A riding risk identification system based on helmet vehicle end data processing

Abstract

The present application relates to the technical field of riding risk identification, and discloses a riding risk identification system based on helmet vehicle end data processing, comprising: a space-time reference alignment module for obtaining helmet inertial navigation data, vehicle motion data and navigation auxiliary data and completing time alignment and coordinate system establishment; a steady straight alignment module for updating attitude bias and generating corrected head attitude under steady straight conditions; a vehicle state integral module for calculating vehicle heading, vehicle roll and vehicle instantaneous curvature; an intention response modeling module for calculating instantaneous angle difference and preview angle difference; a stability margin evaluation module for generating equivalent balance roll angle, attitude stability margin and preview distance; a risk criterion hysteresis module for outputting risk state and risk cause identification; and an event generation reporting module for generating an event package and performing threshold bounded updating. The present application realizes certain identification of the riding risk state.

Description

Technical Field

[0001] This invention belongs to the field of cycling risk identification technology, specifically relating to a cycling risk identification system based on helmet-mounted data processing. Background Technology

[0002] In recent years, with the widespread adoption of shared electric-assisted bicycles, electric two-wheelers, and urban micro-mobility tools, the number of riders has grown rapidly, and the accompanying safety issues have become increasingly prominent. Current road traffic management focuses more on the operational status of vehicles, while its ability to capture changes in individual rider behavior is limited. When riders make actions such as turning their heads, observing, or avoiding obstacles in complex environments, their head posture often changes before the vehicle's. The vehicle's sensors struggle to reflect the rider's immediate intentions in a timely manner, resulting in a delayed response and posing certain risks in scenarios such as changing lanes at intersections, avoiding narrow roads, and meeting oncoming traffic in mixed-traffic areas.

[0003] Current common risk identification methods mostly rely on single-side vehicle sensor data, such as vehicle speed, acceleration, and yaw rate. These methods can only reflect the passive motion state of the vehicle and cannot accurately infer the rider's active steering intentions. Furthermore, some methods use visual sensors to identify head posture, but visual solutions are easily limited by factors such as lighting, occlusion, and cost, resulting in insufficient versatility for practical deployment. On the other hand, while helmet inertial sensors can acquire rider head posture information, if they are not synchronized with the vehicle-side sensors in terms of time and coordinate domain, effective correlation calculations cannot be performed between the two, making it difficult to establish a unified intention response relationship across different platforms.

[0004] In cycling scenarios, vehicle steering exhibits significant dynamic lag and behavioral variability, making it difficult to cover the ever-changing road conditions by relying solely on instantaneous angles or thresholds. Existing risk identification methods often employ empirical rules or static thresholds when dealing with the deviation between the rider's intention and the vehicle's actual response, lacking a comprehensive analysis of vehicle posture stability, accessibility, and anticipation behavior. Therefore, they often fail to provide stable and reliable risk identification results in complex scenarios. Summary of the Invention

[0005] This invention provides a cycling risk identification system based on helmet and vehicle-side data processing, which solves the technical problem in related technologies that the helmet-side and vehicle-side sensor data cannot be correlated and processed in a unified time and spatial domain, resulting in the inaccurate establishment of the relationship between the rider's intention and the vehicle's response, thus making it difficult to achieve reliable cycling risk identification.

[0006] This invention provides a cycling risk identification system based on helmet-mounted data processing, comprising:

[0007] The spatiotemporal reference alignment module is used to acquire helmet inertial navigation data from the helmet end, vehicle motion data from the vehicle end, and dead reckoning data, and to perform time alignment to establish three coordinate systems: road surface, vehicle body, and helmet.

[0008] The steady-state straight alignment module is used to calculate the heading difference and roll difference during steady-state straight travel, update the heading offset and roll offset, obtain the corrected nose heading and nose roll, and output the steady-state straight marker.

[0009] The vehicle state integration module is used to perform time-domain integration based on vehicle motion data to obtain the updated vehicle heading, updated vehicle roll, and instantaneous vehicle curvature.

[0010] The intent response modeling module is used to determine the vehicle's heading at the aiming moment based on the corrected head heading as the intent agent and the updated vehicle heading as the response agent, according to the aiming time and the vehicle yaw rate. It obtains the aiming angle difference and retains the instantaneous angle difference.

[0011] The stability margin assessment module is used to calculate the equivalent balance roll angle based on vehicle speed and instantaneous curvature of the vehicle body, obtain the attitude stability margin based on the updated vehicle roll, and combine the aiming time and vehicle speed to obtain the aiming distance.

