorical feature into a binary vector where only one element is “1” (indicating the presence of that category) and all other elements are “0”. This allows the machine learning model to process categorical data effectively.

[0083] FIG. 5 is a flowchart depicting feature extraction, training, prediction for determining parking spot recommendations.

[0084] Data collector 372 collects input data 551. For example, data collector 372 collects parking entry 552 such as available parking spaces. Based on the parking entry 552, data collector 372 generates available candidates 553 for the training flow 555. Data collectorgenerates parking link chosen spot 554 based on the parking entry 552. Based on the parking link chosen spot 554, data collector 372 generates available candidates 553 for the training flow 555

[0085] After every parking event, regardless of whether it is rule-based or machine learningbased, and whether it is manual or automatic, machine learning model trainer 376 performs a training process as indicated by training flow 555. This training process is based on the parking spot selected by the user and the other candidate spots that were ignored.

[0086] After each parking event, machine learning model trainer 376 initiates a training process. Machine learning model trainer 376 examines the chosen parking spot and all available alternatives. At block 556, machine learning model trainer 376 then creates training data by labeling the chosen spot as “1” (positive example) and all other available spots as “0” (negative examples). The machine learning model is updated incrementally based on the training data, and the feature scalers are adjusted with new values.

[0087] In some embodiments, positive examples include instances where the desired outcome is achieved. In the case of parking spot selection, a positive example is the parking spot chosen by the user. Labeling this spot as “1” helps the machine learning model learn the characteristics and features that make a parking spot preferable to the user. Over time, the machine learning model uses these positive examples to identify patterns and preferences, improving its ability to suggest suitable spots in the future.

[0088] In some embodiments, negative examples include instances where the desired outcome is not achieved. For parking spot selection, negative examples are the spots that were available but not chosen by the user. Labeling these spots as “0” helps the machine learning model understand which features and characteristics are less desirable. By learning from these negative examples, the model can avoid suggesting similar spots in the future, thereby refiningits predictions. This distinction is essential for the machine learning model to make accurate and personalized recommendations based on the user's preferences and past behavior.

[0089] In some embodiments, at block 557, the system employs a Stochastic Gradient Descent (SGD) classifier for the training process. The SGD classifier is used for binary classification, predicting whether a parking spot will be chosen (labeled as “1”) or not (labeled as “0”). The SGD classifier utilizes a logistic regression loss function, which is effective for binary classification tasks. The SGD classifier is trained based on the numerical feature scaling 561 and categorical feature 562 extracted via feature extraction 559 during feature processing 558.

[0090] In some embodiments, SGD classifier may receive new input data associated with a parking event for a training of the trained prediction model. The SGD classifier may iteratively modify parameters of the trained prediction model based on the new input data associated with the parking event. The SGD classifier may adapt the trained prediction model in real-time as additional input data associated with the parking event becomes available. The trained prediction model continuously learns based on adjustment of the parameters of the trained prediction model in response to the new input data associated with the parking event. SGD classifier may discard the new input data after the training of the trained prediction model completes. For example, each training process updates the machine learning model using only newly acquired data. Each training iteration is performed exclusively on a current batch of new data without requiring access to previously processed training data. The system maintains no persistent storage of past training datasets after their respective training iterations are completed. The system updates model parameters while preserving previously acquired knowledge representations without requiring retraining on historical data.

[0091] One of the key advantages of the SGD classifier is its ability to incrementally train with every new entry. This means that the model can continuously learn and update its parameters as new data becomes available, ensuring that it remains accurate and up to date.

[0092] Additionally, the SGD classifier generates probability scores or a ranking score for each of parking space candidates. The SGD classifier uses the probability scores to rank the parking space candidate. These probability scores indicate the likelihood of a parking space being chosen, allowing the system to provide the most relevant and personalized parking recommendations to the user. The ranking score is based on predefined priority rules and the trained prediction model.

