An intelligent vehicle lamp control system with multi-mode interactive feedback function

Abstract

The application discloses a kind of intelligent car light control systems with multimode interactive feedback function, it is related to intelligent driving and vehicle control technical field, including perception unit, configured to obtain the perception data of the environment around vehicle, for identifying external traffic participants and extracting its behavior characteristic data;Intention evaluation module generates intention ambiguity quantization result according to the behavior characteristic data.This application cooperates with perception unit and intention evaluation module, establishes intention ambiguity evaluation mechanism based on position oscillation mode, speed oscillation mode, head orientation angle and light language response consistency and other multi-dimensional feature weighted quantization.Can distinguish between the deceptive game behavior of pedestrian and real hesitation, the identification accuracy is greatly improved compared with existing binary classification scheme, effectively solve the technical problems that existing technology cannot distinguish between pedestrian strategic behavior deviation and passive intention ambiguity, provide fine quantization input for subsequent game state determination.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving and vehicle control technology, specifically to an intelligent vehicle lighting control system with multi-modal interactive feedback function. Background Technology

[0002] In the scenario of a traffic-free zebra crossing at night, the allocation of right-of-way between vehicles and pedestrians highly depends on the real-time observation and implicit negotiation of each other's intentions. Pedestrians often express uncertain or strategic crossing intentions through subtle changes in movement (such as taking a step forward and then back, frequently switching gazes, and fluctuating speed). At the same time, vehicle lights become the only reliable information exchange channel in the nighttime environment, and intelligent vehicle lights can convey driving intentions to pedestrians by projecting different light patterns. In this scenario, the system's accurate understanding of pedestrian behavior patterns and adaptive selection of interaction strategies directly affect traffic efficiency and safety, placing high demands on the refined quantification of pedestrian intentions and the ability to perceive interaction states.

[0003] However, existing vehicle-to-infrastructure (V2I) interaction methods are unable to solve a deep-seated technical problem in the above scenario: pedestrians may adopt legitimate game-theoretic strategies to deliberately create ambiguity of intent in order to gain an advantage in passage. Existing technologies not only cannot distinguish between genuine hesitation and deceptive hesitation, but also lack the ability to identify and respond to the strategic behavior of pedestrians based on game-theoretic motives. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent vehicle lighting control system with multi-modal interactive feedback function to address the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent vehicle lighting control system with multi-modal interactive feedback function, comprising: The sensing unit is configured to acquire sensing data of the environment surrounding the vehicle, for identifying external traffic participants and extracting their behavioral characteristic data. The intent assessment module generates an intent fuzziness quantification result based on the behavioral feature data. The intent fuzziness quantification result is used to characterize the degree to which the current behavior of external traffic participants deviates from the cooperative traffic pattern. The cooperative traffic pattern refers to the behavioral pattern of external traffic participants in accordance with traffic rules and conventional courtesy habits. The trust evolution module establishes and updates the vehicle's first level of trust in external traffic participants and the external traffic participants' second level of trust in the vehicle based on historical interaction data. The strategy decision-making module selects the corresponding light language interaction strategy based on the comprehensive evaluation of the intent ambiguity quantification result and the first trust level and the second trust level. The light language execution unit controls the pixelated headlights to execute the light projection pattern corresponding to the selected light language interaction strategy. The light projection pattern includes at least a deterministic guiding light language for maintaining cooperative equilibrium, a confirmatory probing light language for responding to strategic deviations, and a game information de-masking light language for breaking the game deadlock.

[0006] In a preferred embodiment, the sensing unit includes at least one or more of lidar, millimeter-wave radar, and cameras.

[0007] In a preferred embodiment, the behavioral feature data extracted by the sensing unit includes at least the position oscillation pattern, velocity oscillation pattern, head orientation and movement direction angle of the external traffic participant, and the consistency characteristics of the participant's response to the projected light signal.

[0008] In a preferred embodiment, the intent assessment module uses a multi-feature weighted quantization model to generate the intent ambiguity quantization result. The multi-feature weighted quantization model uses at least a portion of the normalized scores of position oscillation patterns, velocity oscillation patterns, line-of-sight and orientation deviations, and the normalized scores of consistency with light speech responses as input variables.

[0009] In a preferred embodiment, the trust evolution module updates the first trust level and the second trust level respectively using a Bayesian update rule, wherein: The first level of trust is updated based on the consistency of external traffic participants' responses to the vehicle's light signals; The second level of trust is updated based on the change in decision-making time of external traffic participants after the projection of vehicle headlight signals.

[0010] In a preferred embodiment, the game states identified by the strategy decision module include at least cooperative equilibrium, unilateral strategic shift, and two-way game deadlock, and the light language interaction strategy is selected based on the identified game states.

[0011] In a preferred embodiment, the light-text interaction strategy includes at least: Cooperation maintenance strategy: In the case of a state where the quantification result of low intention ambiguity is high and both the trust level of both sides is higher than the preset upper limit, control the light language execution unit to project deterministic guiding light language; Information verification strategy: corresponding to the state of the quantification result of medium intention ambiguity, control the optical language execution unit to project a verification probe optical language; Game-theoretic de-concealing strategy: For states with high intention ambiguity quantification results and determined to be game stalemates, control the light language execution unit to project light language containing game information de-concealing semantics with differentiated light effects.