[0012] The risk criterion hysteresis module is used to establish deterministic risk criteria and construct hysteresis control through entry and exit counts, generating risk status and risk cause identifiers;

[0013] The event generation and reporting module is used to generate event packets on the rising edge when the risk status changes from closed to open, and to write logs and report to the interface. When the steady-state straight-line flag is valid and the risk status is closed, the threshold parameter is updated in a bounded manner.

[0014] Furthermore, the helmet inertial navigation data includes the helmet yaw rate; the vehicle-end motion data includes the vehicle yaw rate, vehicle lateral acceleration, and vehicle speed; and the dead reckoning data includes the dead reckoning heading and dead reckoning position.

[0015] Acquire helmet inertial navigation data from the helmet, vehicle motion data from the vehicle, and dead reckoning data, perform time alignment, and establish three coordinate systems: road surface, vehicle body, and helmet.

[0016] Step 11: Construct a unified timeline with equal intervals using the vehicle-end clock, and bind the helmet inertial navigation data, vehicle-end motion data, and dead reckoning data to the unified timeline index by linear interpolation of timestamps;

[0017] Step 12: On the unified time axis, set the linear drift coefficient to zero, enumerate the time deviation between the helmet end and the vehicle end according to the candidate set, and for each time deviation, calculate the absolute difference between the vehicle body yaw rate and the helmet yaw angular velocity point by point on the unified time axis, and sum them up. Take the minimum summation result as the initial value of the time deviation; use this initial value to jointly search for the time deviation and the linear drift coefficient, and determine the correction parameter pair according to the minimum summation result. Obtain the lateral acceleration consistency residual by subtracting the vehicle body lateral acceleration from the product of the vehicle speed and the vehicle body yaw rate, and mark and remove the unified time axis sampling points that exceed the threshold.

[0018] Step 13: Define the road surface coordinate system, vehicle body coordinate system, and helmet coordinate system as right-handed coordinate systems, fix the forward, left, and upward directions and the origin position respectively, and keep the naming, axis and origin unchanged under a unified time axis index.

[0019] Furthermore, during steady-state straight-ahead travel, the heading difference and roll difference are calculated, the heading offset and roll offset are updated, the corrected nose heading and nose roll are obtained, and the steady-state straight-ahead marker is output, including:

[0020] Step 21: On a unified time axis, calculate the product of vehicle speed and vehicle yaw rate to obtain lateral acceleration; within the sliding window, when the vehicle speed threshold, yaw rate threshold, and lateral acceleration threshold are all satisfied, set the steady-state straight-line marker to valid; otherwise, set it to invalid.

[0021] Step 22: On the samples marked as valid in steady-state straight travel, calculate the difference between the heading and the body heading to obtain the heading difference, and perform angle normalization. Calculate the difference between the roll and the roll of the head to obtain the roll difference. Use the heading difference and roll difference as observations of heading offset and roll offset.

[0022] Step 23: Within the sliding window where the steady-state straight-ahead mark is valid, calculate the window average of the heading difference and roll difference respectively. Calculate the weighted sum of the heading offset from the previous moment and the window average heading difference to obtain the updated heading offset. Calculate the weighted sum of the roll offset from the previous moment and the window average roll difference to obtain the updated roll offset. Calculate the difference between the head heading and the heading offset to obtain the corrected head heading. Calculate the difference between the head roll and the roll offset to obtain the corrected head roll. Output the steady-state straight-ahead mark.

[0023] Furthermore, based on the vehicle motion data, time-domain integration is performed to obtain the updated vehicle heading, updated vehicle roll, and instantaneous vehicle curvature, including:

[0024] Step 31: On a unified time axis, exclude sampling points whose index markers have been removed, use the time interval between adjacent sampling points as the time step, and use the vehicle heading of the first valid sample as the initial value.

[0025] Step 32: At each time step, calculate the product of the vehicle yaw rate and the time step, and add the updated vehicle heading from the previous time step to obtain the updated vehicle heading for the next time step, and then normalize the angle.

[0026] Step 33: At each time step, calculate the ratio of vehicle body yaw rate to vehicle speed to obtain the instantaneous curvature of the vehicle body. When the vehicle speed is lower than the preset lower limit, the preset lower limit is used to replace the vehicle speed. The difference between the product of vehicle speed and vehicle body yaw rate and the vehicle body lateral acceleration is used as the lateral consistency difference. When the difference exceeds the threshold, it is marked as a suspicious sampling point on the unified time axis.