[0093] The SGD classifier may calculate a score for a plurality of scoring factors based on a parking history profile of a user. The SGD classifier may determine a weight for each of the plurality of scoring factors. The SGD classifier may generate an ordered list of the plurality of parking space candidates based on the ranking score. The SGD classifier may select the selected parking space based on the ordered list of the plurality of parking space candidates.

[0094] During the prediction flow 563, the predictor 378 employs a dual-mode operational framework based on parking area classification including known parking areas and unknown parking areas.

[0095] In some embodiments, predictor 378 performs a prediction in two distinct phases: rulebased ranking 565 and ML model-based ranking 566 predictions.

[0096] At block 564, the predictor 378 determines if sufficient data is available for the ML model-based ranking 566 predictions. This determination is common especially when dealing with new users.

[0097] If it is determined that data is not sufficient for the ML model-based ranking 566 predictions, the predictor 378 activates a rule-based ranking 565 predictions. If it is determined that data is sufficient for the ML model-based ranking 566 predictions, the predictor 378 bypasses the rule-based ranking 565 predictions and activates the ML model-based ranking 566 predictions.

[0098] The rule-based ranking 565 predictions rely on predefined logical rules to provide parking space suggestions, ensuring quick and reliable suggestions based on basic criteria.

[0099] In some embodiments, predictor 378 may initiate the predefined priority rules for predefined parking events to generate initial input data. For example, predictor 378 initially applies rule-based prediction during the first N rounds of suggestions.

[0100] In some embodiments, the initial input data may include historical parking data. For example, predictor 378 select the direction of the parking (e.g., whether forward or backward) based on the learnt behavior and previously selected parking spots.

[0101] The predictor 378 may select a direction associated with the selected parking space based on the initial input data.

[0102] Predictor 378 uses rule-based phase to address the “cold start” problem before sufficient data is available for the ML-based predictions phase. Predictor 378 applies rulebased phase during the first N parking events, where N is configurable. During this period, predictor 378 uses the predefined priority rules to rank parking spots.

[0103] In one example, rule-based phase prioritizes front and back spots over side spots when facing forward or backward to an open spot. The priority value is higher when the available parking space includes a front parking space and back parking space when the vehicle is oriented facing forward or backward relative to the available parking space. In another example, rule-based phase prefers the opposite side of the steering wheel, which for left-hand drive (LHD) cars would always choose the right side. The priority value is higher when the available parking space is located on an opposite side of a steering wheel configuration of the vehicle. In a further example, rule-based phase favors spots requiring fewer steering maneuvers or strokes and prefers closer spots when other factors are equal.

[0104] In an additional example, rule-based phase assigns a higher priority value when a distance between the available parking space and a current position of the vehicle is below a predefined distance.

[0105] Further, in some examples, rule-based phase assigns a higher priority value when the available parking space having characteristics matching user-specific requirements. The characteristics include at least one of: a disabled designation, or an electric vehicle charging capability.

[0106] As the user interacts with the system, predictor 378 gradually learns the user parking patterns through multiple parking events. The data collector 372 collects input data 551 on the user’s choices and preferences, allowing the system to build a comprehensive understanding of their parking habits.

[0107] In some embodiments, predictor 378 may apply the initial input data to the trained prediction model after the predefined parking events occur. The train prediction model learns from the selected parking direction and maneuver as described above.

[0108] For example, after the rule-based ranking 566 predictions, predictor 378 activates the machine ML model-based ranking 566 predictions, as the model will have accumulated sufficient history to learn from and make predictions based on the user’s parking habits. Once sufficient data has been gathered, the predictor 378 starts recommending parking spots that match these learned preferences. By leveraging historical data, the predictor 378 can make more accurate and personalized predictions, enhancing the overall user experience.

[0109] In some embodiments, the initial input data may include completed parking events for the vehicle.

[0110] ML model-based ranking 566 predictions begin after N parking events. ML-based predictions phase uses past parking choices to predict future preferences, offering a more sophisticated approach than the rule-based phase. The machine learning model can learncomplex combinations of preferences, adapt to individual user patterns, and consider all features simultaneously.