[0012] In a preferred embodiment, the game-theoretic unmasking strategy further includes a tiered response mechanism: Level 1 Game Unmasking: Triggered when the quantification result of intent ambiguity is greater than the first threshold and less than the second threshold, a game state perception warning pattern is projected with a first brightness multiple and a first flashing frequency; Second-level game de-masking: Triggered when the quantification result of intent ambiguity is greater than or equal to the second threshold, the right-of-way information transparency pattern is projected at the second brightness multiple and the second flashing frequency; Wherein, the second brightness multiple is greater than the first brightness multiple, and the second flashing frequency is greater than the first flashing frequency.

[0013] In a preferred embodiment, the light-text interaction strategy further includes a safety fallback strategy: When the quantification result of intent ambiguity is greater than the third threshold and the second trust level is less than the preset trust lower limit, it is determined that the interactive consensus cannot be reached. The strategy decision module skips the game interaction steps, directly instructs the vehicle to execute the yield braking strategy, and controls the light language execution unit to project a red forced stop light pattern.

[0014] In a preferred embodiment, the trust evolution module further includes a trust trajectory recording submodule, configured as follows: The trajectory of the change between the first trust level and the second trust level is continuously recorded within a preset time window; When multiple interaction records of the same external traffic participant are detected within a historical time window, the trust trajectory of the historical interaction records is input into the trust evolution module as prior information for the current interaction, in order to improve the accuracy of intent ambiguity recognition and game state determination.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention, through the collaborative work of a sensing unit and an intent assessment module, establishes for the first time in the field of vehicle-to-infrastructure (V2I) communication a mechanism for assessing intent ambiguity based on multi-dimensional features weighted quantization, including position oscillation patterns, velocity oscillation patterns, head orientation angle, and consistency of light and speech responses. This mechanism can distinguish between deceptive game-theoretic behavior and genuine hesitation in pedestrians, significantly improving recognition accuracy compared to existing binary classification schemes. It effectively solves the technical problem of existing technologies being unable to distinguish between strategic behavioral deviations and passive intent ambiguity, providing refined quantitative input for subsequent game state determination.

[0016] This invention employs a closed-loop coupling between a trust evolution module and a strategy decision module. It uses Bayesian update rules to establish and dynamically evolve the vehicle's initial trust level towards external traffic participants and the secondary trust level of external traffic participants towards the vehicle. Combined with a trust trajectory recording submodule, it uses historical interaction records of trust trajectories as prior information for the current interaction. This mechanism enables the system to progressively improve the accuracy of predicting the behavior of the same pedestrian in multiple interactions and enhances the response speed to deceptive behavior. It effectively solves the technical problems of existing technologies, such as the lack of historical interaction knowledge accumulation and reuse, and the inability to break the deadlock in the game between vehicles and pedestrians, thus providing a cognitive foundation for game equilibrium convergence.

[0017] This invention, through a strategy decision-making module, adaptively selects a cooperation maintenance strategy, an information verification strategy, or a game-theoretic de-concealment strategy based on a comprehensive evaluation of the fuzziness quantification result and the two-way trust level. The light language execution unit then projects game-theoretic de-concealment light language with differentiated light effects using a hierarchical response mechanism, forcing strategic actors to abandon their ambiguous intentions and return to cooperative equilibrium. Simultaneously, a safety fallback strategy is implemented, forcing the vehicle to perform yield braking and project a red stop light pattern when interactive consensus cannot be reached. This solves the technical problem that existing vehicle light signals only serve as one-way inquiry or warning channels and lack game-theoretic constraint capabilities. In nighttime zebra crossing scenarios without traffic lights, it significantly improves the interaction success rate compared to existing technologies while maintaining an absolute safety boundary with zero collision risk. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1, please refer to Figure 1 As shown in the figure, the intelligent vehicle lighting control system with multi-modal interactive feedback function described in this embodiment includes: The sensing unit is configured to acquire sensing data of the environment surrounding the vehicle, for identifying external traffic participants and extracting their behavioral characteristic data. The intent assessment module generates an intent fuzziness quantification result based on the behavioral feature data. The intent fuzziness quantification result is used to characterize the degree to which the current behavior of external traffic participants deviates from the cooperative traffic pattern. The cooperative traffic pattern refers to the behavioral pattern of external traffic participants in accordance with traffic rules and conventional courtesy habits. The trust evolution module establishes and updates the vehicle's first level of trust in external traffic participants and the external traffic participants' second level of trust in the vehicle based on historical interaction data. The strategy decision-making module selects the corresponding light language interaction strategy based on the comprehensive evaluation of the intent ambiguity quantification result and the first trust level and the second trust level. The light language execution unit controls the pixelated headlights to execute the light projection pattern corresponding to the selected light language interaction strategy. The light projection pattern includes at least a deterministic guiding light language for maintaining cooperative equilibrium, a confirmatory probing light language for responding to strategic deviations, and a game information de-masking light language for breaking the game deadlock.