[0027] Step 34: At each time step, calculate the product of the square of the vehicle speed and the instantaneous curvature of the vehicle body, divide it by the gravitational acceleration, take the arctangent to obtain the principal value of the roll angle, determine the roll direction according to the sign of the instantaneous curvature of the vehicle body, and obtain the updated vehicle roll.

[0028] Furthermore, the vehicle's heading at the moment of aiming is determined, the aiming angle difference is obtained, and the instantaneous angle difference is retained, including:

[0029] Step 41: On the unified timeline, record the corrected head heading as the intent proxy field and the updated vehicle heading as the response proxy field, and establish the correspondence between the field names and time indexes of the two in the unified timeline index.

[0030] Step 42: At each time point, calculate the difference between the corrected head heading and the updated vehicle heading to obtain the instantaneous angle difference. Add the product of the aiming time and the vehicle yaw rate to the updated vehicle heading to obtain the vehicle heading at the aiming time, and perform angle normalization processing respectively.

[0031] Step 43: At each time point, calculate the difference between the corrected head heading and the vehicle heading at the aiming time to obtain the aiming angle difference, and use the product of the vehicle speed and the aiming time as the aiming distance, and output the instantaneous angle difference, aiming angle difference, aiming distance and vehicle heading at the aiming time.

[0032] Furthermore, the equivalent balance roll angle is calculated based on the vehicle speed and instantaneous curvature of the vehicle body. The attitude stability margin is obtained based on the updated vehicle roll. The aiming time and vehicle speed are combined to obtain the aiming distance, including:

[0033] Step 51: At each time point, multiply the square of the vehicle speed by the absolute value of the instantaneous curvature of the vehicle body, divide by the gravitational acceleration as the denominator, and take the arctangent as the principal value of the equivalent equilibrium roll angle.

[0034] Step 52: At each time point, the attitude stability margin is obtained by subtracting the absolute value of the updated vehicle roll from the maximum allowable roll angle.

[0035] Step 53: At each time point, the product of the aiming time and the vehicle speed is used as the aiming distance, and the ratio of the aiming angle difference to the aiming distance is used as the reachable curvature. The equivalent balance roll angle, attitude stability margin, aiming distance and reachable curvature are output.

[0036] Furthermore, deterministic risk criteria are established, and hysteresis control is constructed through entry and exit counts to generate risk status and risk cause identifiers, including:

[0037] Step 61: Obtain the aiming angle difference, instantaneous curvature of the vehicle body, vehicle speed, reachable curvature and attitude stability margin on a unified time axis, and simultaneously read the safety angle difference threshold, minimum effective curvature threshold, stability margin safety reduction amount and gravitational acceleration constant; if any input is missing, pause the risk assessment.

[0038] Step 62: Establish a deterministic risk criterion that includes the risk of unresponsive intent, the risk of reverse response, and the risk of unreachability. If any of these risks is met, the risk trigger signal is set to valid.

[0039] Step 63: When the risk trigger signal is valid within a preset time, the counter increments; when the risk trigger signal remains invalid, the counter increment stops; when the cumulative counter value reaches the corresponding threshold, the risk state is triggered to open or close.

[0040] Step 64: When the risk status changes from closed to open, determine the risk cause identifier based on the source of the trigger signal, and classify the risk cause identifier into three categories according to the preset priority: no response to intent, reverse response, and unreachable risk.

[0041] Furthermore, the deterministic risk criterion specifically includes:

[0042] When the absolute value of the pre-aiming angle difference exceeds the safe angle difference threshold and the instantaneous curvature of the vehicle body is lower than the minimum effective curvature threshold, it is determined as a risk of unresponsive intent.

[0043] When the pre-aiming angle difference is opposite to the direction of the instantaneous curvature of the vehicle body and the absolute value of the pre-aiming angle difference exceeds the safety angle difference threshold, it is identified as a reverse response risk.

[0044] The difference between the attitude stability margin and the stability margin safety reduction is calculated. When the equivalent balance roll angle calculated based on the vehicle speed and reachability curvature is greater than this difference, it is determined to be an unreachable risk.