[0111] The predictor 378 continues to adapt if the user’s preferences change over time.Continuous learning ensures that the system remains responsive to evolving habits, updating its recommendations to reflect the user's current preferences.

[0112] The predictor 378 then generates a ranking for the parking space. If the system is in the rule-based phase, predictor 378 applies priority rules. If it is in the machine learning phase, predictor 378 uses the machine learning model to predict the probability for each parking space.

[0113] At block 567, the predictor 378 sorts the parking spaces by ranking or probability. The predictor generates ranked candidates based on the ranking or the probability. Based on the ranked candidates, the predictor 378 recommends the top parking space to the user via a parking prompt.

[0114] FIG. 6 is a diagram 600 of a user journey in conjunction with the intuitive parking system according to one embodiment.

[0115] In some embodiments, the intuitive parking system 102 may determine using the trained prediction model and based on the plurality of parking space candidates, a selected parking space within the predefined geofenced area. The selected parking space may be selected based on an occupancy status of the plurality of parking space and a type of the predefined geofenced area. For example, the intuitive parking system 102 may select a disabled parking spot if the driver frequently parks in disabled areas. In some other examples, the intuitive parking system 102 may select a parking spot that matches user requirements with spot-specific attributes (e.g., electric vehicle (EV) type). The intuitive parking system 102 may define “available spots” as regular parking spaces by default. However, as the driver parks morefrequently, the system learns their preferences and adapts accordingly. For example, the system can also consider dedicated parking areas, depending on the driver’s past behavior.

[0116] In some embodiments, in response to determining the selected parking space within the predefined geofenced area, the intuitive parking system 102 may activate a notification. The intuitive parking system 102 may include a graphical user interface (GUI) 602 that displays a notification or a prompt when the highest-ranking parking space is determined. For example, the GUI displays a notification notifying the user that a parking space near the destination has been found.

[0117] In some embodiments, the intuitive parking system 102 displays a notification when the location of the vehicle and the determined parking space is within a predefined distance. The system automatically selects the best available parking spot based on predefined criteria and user preferences. This selection is facilitated by a model that considers known parking areas and driver parking history as described above. For example, approximately 20 meters from the parking destination, the intuitive parking system 102 initiates the camera and parking application 604 to start the automatic parking application. In another example, the user initiates the camera and parking application 604 to start the automatic parking application.

[0118] A pop-up recommendation may be displayed to prepare the user for the auto-parking initiation. This pop-up includes an interactive button that the user can click to confirm the selection. Implementing this feature is expected to significantly increase the adoption rate of the parking assistance functionality. For example, referring to Fig 6, the intuitive parking system 102 displays a start automatic parking pop-up 606 displaying the optimal parking space recommendation including options “dismiss” and “accept” for the user to respond. The user can provide a user response responsive to the notification. For example, the user can initiate parking by clicking “accept,” which triggers the parking assistance features. On the other hand, the user can decline the recommendation by clicking “dismiss.”

[0119] In some embodiments, the system generates an in-app prompt while the user is using a navigation application. For example, referring to FIG. 6, the system displays a pop-up prompt in the navigation application 608 providing a real-time update on the recommended parking space and allows the user to accept or decline the suggestion directly within the navigation application.

[0120] Note that various notification methods can be utilized to provide the optimal parking space recommendation to the user. In one example, for users with voice-enabled devices, the system can provide voice alerts announcing the recommended parking space. The voice alert includes details about the parking spot and prompts the user to confirm their choice using voice commands. In another example, in vehicles equipped with advanced infotainment systems, the recommendation can be displayed on the dashboard screen. The display includes a map highlighting the recommended parking space and an option to navigate to the spot. In a further example, the system sends a push notification to the user's mobile device, alerting them of the recommended parking space. The notification includes a brief description of the parking spot and a link to open the detailed pop-up interface.