[0022] Example 2, based on Example 1 above, provides a detailed description of the specific structure of the sensing unit and its data acquisition mechanism. The sensing unit is configured to acquire sensing data of the vehicle's surrounding environment. Optionally, the sensing unit includes at least one or more of LiDAR, millimeter-wave radar, and cameras.

[0023] In some implementations, the sensing unit can use only a monocular camera as the data acquisition front end, and use image recognition algorithms to detect and track external traffic participants. In other implementations, in order to improve the reliability and robustness of detection, a multi-sensor fusion scheme can be adopted, such as deploying LiDAR and cameras simultaneously. LiDAR is used to obtain accurate distance and depth information, millimeter-wave radar can be used to supplement the detection of long-distance targets, and cameras are used to extract semantic-level behavioral features such as facial orientation and body movements.

[0024] Optionally, when multi-sensor fusion is used, the data from each sensor can be aligned by timestamps and then fused at the data level or feature level within the sensing unit to form a unified sensing data stream that is output to the intent evaluation module.

[0025] The sensing unit is further configured to identify external traffic participants and extract their behavioral characteristic data. Optionally, external traffic participants may include pedestrians, non-motorized vehicle riders, and other vehicle occupants.

[0026] In some implementations, the sensing unit can employ a deep learning-based object detection network, such as the YOLO series or Faster R-CNN, to perform instance segmentation and object classification on the image data acquired by the camera, thereby locating the pixel-level regions of external traffic participants.

[0027] Optionally, for lidar point cloud data, network structures such as PointPillars or PointNet++ can be used for 3D target detection to obtain the spatial position and velocity of the participant in the vehicle coordinate system. Behavioral feature data should include at least the positional oscillation pattern, velocity oscillation pattern, head orientation angle with the direction of movement, and the participant's consistent response to the projected optical signal.

[0028] Optionally, the positional oscillation pattern can be calculated by continuously tracking the sequence of positional changes of the participant in the traversing direction, and is represented by the variance or standard deviation of the positional data; the velocity oscillation pattern can be obtained by performing time-domain analysis on velocity sensor data, and is represented by the degree of fluctuation of the velocity curve; the angle between the head orientation and the direction of movement can be obtained by analyzing the posture data of human key points in the camera image, reflecting the consistency between the participant's gaze direction and the actual movement trend; the response consistency feature needs to be jointly analyzed by combining the timestamp of the projected light signal and the time delay of the participant's subsequent behavioral changes.

[0029] Example 3, based on the aforementioned examples, provides a detailed description of the specific working mechanism of the intent assessment module. The intent assessment module generates an intent fuzziness quantification result based on the behavioral feature data output by the perception unit. The intent fuzziness quantification result is used to characterize the degree to which the current behavior of external traffic participants deviates from the cooperative traffic pattern. Optionally, this quantification result can be output in continuous value form, with the numerical range set to a floating-point number between 0 and 1, where 0 indicates complete compliance with the cooperative traffic pattern, and 1 indicates a high degree of deviation from the cooperative traffic pattern, potentially indicating a strategic game-theoretic intent.

[0030] In some implementations, cooperative traffic patterns can be defined as behavioral patterns in which external traffic participants proceed in accordance with traffic rules and conventional courtesy, such as pedestrians slowing down and maintaining a steady pace at crosswalks, or non-motorized cyclists maintaining a reasonable lateral distance from lane lines.

[0031] The intent assessment module uses a multi-feature weighted quantization model to generate intent ambiguity quantification results. Optionally, the multi-feature weighted quantization model uses at least a portion of the normalized scores of position oscillation patterns, velocity oscillation patterns, line-of-sight and orientation deviations, and the normalized scores of consistency with light speech responses as input variables.

[0032] In some implementations, the normalization of each input variable can be achieved using the Min-Max normalization method, which maps the original feature data to the 0-1 interval. For example, the normalization score of the position oscillation mode can be obtained by calculating the ratio of the position variance of the current frame to the preset maximum position variance.

[0033] Optionally, the weighted quantization model can take the form of a linear weighted summation, and its mathematical expression can be: ,in The result indicates the fuzzy quantification of intent. This represents the weight coefficient of the i-th input feature. This represents the normalized score of the i-th input feature. The value of i can be an integer between 1 and 4, corresponding to the four features mentioned above.

[0034] Optionally, the weight coefficients can be determined based on statistical learning of historical interaction data, or they can be preset based on manual parameter tuning experience for specific application scenarios.

[0035] The intent ambiguity quantification results output by the intent assessment module can be used as input data for the subsequent trust evolution module and strategy decision module.

[0036] Specifically, the intent assessment module and the trust evolution module can be connected via a preset data interface, using the intent ambiguity quantification result as one of the input parameters for trust level updates. Optionally, the intent assessment module and the policy decision module can be connected via a policy trigger signal, triggering the policy decision module to perform the corresponding policy selection action when the intent ambiguity quantification result meets preset conditions.

[0037] Example 4, based on the foregoing examples, provides a detailed explanation of the specific working mechanism of the trust evolution module. The trust evolution module establishes and updates the vehicle's first level of trust in external traffic participants and the external traffic participants' second level of trust in the vehicle based on historical interaction data.