[0045] Furthermore, an event packet is generated on the rising edge when the risk state changes from closed to open. The event packet is written to a log file and reported through an interface. When the risk state stabilizes at closed and the steady-state straight-line marker is valid, a bounded adaptive update is performed on the safety angle difference threshold and the minimum effective curvature threshold. Within the sliding time window, the median of the absolute values ​​of the instantaneous angle difference and the median of the absolute values ​​of the instantaneous curvature of the vehicle body are taken respectively, and the current threshold is weighted and summed in a fixed ratio to form a candidate threshold. The candidate threshold is then limited to a preset allowable range through interval projection. The event packet includes the current time index, vehicle speed, instantaneous angle difference, pre-aiming angle difference, instantaneous curvature of the vehicle body, reachable curvature, attitude stability margin, and risk cause identifier.

[0046] Furthermore, when abnormal input data is detected or the proportion of invalid samples within the window exceeds a preset proportion threshold, the threshold parameter update is immediately frozen and the freeze event is recorded. When no abnormal data is detected within multiple consecutive sampling periods and the proportion of invalid samples within the window is lower than the preset proportion threshold, the freeze state is lifted. Here, the invalid sampling proportion is the proportion of sampling points whose index markers have been removed to the sampling points in the window.

[0047] The beneficial effects of this invention are as follows: By unifying the time alignment, defining the coordinate system, and correcting the posture bias of multi-source sensor data from the helmet and vehicle ends, this invention establishes a correspondence between the rider's intention and the vehicle's response within the same reference domain, solving the difficulty of directly linking head posture and vehicle dynamics in existing technologies. By integrating vehicle state in the time domain, continuous vehicle heading, roll, and instantaneous curvature are obtained, making vehicle posture quantities traceable and temporally consistent, providing a reliable foundation for intention response modeling and stability analysis. By combining the corrected head heading, aiming time, vehicle yaw rate, and reachable curvature to construct aiming angle difference and aiming distance, a unified expression of intention quantities and vehicle trajectory requirements is achieved. Deterministic risk criteria are formed through posture stability margin, equivalent balance roll angle, and multiple risk triggering conditions, and a hysteresis mechanism is established using entry and exit counting, ensuring stable output characteristics for risk states. Through event packet reporting and bounded adaptive updates of thresholds, the system maintains a consistent data structure and parameter benchmark during long-term operation. Overall, this invention can complete the entire process of cycling risk identification under a unified data link. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of a cycling risk identification system based on helmet-mounted data processing according to the present invention. Detailed Implementation

[0049] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0050] like Figure 1 As shown, a cycling risk identification system based on helmet-mounted data processing includes:

[0051] The spatiotemporal reference alignment module 1 is used to acquire helmet inertial navigation data from the helmet end, vehicle motion data from the vehicle end, and dead reckoning data, and to perform time alignment to establish three coordinate systems: road surface, vehicle body, and helmet.

[0052] The steady-state straight alignment module 2 is used to calculate the heading difference and roll difference during steady-state straight travel, update the heading offset and roll offset, obtain the corrected nose heading and nose roll, and output the steady-state straight mark.

[0053] Vehicle state integration module 3 is used to perform time-domain integration based on vehicle motion data to obtain updated vehicle heading, updated vehicle roll and instantaneous vehicle curvature.

[0054] The intent response modeling module 4 is used to determine the vehicle heading at the aiming moment based on the aiming time and the vehicle yaw rate, using the corrected head heading as the intent agent and the updated vehicle heading as the response agent, to obtain the aiming angle difference and retain the instantaneous angle difference.

[0055] Stability margin assessment module 5 is used to calculate the equivalent balance roll angle based on vehicle speed and instantaneous curvature of the vehicle body, obtain the attitude stability margin based on the updated vehicle body roll, and combine the aiming time and vehicle speed to obtain the aiming distance.

[0056] Risk criterion hysteresis module 6 is used to establish deterministic risk criteria and construct hysteresis control through entry count and exit count, generating risk status and risk cause identifiers;

[0057] The event generation and reporting module 7 is used to generate event packets on the rising edge when the risk state changes from closed to open, and to write logs and report to the interface. When the steady-state straight-line flag is valid and the risk state is closed, the threshold parameter is updated in a bounded manner.

[0058] In one embodiment of the present invention, the helmet inertial navigation data includes helmet yaw rate, which is used to characterize the angular velocity of the rider's head about the vertical axis; the vehicle motion data includes vehicle yaw rate, vehicle lateral acceleration and vehicle speed, which are used to describe the vehicle's own steering motion; and the dead reckoning data includes dead reckoning heading and dead reckoning position, which are used to help correct the cumulative error of the vehicle inertial navigation system.

[0059] Acquire helmet inertial navigation data from the helmet, vehicle motion data from the vehicle, and dead reckoning data, perform time alignment, and establish three coordinate systems: road surface, vehicle body, and helmet.