[0121] In some embodiments, the autonomous parking assist can determine a navigation path based on the user response. The autonomous parking assist can navigate the vehicle to the selected parking space within the predefined geofenced area based on the navigation path. Note that “geofenced area” may refer to an area surrounding previous destinations or a target destination during a navigation. Throughout this disclosure, the terms “geofenced area” and “parking area” may be used interchangeably and should be understood to refer to the same component.

[0122] FIG. 7 is an exemplary recommendation of the highest-ranking parking space selected based on the machine learning model according to one embodiment.

[0123] In some embodiments, the system utilizes historical data to recommend parking spaces based on the user’s past preferences and predefined criteria. The system presents the parkingspace recommendations via a pop-up interface as described above, which includes an interactive button for the user to confirm the recommended parking space 704. The system’s ability to learn and adapt to the user's parking preferences aims to enhance user satisfaction and increase the adoption rate of the parking assistance feature.

[0124] The system creates geofences around common locations where parking is typically needed, such as grocery stores and offices. These geofences are defined by road boundaries and vary in size based on frequency of use.

[0125] Points of interest are highlighted on a map with icons or markers, aiding in quick identification of potential parking areas.

[0126] Parking Spot Detection and Prompt Activation:

[0127] Within the geofence, the system detects available parking spots using symbols or labels. As the vehicle enters the geofence, a parking prompt is activated, guiding the user through the parking process.

[0128] FIG. 8 depicts an example of determination of known parking areas according to one embodiment.

[0129] In some embodiments, known parking areas include areas associated with a parking history profile of a user. Parking space detection module 242 can recognize these areas through various indicators. In one example, parking space detection module 242 utilizes active route guidance, where the vehicle is navigating towards a destination using the navigation system. In another example, parking space detection module 242 considers the proximity to past or favorite destinations, identifying locations frequently visited by the driver or marked as favorites. This information is obtained by recording the driver’s past visited destinations. In a further example, parking space detection module 242 considers the proximity to previous parking locations, recognizing areas where the driver has parked before by maintaining records of past parking spots.

[0130] Parking space detection module 242 establishes virtual perimeter boundaries (geofences) with variable radii around identified known parking areas. Geofence activation triggers are determined by pre-programmed baseline parameters and machine learning algorithms incorporating historical parking patterns.

[0131] In some embodiments, the radius of the geofence can be predefined based on practical considerations and dynamically adjusted overtime using machine learning model to learn from the driver’s parking history 802 and preferences. For example, referring to FIG. 8, parking space detection module 242 determines the geofence radius by considering the frequency and certainty of the driver’s visits to specific locations.

[0132] For past parking spots, the radius can be adjusted based on the frequency of parking in that area and the size of the entire parking lot. For example, if a parking lot is 100 square meters and the driver has parked there three times, parking space detection module 242 can define a geofence 804 with a radius of 200 meters around the center of the parking lot. In another example, a smaller and more concentrated radius (e.g., 100 meters) of the geofence 806 can be used for frequently visited or navigated destinations. Once the driver enters this geofence, parking space detection module 242 sends a signal to the interface system 116 to display a parking prompt to the user.

[0133] FIG. 9 depicts an example of detection of unknown parking areas according to one embodiment.

[0134] In some embodiments, predefined geofenced area may include unknown parking areas. For example, unknow parking areas include locations where the driver has no prior parking history. In some embodiments, predefined geofenced area may include an area associated with vehicular parking. The parking space detection module 242 identifies such areas using GPS and detects when the vehicle enters a parking-only location or a place where parking is typically needed, such as grocery stores, charging stations, offices, and parking garages. In someembodiments, predefined geofenced area may include an area with a parking density exceeds a predefined threshold. For example, parking space detection module 242 identifies locations where there is a high demand for parking due to the concentration of people and activities. For example, areas with a parking density exceeds a predefined threshold shopping include shopping malls, office complexes, airports, stadiums, downtown areas, universities, hospitals, and public transport hubs. Shopping malls attract large numbers of shoppers, especially during weekends and holidays, leading to high parking demand. Office complexes experience heavy parking usage as employees and visitors require parking spaces. Airports have significant parking needs with travelers, often with multiple parking lots and garages. Stadiums see a surge in parking demand during events such as sports games or concerts. Downtown areas often have limited parking availability due to high density of businesses, restaurants, and entertainment venues. Universities require extensive parking facilities for students, faculty, and visitors. Hospitals need ample parking for patients, staff, and visitors, often operating 24 / 7. Public transport hubs require parking for commuters who use public transportation.