[0038] Optionally, the first trust level is used to characterize the vehicle's trust assessment of whether external traffic participants honestly express their intention to pass, and the value range can be set to a continuous value between 0 and 1, where 0 represents no trust and 1 represents full trust; the second trust level is used to characterize the external traffic participants' trust assessment of whether the vehicle complies with the passage agreement, and the value range can also be set to between 0 and 1.

[0039] Optionally, the initial trust level of the trust evolution module can be preset according to the traffic scenario type. For example, in areas with a high concentration of vulnerable groups, such as around schools, the second trust level can be initialized to a higher value to reflect the attitude of yielding to pedestrians.

[0040] The trust evolution module uses Bayesian update rules to update the first trust level and the second trust level respectively.

[0041] In some implementations, the Bayesian update rule can take the following form: P(H|E)=P(E|H)×P(H) / P(E), where P(H) represents the prior probability, i.e. the current level of trust, P(H|E) represents the posterior probability, i.e. the updated level of trust, and P(E|H) represents the likelihood function, i.e. the behavioral observation evidence of external traffic participants.

[0042] Here, P(E) represents the marginal likelihood, which is the total probability of observing behavioral evidence E. In the actual calculation of trust updates, P(E), as a normalization constant, can be obtained by integrating the product of the likelihood function and the prior distribution over the total probability space, i.e.: P(E) = ∫P(E|H)·P(H)dH; In the discretized implementation (where the trust level is divided into a finite number of discrete values), P(E) can be expressed as a weighted sum under all possible trust level assumptions: ; in, This represents the i-th possible value of the trust level. This represents the prior probability of that value. Let E be the likelihood probability of observing behavioral evidence E at this level of trust. Since P(E) only serves as a normalization function and does not affect the relative magnitude of the posterior confidence, in engineering implementation, the unnormalized posterior probability P(H|E)∝P(E|H)×P(H) can be directly calculated, and then a normalization factor can be used to ensure that the sum of the probabilities of all hypotheses is 1.

[0043] Optionally, to reduce computational complexity, the system can directly update the confidence level using point estimation methods (such as maximum a posteriori estimation) without explicitly calculating P(E). Regardless of the calculation method used, the engineering essence of P(E) is a scaling factor that normalizes the posterior probability. Its specific value does not affect the trend of the confidence level change, but is only used to ensure that the updated confidence level is within the interval [0, 1].

[0044] Optionally, the parameterization of the likelihood function can be designed specifically for the type of behavioral evidence. For example, for evidence of response consistency, the likelihood function can adopt a binary Bernoulli distribution, while for evidence of decision time, the likelihood function can adopt an exponential or Gaussian distribution.

[0045] The initial trust level is updated based on the consistency of external traffic participants' responses to the vehicle's light signals. Optionally, response consistency can be defined as the degree to which the subsequent behavior of external traffic participants after receiving the vehicle's light signal is consistent with the intent conveyed by the light signal.

[0046] For example, when a vehicle's headlights project a deterministic guide light signal indicating its intention to yield, if the pedestrian immediately begins to cross at a constant speed, it can be considered a highly consistent response. In this case, the likelihood function can assign a higher conditional probability value, thereby positively updating the first confidence level. Conversely, if the pedestrian continues to hesitate or attempts to speed up after receiving the yield signal, it can be considered an inconsistent response, and the first confidence level can be lowered accordingly.

[0047] Optionally, the first level of trust can be updated using incremental Bayesian updates. Each time new response evidence is observed, an incremental update can be performed based on the current level of trust, without the need to store all historical data.

[0048] The second level of trust is updated based on the change in decision time for external traffic participants after the headlight signal is projected. Optionally, decision time can be defined as the time interval between the moment the headlight signal is projected and the moment the external traffic participant makes a decision to proceed.

[0049] In some implementations, if external traffic participants can quickly make a clear decision to proceed after receiving the vehicle light signal, such as starting to pass or stopping within a short time (e.g., within 2 seconds), then the participant can be considered to have a high level of acceptance of the interaction process, and the second level of trust can be positively updated. Conversely, if a participant fails to make a clear decision for a long time (e.g., more than 5 seconds), it may indicate that they have doubts about the reliability of the vehicle light signal, and the second level of trust can be lowered accordingly. Optionally, the normalization of decision time variation can be performed using a relative rate of change, that is, comparing the deviation of the current interaction's decision time with the historical average decision time to eliminate inherent differences in reaction speed among different traffic participants.

[0050] The trust evolution module also includes a trust trajectory recording submodule, configured to continuously record the change trajectory of the first and second trust levels within a preset time window. Optionally, the preset time window length can be set according to the typical interaction duration of the application scenario, for example, it can be set to between 30 and 120 seconds. When multiple interaction records of the same external traffic participant are detected within the historical time window, the trust trajectory recording submodule can input the trust trajectory of the historical interaction records as prior information for the current interaction into the trust evolution module to improve the accuracy of intent ambiguity recognition and game state determination. Optionally, the injection of prior information can be achieved by using the statistical characteristics of historical trust trajectories (such as mean, trend slope, etc.) as prior distribution parameters for Bayesian updates, thereby realizing knowledge transfer and continuous learning across interaction sessions.