[0060] Step 11: Construct a unified time axis with equal intervals using the vehicle-end clock to eliminate the differences in sampling frequency and time drift between different data sources. Bind the helmet inertial navigation data, vehicle-end motion data and dead reckoning data to the unified time axis index by linear interpolation of timestamps to achieve a one-to-one correspondence between the data under a unified time reference.

[0061] Step 12: On the unified time axis, set the linear drift coefficient to zero, enumerate the time deviation between the helmet end and the vehicle end according to the candidate set. The candidate set is a set of discrete hypothetical values ​​used to enumerate and search for the initial value of the time deviation. For each time deviation, calculate the absolute difference between the vehicle body yaw rate and the helmet yaw angular velocity point by point on the unified time axis, and sum them up. Take the minimum summation result as the initial value of the time deviation. Use this initial value to jointly search for the time deviation and the linear drift coefficient, and determine the correction parameter pair according to the minimum summation result. The correction parameter pair includes the time deviation and the linear drift coefficient. The linear drift coefficient represents the proportional deviation between the clock frequencies of the helmet and the vehicle end. Obtain the lateral acceleration consistency residual by calculating the product of the vehicle speed and the vehicle body yaw rate and subtracting the vehicle body lateral acceleration. Mark and remove unified time axis sampling points that exceed the threshold to eliminate abnormal data interference.

[0062] Step 13: Define the road surface coordinate system, vehicle body coordinate system, and helmet coordinate system as right-handed coordinate systems, fix the forward, left, and upward directions and the origin position respectively, and keep the naming, axis and origin unchanged under a unified time axis index.

[0063] Through the above steps, this invention achieves unified temporal and spatial correction of multi-source sensor data from the helmet and vehicle ends, and establishes a data benchmark for co-domain alignment. In the temporal domain, it achieves synchronization of multiple sensors through joint correction of time deviation and drift compensation. In the spatial domain, it achieves geometric correspondence between gaze intention and vehicle response through the unified definition of coordinate system. Residual-driven adaptive calibration improves the spatiotemporal alignment accuracy and robustness.

[0064] In one embodiment of the present invention, during steady-state straight-ahead travel, the heading difference and roll difference are calculated, the heading offset and roll offset are updated, the corrected nose heading and nose roll are obtained, and a steady-state straight-ahead marker is output, including:

[0065] Step 21: On a unified time axis, calculate the product of vehicle speed and vehicle yaw rate to obtain lateral acceleration, which reflects the centrifugal component generated by the vehicle's steering. Within the sliding window, when the vehicle speed threshold, yaw rate threshold, and lateral acceleration threshold are all satisfied simultaneously, the steady-state straight-line marker is set to valid; otherwise, it is set to invalid. Specifically, when the vehicle speed is higher than the set vehicle speed threshold, the absolute value of the vehicle yaw rate is less than the yaw rate threshold, and the absolute value of the lateral acceleration is less than the lateral acceleration threshold, the steady-state straight-line marker is set to valid. A valid steady-state straight-line marker indicates that the vehicle's current motion state is in a stable straight-line range. When any condition is not satisfied, the steady-state straight-line marker is set to invalid.

[0066] Step 22: On samples marked as valid for steady-state straight travel, calculate the difference between the head heading and the vehicle body heading to obtain the heading difference, and perform angle normalization to limit the heading difference to within a certain range. Within the interval, to avoid jumps caused by crossing full circumference angles, the difference between head roll and vehicle roll is calculated to obtain the roll difference; the heading difference and roll difference are used as observations of heading offset and roll offset, respectively, to characterize the offset between the rider's head posture and the vehicle's own posture.

[0067] Step 23: Within the sliding window where the steady-state straight-ahead marker is valid, calculate the window average of the heading difference and roll difference respectively. Then, weighted sum the heading offset from the previous moment with the window average heading difference to obtain the updated heading offset. Similarly, weighted sum the roll offset from the previous moment with the window average roll difference to obtain the updated roll offset. Calculate the difference between the head heading and the heading offset to obtain the corrected head heading, and calculate the difference between the head roll and the roll offset to obtain the corrected head roll. Output the steady-state straight-ahead marker. This process eliminates the long-term head attitude deviation over time through gradual correction, ensuring that the corrected head heading and head roll directly correspond to the vehicle attitude.