[0135] Similar to known areas as described above, a geofence is created around these locations. The parking prompt is only shown if available spots are detected within this geofence. For example, referring to FIG. 9, the system displays the pop-up to the user as the vehicle enters the radius of the geofence configured around the grocery store. In some embodiments, the radius of the geofence for these areas may be defined by the road boundaries around the point of interest (POI), considering the size of the parking lot. For example, as illustrated in FIG.9, the system determines the radius of the geofence around the office 904 is larger than the geofence around the grocery store 902 because the office’s parking lot is larger than the grocery store’s parking lot.

[0136] FIG. 10 is a flow diagram illustrating a process of machine learning model-based parking space determination according to one embodiment.

[0137] Process 1000 may be performed by processing logic which may include software, hardware, or a combination thereof. For example, process 1000 may be performed by parking space recommendation module 246 of FIG. 3. Referring to FIG. 10, in operation 1001, processing logic determines, by a processor based on a current location of a vehicle, that the vehicle has entered a predefined geofenced area. In operation 1002, processing logic determines, by the processor using a trained prediction model, a plurality of parking space candidates within the predefined geofenced area. In operation 1003, processing logic determines, by the processor using the trained prediction model and based on the plurality of parking space candidates, a selected parking space within the predefined geofenced area. The selected parking space may be selected based on an occupancy status of the plurality of parking space and a type of the predefined geofenced area. The selected parking space may also be selected based on ranking score provided by the prediction model. In operation 1004, processing logic determines, by the processor, a first binary representation for the plurality of parking space candidates indicating the selected parking space. In operation 1005, processing logic determines, by the processor, a second binary representation for the plurality of parking space candidates indicating one or more unselected parking spaces, wherein the first binary representation includes a value of 1 , and wherein the second binary representation includes a value of 0. The first and second binary representations are labels that may be assigned to each parking space. Label 1 may be assigned to a selected parking space and label 0 may be assigned to the unselected parking space. The label 1 and label 0 may be used as new input data to train the prediction model. In operation 1006, processing logic, in response to determining the selected parking space within the predefined geofenced area, activates, by the processor, a notification. In operation 1007, processing logic receives, by the processor, a user response to the notification. In operation 1008, processing logic navigates, by the processor,the vehicle to the selected parking space within the predefined geofenced area based on the user response.

[0138] Note that some or all of the components as shown and described above may be implemented in software, hardware, or a combination thereof.

[0139] FIG. 11 is a block diagram illustrating an example of a data processing system which may be used with one embodiment of the disclosure. For example, system 1100 may represent any of data processing systems described above performing any of the processes or methods described above, such as, for example, operating functions system 118 or any of servers 104 and servers 106 of FIG. 1. System 1100 can include many different components. These components can be implemented as integrated circuits (ICs), portions thereof, discrete electronic devices, or other modules adapted to a circuit board such as a motherboard or add-in card of the computer system, or as components otherwise incorporated within a chassis of the computer system.

[0140] In one embodiment, system 1100 includes processor 1101, memory 1103, and devices 1105-1108 connected via a bus or an interconnect 1110. Processor 1101 may represent a single processor or multiple processors with a single processor core or multiple processor cores included therein.

[0141] System 1100 may further include a graphics interface that communicates with optional graphics subsystem 1104, which may include a display controller, a graphics processor, and / or a display device.