[0051] Example 5, based on the foregoing examples, provides a detailed explanation of the specific working mechanism of the strategy decision-making module. The strategy decision-making module selects the corresponding optical language interaction strategy based on a comprehensive evaluation of the intent ambiguity quantification result and the first and second levels of trust.

[0052] Optionally, the comprehensive evaluation can employ a multi-dimensional scoring mechanism. This involves assigning different weight coefficients to the quantified results of intent ambiguity and the two-way trust level, then summing them using weighted averages to obtain a comprehensive game situation score. The corresponding strategy branch is then selected based on the score's interval division. Optionally, the allocation of weight coefficients can reflect the relative importance of each factor in the decision-making process. For example, in scenarios with high security requirements, the weight coefficient of the second level of trust can be appropriately increased to prioritize ensuring the vehicle's passage safety.

[0053] The game states identified by the strategy decision-making module include at least cooperative equilibrium, unilateral strategic shift, and two-way game deadlock.

[0054] Optionally, a cooperative equilibrium state can be defined as a game situation where both parties exhibit stable cooperative intentions, the fuzziness quantification result of intentions is low, and both sides have high levels of trust. A unilateral strategic shift state can be defined as one party's behavior exhibiting obvious strategic characteristics, such as an external traffic participant having a high fuzziness quantification result of intentions but still high first trust, possibly indicating an attempt to gain a passage advantage through ambiguous intentions. A two-way game stalemate state can be defined as a state where both parties are in a state of high intention fuzziness and low trust, such as a high fuzziness quantification result of intentions and a second trust below a preset lower limit, indicating that both parties have serious doubts about the reliability of each other's intentions. Optionally, the identification of different game states can be based on state machine switching according to preset decision boundaries, or it can be automatically identified using a machine learning-based classifier.

[0055] The light-text interaction strategy is selected based on the identified game state. The light-text interaction strategy includes at least a cooperation maintenance strategy, an information verification strategy, and a game de-concealment strategy. Optionally, the cooperation maintenance strategy corresponds to a state with low intention ambiguity quantification results and both bidirectional trust levels exceeding a preset upper limit. In this state, the strategy decision module can control the light-text execution unit to project deterministic guiding light-text to clearly convey the vehicle's intention to yield or proceed, strengthening the continuity of the cooperative equilibrium. The information verification strategy corresponds to a state with medium intention ambiguity quantification results. In this state, the strategy decision module can control the light-text execution unit to project confirmatory probing light-text, testing the true intentions of external traffic participants through changes in the light-text, while simultaneously collecting more behavioral observation data for subsequent trust level updates. The game de-concealment strategy corresponds to a state with high intention ambiguity quantification results and is determined to be a game stalemate. In this state, the strategy decision module can control the light-text execution unit to project light-text containing game information de-concealment semantics with differentiated light effects, breaking the game stalemate by enhancing information transparency.

[0056] Example 6, based on the aforementioned examples, provides a detailed description of the hierarchical response mechanism of the game-playing de-concealment strategy. The game-playing de-concealment strategy further includes a hierarchical response mechanism. Optionally, this hierarchical response mechanism can set multiple response levels according to the increasing degree of intent ambiguity quantification results. Each level corresponds to different light effect parameter configurations to achieve a progressive response to different levels of game tension. First-level game-playing de-concealment: triggered when the intent ambiguity quantification result is greater than a first threshold and less than a second threshold. Optionally, the specific values ​​of the first and second thresholds can be set based on experimental calibration or simulation test results. For example, the first threshold can be set to 0.6, and the second threshold can be set to 0.8. When the first-level game-playing de-concealment is triggered, the strategy decision module can instruct the light signal execution unit to project a game state perception warning pattern at a first brightness multiple and a first flashing frequency. Optionally, the first brightness multiple can be set to 1.2 to 1.5 times the base brightness, and the first flashing frequency can be set to 2 to 4 Hz to convey warning information without causing visual interference.

[0057] The second level of game-theoretic de-concealment is triggered when the quantification result of the intent ambiguity is greater than or equal to a second threshold. Upon triggering the second level of game-theoretic de-concealment, the strategy decision-making module can instruct the optical signal execution unit to project a right-of-way information transparency pattern at a second brightness multiple and a second flashing frequency. Optionally, the second brightness multiple can be set to 1.8 to 2.5 times the base brightness, and the second flashing frequency can be set to 5 to 8 Hz, where the second brightness multiple is greater than the first brightness multiple, and the second flashing frequency is greater than the first flashing frequency. The right-of-way information transparency pattern can include explicit traffic direction indications, an estimated yield time countdown, and other semantic information to forcibly increase the transparency of the interaction information, forcing the strategic actor to abandon its ambiguous intent.

[0058] The effect of the tiered response mechanism is that it gradually adjusts the intensity of light effects and information transparency according to the tension of the game situation. This can avoid the breakdown of interaction caused by prematurely exposing a tough stance, and gradually increase information pressure as the game deadlock continues to deepen, thereby providing clear behavioral guidance signals for external traffic participants.

[0059] Alternatively, in some alternative implementations, the hierarchical response mechanism can also be configured with other parameter dimensions, such as the size of the projected pattern, color saturation, and flashing mode (e.g., the difference between continuous flashing and intermittent flashing), to provide richer means of expressing the game state.