[0068] Through the above steps, this embodiment constructs a head posture alignment mechanism based on steady-state straight-line driving conditions. In the time domain, by jointly determining vehicle speed, vehicle yaw rate, and lateral acceleration, the bias update is limited to steady-state driving conditions. In the attitude domain, through windowed estimation of heading difference and roll difference and progressive weighted updates, continuous alignment between the helmet end and vehicle end attitudes is achieved. By generating the corrected head heading and head roll, an accurate, consistent, and traceable attitude benchmark is provided for subsequent intent response modeling.

[0069] In one embodiment of the present invention, time-domain integration is performed based on vehicle motion data to obtain updated vehicle heading, vehicle roll, and instantaneous vehicle curvature, including:

[0070] Step 31: On a unified time axis, examine each time point in the sampling sequence, excluding sampling points marked as removal indices, and retaining only unmarked valid sampling points as integration input. Calculate the time interval between adjacent valid sampling points and use this interval as the time step for subsequent progressive integration. Based on this, select the vehicle heading of the first valid sampling point in the sequence as the initial heading value for integration, to begin constructing the updated vehicle heading sequence.

[0071] Step 32: In each time step, read the vehicle yaw rate at that time point and calculate the product of the vehicle yaw rate and the time step size, which represents the heading increment within the current time step. Then, add this heading increment to the updated vehicle heading from the previous time step to obtain the updated vehicle heading for the next time step. To avoid jumps caused by angles spanning entire revolutions, the obtained heading value is normalized to keep it within a defined range of principal angle values.

[0072] Step 33: At each time step, the instantaneous curvature of the vehicle body is obtained by calculating the ratio between the vehicle body yaw rate and the vehicle speed. When the vehicle speed is lower than a preset lower speed limit, this lower speed limit is used instead of the original vehicle speed for calculation to avoid unstable curvature results due to excessively low speeds. Simultaneously, the lateral consistency difference at that time point is obtained by calculating the product of the vehicle speed and the vehicle body yaw rate and subtracting the vehicle body lateral acceleration. This value reflects the consistency between the vehicle's lateral dynamics. When the lateral consistency difference exceeds a preset threshold, the current time point is marked as a suspicious sampling point on the unified time axis.

[0073] Step 34: Within each time step, calculate the square of the vehicle speed and multiply this squared value by the absolute value of the instantaneous curvature of the vehicle body to obtain the equivalent lateral acceleration component of the vehicle under the current curvature and speed conditions. Divide this result using gravitational acceleration as the divisor and convert it into the principal value of the roll angle using the arctangent method, representing the theoretical equilibrium roll angle of the vehicle under the current conditions. Determine the roll direction based on the sign of the instantaneous curvature of the vehicle body, assign the corresponding positive or negative sign to the principal value of the roll angle, and obtain the updated vehicle roll angle.

[0074] Through the above steps, this embodiment constructs a time-domain integration process for vehicle attitude under a unified time axis constraint. Based on sample selection, the updated vehicle heading is progressively advanced, the instantaneous curvature of the vehicle is calculated, and the updated vehicle roll is derived, ensuring that the vehicle's heading, curvature, and roll have a consistent definition system in the time domain. This process provides continuous, unified, and traceable vehicle attitude quantities for intent response modeling and stability margin assessment.

[0075] In one embodiment of the present invention, the corrected head heading is used as the intent agent, and the updated vehicle heading is used as the response agent. The vehicle heading at the aiming moment is determined based on the aiming time and the vehicle yaw rate to obtain the aiming angle difference, and the instantaneous angle difference is retained, including:

[0076] Step 41: On the unified timeline, record the corrected head heading as an intent proxy field to represent the rider's head direction at the current moment; record the updated vehicle heading as a response proxy field to describe the vehicle's orientation under the same time index. Subsequently, establish a correspondence between the intent proxy field and the response proxy field in the unified timeline index, so that the two can progress over time in a one-to-one correspondence.

[0077] Step 42: At each time point, calculate the difference between the corrected head heading and the updated vehicle heading, and use this difference as the instantaneous angle difference to characterize the rider's current gaze deviation. To avoid jumps caused by angles spanning an entire circumference, the instantaneous angle difference is normalized.

[0078] The vehicle's heading increment within the advance distance after the advance time is calculated by multiplying the preview time by the vehicle's yaw rate. This increment is then added to the updated vehicle heading to obtain the vehicle heading at the preview time. Similarly, the vehicle heading at the preview time is also angle-normalized to make it comparable across different time points.