[0142] Processor 1101 may communicate with memory 1103, which in one embodiment can be implemented via multiple memory devices to provide for a given amount of system memory.

[0143] Input device(s) 1106 may include a mouse, a touch pad, a touch sensitive screen (which may be integrated with optional graphic subsystem 1104), a pointer device such as astylus, and / or a keyboard (e.g., physical keyboard or a virtual keyboard displayed as part of a touch sensitive screen). For example, input device 1106 may include a touch screen controller coupled to a touch screen.

[0144] IO devices 1107 may include an audio device. An audio device may include a speaker and / or a microphone to facilitate voice-enabled functions, such as voice recognition, voice replication, digital recording, and / or telephony functions.

[0145] Storage device 1108 may include computer-accessible storage medium 1109 (also known as a machine-readable storage medium or a computer-readable medium) on which is stored one or more sets of instructions or software (e.g., module, unit, and / or logic 1128) embodying any one or more of the methodologies or functions described herein. Processing module / unit / logic 1128 may represent any of the components described above, such as, for example, parking space detection module 242, and / or parking space recommendation module 246. Processing module / unit / logic 1128 may also reside, completely or at least partially, within memory 1103 and / or within processor 1101 during execution thereof by data processing system 1100, memory 1103 and processor 1101 also constituting machine-accessible storage media. Processing module / unit / logic 1128 may further be transmitted or received over a network via network interface device 1105.

[0146] Computer-readable storage medium 1109 may also be used to store some software functionalities described above persistently.

[0147] Processing module / unit / logic 1128, components and other features described herein can be implemented as discrete hardware components or integrated in the functionality of hardware components such as ASICS, FPGAs, DSPs or similar devices. In addition, processing module / unit / logic 1128 can be implemented as firmware or functional circuitry within hardware devices. Further, processing module / unit / logic 1128 can be implemented in any combination hardware devices and software components.

[0148] In the foregoing specification, embodiments of the disclosure have been described with reference to specific exemplary embodiments thereof. It will be evident that various modifications may be made thereto without departing from the broader spirit and scope of the disclosure as set forth in the following claims. The specification and drawings are, accordingly, to be regarded in an illustrative sense rather than a restrictive sense.

Claims

Mercedes-Benz Group AGCLAIMSWhat is claimed is:

1. A parking assistance method, comprising:determining, by a processor based on a current location of a vehicle, that the vehicle has entered a predefined geofenced area;determining, by the processor using a trained prediction model, a plurality of parking space candidates within the predefined geofenced area;determining, by the processor using the trained prediction model and based on the plurality of parking space candidates, a selected parking space within the predefined geofenced area, the selected parking space being selected based on an occupancy status of the plurality of parking space and a type of the predefined geofenced area;determining, by the processor, a first binary representation for the plurality of parking space candidates indicating the selected parking space;determining, by the processor, a second binary representation for the plurality of parking space candidates indicating one or more unselected parking spaces, wherein the first binary representation includes a value of 1 , and wherein the second binary representation includes a value of 0;in response to determining the selected parking space within the predefined geofenced area, activating, by the processor, a notification;receiving, by the processor, a user response to the notification;andnavigating, by the processor, the vehicle to the selected parking space within the predefined geofenced area based on the user response.

2. The method of claim 1 , wherein determining the selected parking space within the predefined geofenced area further comprises:generating, by the processor, a ranking score for each of the plurality of parking space candidates, wherein the ranking score is based on predefined priority rules and the trained prediction model, wherein generating the ranking score further comprises:calculating, by the processor, a score for a plurality of scoring factors based on a parking history profile of a user, anddetermining, by the processor, a weight for each of the plurality of scoring factors; generating, by the processor, an ordered list of the plurality of parking space candidates based on the ranking score; andselecting, by the processor, the selected parking space based on the ordered list of the plurality of parking space candidates.