[0060] Example 7, based on the aforementioned examples, provides a detailed explanation of the specific working mechanism of the safety fallback strategy. The light-language interaction strategy also includes a safety fallback strategy, which serves as the final guarantee mechanism for the game-theoretic interaction process, used to address extreme situations where consensus cannot be reached. When the quantification result of intent ambiguity exceeds the third threshold and the second trust level is less than the preset trust lower limit, the strategy decision module can determine that consensus cannot be reached. Optionally, the third threshold can be set with reference to the second threshold of the game-theoretic de-concealment strategy, for example, it can be set to 0.85 to indicate a high deviation from the cooperative traffic pattern; the preset trust lower limit can be set to 0.3 to indicate that the trust level of external traffic participants in this vehicle has decreased to a dangerous level.

[0061] Upon determining that an interactive consensus cannot be reached, the strategy decision-making module can skip the game-theoretic interaction steps to avoid time delays and risk accumulation caused by further attempts at persuasion or probing. Optionally, skipping the game-theoretic interaction steps means that the strategy decision-making module no longer waits for the behavioral responses of external traffic participants, but directly enters the safety response mode. The strategy decision-making module can directly instruct the vehicle to execute a yielding braking strategy. Optionally, the yielding braking strategy can include specific actions such as decelerating to a stop or maintaining a low-speed following position, depending on the vehicle's current driving state and relative distance to external traffic participants. Simultaneously, the strategy decision-making module can control the light signal execution unit to project a red forced stop light pattern. Optionally, the red forced stop light pattern can use a high-brightness static red projection to clearly convey the vehicle's intention to stop, ensuring driving safety.

[0062] The effectiveness of this safety fallback strategy lies in preventing potential accidents by applying forced braking and clear stop signals when the system detects that the risk of an interaction has reached a critical level. Optionally, this strategy can also be linked with the vehicle's underlying safety control system, such as projecting a stop light pattern in advance before the Automatic Emergency Braking (AEB) system is activated, providing sufficient reaction time for surrounding road users.

[0063] Example 8, based on the foregoing examples, provides a detailed description of the specific working mechanism of the optical language execution unit. The optical language execution unit controls the pixelated vehicle lights to execute the light projection pattern corresponding to the selected optical language interaction strategy. Optionally, the pixelated vehicle lights can employ technologies such as digital micromirror devices (DMD), liquid crystal on silicon (LCOS), or Micro LED arrays to achieve high-resolution light projection. The number of pixels can be set between several thousand and hundreds of thousands to support clear projection of complex patterns. The optical language execution unit can select the corresponding light projection pattern from a predefined pattern library for projection based on the strategy instructions output by the strategy decision module.

[0064] The light projection patterns include at least deterministic guiding light signals for maintaining cooperative equilibrium, confirmatory probing light signals for responding to strategic shifts, and game information decryption light signals for breaking game deadlocks. Optionally, deterministic guiding light signals can adopt simple and clear pattern designs, such as a combination of a green checkmark and a forward arrow, indicating that the vehicle confirms its willingness to yield and guide the other vehicle through; confirmatory probing light signals can adopt dynamically changing pattern designs, such as a combination of a question mark and a head-shaking animation, indicating that the vehicle has doubts about the other vehicle's intentions and requests a clear response; game information decryption light signals can adopt high-contrast warning pattern designs, such as a combination of a red exclamation mark and a flashing border, indicating that the vehicle has identified a game deadlock and requests a clear statement from the other vehicle.

[0065] The optical signal execution unit and the strategy decision module are connected via a strategy signal interface. The strategy signal may include pattern type encoding, light effect parameter set, and projection timing information. Optionally, after being triggered, the strategy signal can be transmitted to the optical signal execution unit via an in-vehicle communication network such as CAN bus or vehicle Ethernet. After receiving the strategy signal, the optical signal execution unit can complete the pattern switching action within a preset response delay time (e.g., within 100 milliseconds).

[0066] Example 9, based on the aforementioned examples, provides a comprehensive description of the collaborative working mechanism between the various modules of the system. The data flow and signal transmission between the system modules work collaboratively according to a predetermined temporal logic. First, the perception unit continuously collects perception data of the vehicle's surrounding environment and performs target detection and behavioral feature extraction in real time. The generated behavioral feature data can be updated at a preset frequency (e.g., 10 Hz) and output to the intent evaluation module. After receiving the behavioral feature data, the intent evaluation module can calculate the intent ambiguity quantification result in real time and output the result to the trust evolution module and the strategy decision module.

[0067] After receiving the fuzzy quantification result of intent, the trust evolution module can update the bidirectional trust level by combining the current optical language projection state and historical interaction records with Bayesian update rules.

[0068] Optionally, the trust level update can be performed immediately upon receiving new behavioral observation evidence, or it can be cumulatively updated in batches over a preset time interval (e.g., 1 second). The updated first and second trust levels can be used as input parameters for the strategy decision-making module, participating in game state identification and strategy selection together with the intent fuzziness quantification results.

[0069] After receiving the fuzzy quantification result of intent and the two-way trust level, the strategy decision module can determine the light language interaction strategy to be adopted based on the predefined strategy selection logic.