[0079] Step 43: At each time point, the difference between the corrected head heading and the vehicle heading at the pre-aiming time is calculated to obtain the pre-aiming angle difference, which describes the degree of deviation between the rider's future line of sight and the vehicle's predicted direction of travel. Simultaneously, the vehicle speed is multiplied by the pre-aiming time to obtain the pre-aiming distance, which represents the expected path length of the vehicle within the pre-aiming time.

[0080] Finally, the instantaneous angle difference, the pre-aiming angle difference, the pre-aiming distance, and the vehicle's heading at the pre-aiming moment are output with the same time index, providing consistent data input for the subsequent stability margin assessment module and risk criterion hysteresis module.

[0081] Through the steps described above, this embodiment establishes a correlation between intent and response on a unified timeline. By jointly modeling the current line-of-sight difference and short-term predicted response, it generates two core quantities: instantaneous angle difference and pre-aiming angle difference. This part expresses the cyclist's head direction, the vehicle's current posture, and the future short-term heading in a unified manner, enabling subsequent risk identification processes to conduct deterministic analysis based on consistent temporal and spatial angle information, thus laying a clear data foundation for the accurate identification of cycling risks.

[0082] In one embodiment of the present invention, the equivalent balance roll angle is calculated based on the vehicle speed and the instantaneous curvature of the vehicle body, the attitude stability margin is obtained based on the updated vehicle roll, and the preview distance is obtained by combining the preview time and the vehicle speed, including:

[0083] Step 51: At each time point, first calculate the square of the vehicle speed to obtain the basic lateral force that the vehicle may generate at the current speed. Then, multiply the square of the vehicle speed by the absolute value of the instantaneous curvature of the vehicle body to characterize the lateral acceleration required for the vehicle to achieve lateral balance at that curvature. Next, use gravitational acceleration as the divisor to perform a ratio operation on the above product, and perform an inverse-switching operation on the ratio result to obtain the principal value of the roll angle, which is used as the equivalent balance roll angle of the vehicle at that speed and curvature combination.

[0084] It should be noted that although both steps 34 and 51 use vehicle speed and instantaneous vehicle curvature as inputs, their physical semantics, computational domain affiliation, and output properties are different. Step 34 belongs to the vehicle attitude reconstruction process. Its calculation is based on the geometric features of the current driving trajectory represented by the instantaneous curvature of the vehicle body, and combines the vehicle speed to derive the lateral dynamic state of the vehicle at that moment, thereby obtaining the updated vehicle roll, which describes the actual roll attitude of the vehicle at the current moment. This attitude quantity directly reflects the degree of tilt of the vehicle under real-world conditions and is the basis for subsequent attitude stability margin calculations.

[0085] Step 51 belongs to the stability margin assessment process. Its calculation is based on the vehicle's balance requirements under the current speed and curvature conditions. The calculated equivalent balance roll angle does not represent the vehicle's actual roll attitude at that moment, but rather represents the roll angle that the vehicle should theoretically have when it achieves lateral force balance under the current speed and curvature combination. It is used to compare with the attitude stability margin in the risk criterion to determine whether the current aiming path is within the range achievable by the vehicle.

[0086] Step 52: At each time point, based on the maximum permissible roll angle set before deployment, subtract the updated absolute value of the vehicle roll from the maximum permissible roll angle to obtain the attitude stability margin. The attitude stability margin reflects the remaining available roll space of the vehicle without exceeding the attitude safety limits, and can be used to distinguish whether the vehicle's attitude under current driving conditions is within a critical range.

[0087] Step 53: At each time point, multiply the aiming time by the vehicle speed to represent the distance the vehicle is expected to travel within the aiming time. This product is used as the aiming distance. When the vehicle speed is low, the original speed can be replaced by a preset lower limit constraint to ensure the availability of the aiming distance. After obtaining the aiming distance, the ratio of the aiming angle difference to the aiming distance is used as the reachable curvature, which describes the curvature required for the vehicle to satisfy the aiming angle difference within the aiming distance.

[0088] Through the above steps, this embodiment transforms vehicle motion state parameters into equivalent roll angle and attitude stability margin in attitude space, and generates preview distance and reachable curvature by combining preview time and vehicle speed, thereby constructing a unified analytical foundation for the vehicle attitude domain and preview geometry domain. Simultaneously, all calculations are performed based on a unified time axis, ensuring a consistent and traceable correspondence between preview information, attitude information, and curvature information, providing clear data input for the construction of deterministic risk criteria.