3. The method of claim 2, further comprising:initiating, by the processor, the predefined priority rules for predefined parking events to generate initial input data, wherein the initial input data includes historical parking data;selecting, by the processor, a direction associated with the selected parking space based on the initial input data;applying, by the processor, the initial input data to the trained prediction model after the predefined parking events occur; andtraining, by the processor, the trained prediction model based on the initial input data.

4. The method of claim 3, wherein the initial input data include completed parking events for the vehicle, andwherein initiating the predefined priority rules includes:assigning, by the processor, a higher priority value to an available parking space based on the predefined priority rules including at least one of:the priority value is higher when the available parking space includes a front parking space and back parking space when the vehicle is oriented facing forward or backward relative to the available parking space;the priority value is higher when the available parking space is located on an opposite side of a steering wheel configuration of the vehicle;the priority value is higher when the available parking space requires a predefined steering maneuvers being below a threshold;the priority value is higher when a distance between the available parking space and a current position of the vehicle is below a predefined distance; orthe priority value is higher when the available parking space having characteristics matching user-specific requirements, wherein the characteristics include at least one of: a disabled designation, or an electric vehicle charging capability.

5. The method of claim 1 , further comprising:determining, by the processor, a scaled cumulative value for a first type feature based on new input data;determining, by the processor, a binary representation for a second type feature; and training, by the processor, the trained prediction model using a stochastic gradient descent (SGD) based on the scaled cumulative value and the binary representation.

6. The method of claim 1 , further comprising:receiving, by the processor, new input data associated with a parking event for a training of the trained prediction model;iteratively modifying, by the processor, parameters of the trained prediction model based on the new input data associated with the parking event;adapting, by the processor, the trained prediction model in real-time as additional input data associated with the parking event becomes available, wherein the trained prediction model continuously learns based on adjustment of the parameters of the trained prediction model in response to the new input data associated with the parking event; anddiscarding the new input data after the training of the trained prediction model completes.

7. The method of claim 1, wherein the predefined geofenced area includes one or more of:an area associated with a parking history profile of a user;an area with a parking density exceeds a predefined threshold; oran area associated with vehicular parking.

8. A parking assistance system, comprising:a processor, anda memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations, the operations comprising:determining, based on a current location of a vehicle, that the vehicle has entered a predefined geofenced area;determining, by using a trained prediction model, a plurality of parking space candidates within the predefined geofenced area;determining, by using the trained prediction model and based on the plurality of parking space candidates, a selected parking space within the predefined geofenced area, the selectedparking space being selected based on an occupancy status of the plurality of parking space and a type of the predefined geofenced area;determining a first binary representation for the plurality of parking space candidates indicating the selected parking space;determining a second binary representation for the plurality of parking space candidates indicating one or more unselected parking spaces, wherein the first binary representation includes a value of 1, and wherein the second binary representation includes a value of 0; in response to determining the selected parking space within the predefined geofenced area, activating a notification;receiving a user response to the notification;andnavigating the vehicle to the selected parking space within the predefined geofenced area based on the user response.

9. The system of claim 8, wherein determining the selected parking space within the predefined geofenced area further comprises:generating a ranking score for each of the plurality of parking space candidates, wherein the ranking score is based on predefined priority rules and the trained prediction model, wherein generating the ranking score further comprises:calculating a score for a plurality of scoring factors based on a parking history profile of a user, anddetermining a weight for each of the plurality of scoring factors;generating an ordered list of the plurality of parking space candidates based on the ranking score; andselecting the selected parking space based on the ordered list of the plurality of parking space candidates.

10. The system of claim 9, further comprising:initiating the predefined priority rules for predefined parking events to generate initial input data, wherein the initial input data includes historical parking data;selecting a direction associated with the selected parking space based on the initial input data;applying the initial input data to the trained prediction model after the predefined parking events occur; andtraining the trained prediction model based on the initial input data, wherein each training iteration is performed on a current batch of the initial input data without requiring access to previously processed batch of the initial input data.