[0070] Optionally, the strategy decision-making module can implement the state transition logic using a finite state machine, automatically switching to the corresponding strategy branch based on the current game situation. After determining the strategy, the strategy decision-making module can generate the corresponding strategy signal and send it to the optical language execution unit to trigger the optical language projection action.

[0071] After receiving the strategy signal, the light language execution unit can control the pixelated headlights to execute the corresponding light projection pattern.

[0072] Optionally, the optical signal execution unit can also receive real-time feedback signals from the sensing unit to determine whether the currently projected pattern has been observed by external traffic participants, thus providing a basis for subsequent trust level updates. The entire system's workflow can be continuously executed in a loop until the external traffic participants complete their passage or the system determines that the interaction needs to be terminated.

[0073] The overall technical effects resulting from the collaborative work among the modules include: through multi-level processing by the perception unit, intent assessment module, trust evolution module, and strategy decision-making module, accurate identification of the intentions of external traffic participants and comprehensive assessment of the game situation are achieved; through multi-level light effect projection by the light language execution unit, progressive responses to different game states are achieved, effectively addressing strategic deviations and game stalemates while maintaining cooperative equilibrium; and through the setting of a safety fallback strategy, priority is given to ensuring driving safety in extreme situations.

[0074] Example 10: Based on the aforementioned examples, this example provides an illustrative description of the specific application process of the system in a nighttime zebra crossing scenario without traffic lights. When the vehicle travels to a nighttime zebra crossing area without traffic lights, the sensing unit can begin collecting target data for the area ahead.

[0075] Optionally, the sensing unit can first use lidar to scan and determine the passable area in front of the zebra crossing, and detect whether pedestrians or other traffic participants have entered the area. Once an external traffic participant is detected, the sensing unit can extract the participant's behavioral characteristic data, including position oscillation patterns, velocity oscillation patterns, and the angle between head orientation and direction of movement.

[0076] The intent assessment module can calculate the intent ambiguity quantification result based on behavioral feature data. In an exemplary scenario, if a pedestrian exhibits obvious hesitation characteristics at a zebra crossing, such as repeatedly moving forward and backward (higher score for positional oscillation pattern), fluctuating speed (higher score for speed oscillation pattern), and frequently switching gaze (higher score for gaze and orientation deviation), the intent ambiguity quantification result may be calculated as 0.75, which belongs to a moderately high level of ambiguity.

[0077] The trust evolution module can update the two-way trust level by combining historical interaction data. For example, if a pedestrian has previously accelerated and attempted to cross the road after responding to a vehicle's yield signal in the current zebra crossing area's historical interactions, the first trust level might be updated to around 0.5, lower than the preset upper threshold. If the pedestrian demonstrates sustained attention to the vehicle's projected light signal and relatively stable decision-making time during this interaction, the second trust level might be updated to around 0.6.

[0078] The strategy decision-making module comprehensively evaluates the intent ambiguity quantification result (0.75) and the two-way trust level (0.5 and 0.6), determining the current game state to be a moderate strategic shift and triggering an information verification strategy. The optical language execution unit can project confirmatory probe optical language, such as a pattern with a dynamic question mark effect, and observe the pedestrian's subsequent reaction. If the pedestrian begins to cross at a constant speed after receiving the confirmatory probe optical language, it can be considered that the intent ambiguity has decreased, and the first and second trust levels can be adjusted accordingly, allowing the strategy decision-making module to switch to a cooperative maintenance strategy. If the pedestrian continues to hesitate or attempts to rush across, the intent ambiguity quantification result may further increase, and the system can gradually upgrade to a game de-concealment strategy.

[0079] Example 11, based on the aforementioned examples, extends the application of the system in other typical traffic scenarios. Besides nighttime crosswalks without traffic lights, this system can also be applied to other scenarios requiring implicit negotiation between vehicles and traffic participants, such as nighttime traffic within residential areas, traffic around schools, and oncoming traffic at intersections without traffic lights. In different application scenarios, the specific parameter configurations of each module can be adjusted. For example, in high-speed highway merging scenarios, the target detection range of the sensing unit can be expanded accordingly, and the time window for determining the game state can be shortened accordingly; in areas with a high concentration of vulnerable groups (such as around schools and kindergartens), the initial value and lower limit of the second level of trust can be appropriately increased to reflect a higher degree of courtesy towards pedestrians.

[0080] In some implementations, the system can also be integrated with other in-vehicle systems, such as high-precision map systems to obtain road type and zebra crossing location information, V2X communication systems to obtain information on traffic participants at greater distances, and automatic emergency braking (AEB) systems to achieve linkage between safety fallback strategies and underlying safety controls. Optionally, these integration functions can be implemented through standardized interface definitions, allowing the system to flexibly choose to enable or disable them based on actual configuration.

[0081] Example 12, based on the previous examples, describes the parameter configuration and optimization methods for the system. The system involves multiple configurable parameters, including but not limited to weight coefficients in the intent fuzzy quantization model, threshold parameters for triggering each strategy, brightness multiples of the light effect projection, and flicker frequency parameters. The values ​​of these parameters can significantly affect the overall performance of the system.

[0082] In some implementations, the initial values ​​of the parameters can be preset based on theoretical derivation or simulation test results, and then optimized using real vehicle test data.

[0083] Optionally, parameter tuning can employ a combination of offline optimization and online learning. The offline optimization phase can utilize large-scale historical interaction data for batch parameter optimization, such as using Bayesian optimization or reinforcement learning algorithms to search for the optimal parameter combination. The online learning phase can leverage real-time interaction data for incremental parameter updates, such as automatically adjusting the learning rate based on the interaction success rate. Optionally, to ensure system stability and interpretability, an upper limit can be set on the parameter update magnitude during online learning to prevent behavioral instability caused by drastic parameter fluctuations.

[0084] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent vehicle lighting control system with multi-modal interactive feedback function, characterized in that, include: The sensing unit is configured to acquire sensing data of the environment surrounding the vehicle, for identifying external traffic participants and extracting their behavioral characteristic data. The intent assessment module generates an intent fuzziness quantification result based on the behavioral feature data. The intent fuzziness quantification result is used to characterize the degree to which the current behavior of external traffic participants deviates from the cooperative traffic pattern. The cooperative traffic pattern refers to the behavior pattern of external traffic participants in accordance with traffic rules and conventional courtesy habits. The trust evolution module establishes and updates the vehicle's first level of trust in external traffic participants and the external traffic participants' second level of trust in the vehicle based on historical interaction data. The strategy decision-making module selects the corresponding light language interaction strategy based on the comprehensive evaluation of the intent ambiguity quantification result and the first trust level and the second trust level. The light language execution unit controls the pixelated headlights to execute the light projection pattern corresponding to the selected light language interaction strategy. The light projection pattern includes at least a deterministic guiding light language for maintaining cooperative equilibrium, a confirmatory probing light language for responding to strategic deviations, and a game information de-masking light language for breaking the game deadlock.

2. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 1, characterized in that: The sensing unit includes at least one or more of lidar, millimeter-wave radar, and cameras.

3. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 1, characterized in that: The behavioral feature data extracted by the sensing unit includes at least the position oscillation pattern, velocity oscillation pattern, head orientation and movement direction angle of the external traffic participant, and the consistency characteristics of the participant's response to the projected light signal.

4. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 1, characterized in that: The intent assessment module uses a multi-feature weighted quantization model to generate the intent ambiguity quantization result. The multi-feature weighted quantization model uses at least a portion of the normalized scores of position oscillation patterns, velocity oscillation patterns, line-of-sight and orientation deviation, and the normalized scores of consistency with light speech response as input variables.

5. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 1, characterized in that: The trust evolution module uses Bayesian update rules to update the first trust level and the second trust level respectively, wherein: The first level of trust is updated based on the consistency of external traffic participants' responses to the vehicle's light signals; The second level of trust is updated based on the change in decision-making time of external traffic participants after the projection of vehicle headlight signals.

6. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 1, characterized in that: The game states identified by the strategy decision-making module include at least cooperative equilibrium, unilateral strategic shift, and two-way game deadlock. The light-language interaction strategy is selected based on the identified game states.

7. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 1, characterized in that: The light-text interaction strategy includes at least the following: Cooperation maintenance strategy: In the case of a state where the quantification result of low intention ambiguity is high and both the trust level of both sides is higher than the preset upper limit, control the light language execution unit to project deterministic guiding light language; Information verification strategy: corresponding to the state of the quantification result of medium intention ambiguity, control the optical language execution unit to project a verification probe optical language; Game-theoretic de-concealing strategy: For states with high intention ambiguity quantification results and determined to be game stalemates, control the light language execution unit to project light language containing game information de-concealing semantics with differentiated light effects.

8. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 7, characterized in that: The game-theoretic unmasking strategy further includes a tiered response mechanism: Level 1 Game Unmasking: Triggered when the quantification result of intent ambiguity is greater than the first threshold and less than the second threshold, a game state perception warning pattern is projected with a first brightness multiple and a first flashing frequency; Second-level game de-masking: Triggered when the quantification result of intent ambiguity is greater than or equal to the second threshold, the right-of-way information transparency pattern is projected at the second brightness multiple and the second flashing frequency; Wherein, the second brightness multiple is greater than the first brightness multiple, and the second flashing frequency is greater than the first flashing frequency.

9. The intelligent vehicle lighting control system with multi-mode interactive feedback function according to claim 7, characterized in that: The light-text interaction strategy also includes a safety fallback strategy: When the quantification result of intent ambiguity is greater than the third threshold and the second trust level is less than the preset trust lower limit, it is determined that the interactive consensus cannot be reached. The strategy decision module skips the game interaction steps, directly instructs the vehicle to execute the yield braking strategy, and controls the light language execution unit to project a red forced stop light pattern.

10. The intelligent vehicle lighting control system with multi-modal interactive feedback function according to claim 1, characterized in that: The trust evolution module also includes a trust trajectory recording submodule, configured as follows: The trajectory of the change between the first trust level and the second trust level is continuously recorded within a preset time window; When multiple interaction records of the same external traffic participant are detected within a historical time window, the trust trajectory of the historical interaction records is input into the trust evolution module as prior information for the current interaction, in order to improve the accuracy of intent ambiguity recognition and game state determination.

Images