[0089] In one embodiment of the present invention, a deterministic risk criterion is established, and hysteresis control is constructed through entry and exit counts to generate risk status and risk cause identifiers, including:

[0090] Step 61: On a unified timeline, first acquire the aiming angle difference, instantaneous vehicle curvature, vehicle speed, achievable curvature, and attitude stability margin, and simultaneously read the safety angle difference threshold, minimum effective curvature threshold, stability margin safety reduction, and gravitational acceleration constant. If any input is missing, risk assessment is temporarily suspended to avoid misjudgments based on incomplete input.

[0091] Step 62: At the point in time when complete input data is available, establish three types of deterministic risk criteria, including risk of unresponsive intent, risk of reverse response, and risk of unreachability.

[0092] The specific criteria for determining risk include:

[0093] When the absolute value of the pre-aiming angle difference exceeds the safe angle difference threshold and the instantaneous curvature of the vehicle body is lower than the minimum effective curvature threshold, it indicates that the rider's gaze has significantly deviated from the vehicle's direction of travel, and is identified as a risk of unresponsive intent.

[0094] When the pre-aiming angle difference is opposite to the instantaneous curvature of the vehicle body and the absolute value of the pre-aiming angle difference exceeds the safety angle difference threshold, it indicates that the deviation has a significant magnitude and is identified as a reverse response risk.

[0095] The difference between the attitude stability margin and the stability margin safety reduction is calculated. This difference represents the vehicle's remaining roll space after considering attitude safety constraints. When the equivalent balance roll angle calculated based on vehicle speed and reachability curvature is greater than this difference, it is determined to be an unreachable risk, indicating that the vehicle cannot meet the pre-aiming path requirements under attitude safety conditions.

[0096] Step 63: In the continuous time sequence, when the risk trigger signal remains valid, the entry counter increments sequentially according to the time step, and the exit counter is reset to zero; when the risk trigger signal remains invalid, the exit counter increments sequentially, and the entry counter is reset to zero. When the entry counter reaches the preset entry threshold, the risk state is set to open; when the exit counter reaches the preset exit threshold, the risk state is set to closed. Through the above control method, the risk state avoids frequent switching in the time dimension, maintaining logical continuity and hysteresis.

[0097] Step 64: At the point when the risk status changes from closed to open, identify the type of risk that triggered this change, and assign a corresponding category to the risk cause identifier based on the source of the triggering risk. When multiple risk criteria are triggered simultaneously, the final risk cause identifier is determined by a preset priority, ensuring that each opening of the risk status has a clear cause attribution. The risk cause identifier is ultimately classified into three categories: unresponsive intent, reverse response, and unreachable risk, and recorded under a unified timeline index.

[0098] Through the above steps, this embodiment constructs a deterministic risk criterion based on aiming quantity, vehicle response quantity, and attitude margin on a unified time axis; and forms a risk state switching mechanism with hysteresis characteristics through entry counting and exit counting. Under the premise of data integrity confirmation, this module classifies and identifies different risk types, ensuring that risk identification remains consistent in both time and logic, and providing clear information on the source of riding risks for event generation and adaptive threshold updates.

[0099] In one embodiment of the present invention, on a unified timeline, when the risk state changes from closed to open, i.e., at the rising edge of the risk state, an event packet is generated and written to a log file to provide a complete record of the risk occurrence, which is then reported via an interface. The event packet records key data content that triggers the change in risk state, including the current time index, vehicle speed, instantaneous angle difference, pre-aiming angle difference, instantaneous curvature of the vehicle body, reachability curvature, attitude stability margin, and risk cause identifier.

[0100] When the risk status remains off and the steady-state straight-line marker is valid, the safety angle difference threshold and the minimum effective curvature threshold are updated. Specifically, within a set sliding time window, the median value of the absolute value of the instantaneous angle difference and the absolute value of the instantaneous vehicle curvature are calculated to obtain a statistically representative value within the time window. This median value is then weighted and summed with the current threshold in a fixed ratio to generate a candidate threshold. This candidate threshold is then limited to a preset allowable range using interval projection. If a candidate threshold exceeds the upper or lower limit of the allowable interval, it is adjusted to the boundary value of that interval. Interval projection is used to prevent the threshold from shifting excessively in extreme directions, thereby keeping the threshold within a reasonable range. The final updated threshold serves as the new safety angle difference threshold and the minimum effective curvature threshold, and is directly referenced in subsequent risk assessments.

[0101] Through the aforementioned event generation and threshold adaptive update mechanism, this invention fully records key input quantities and risk causes at the time points of risk state changes, making the risk process traceable. Simultaneously, when the risk state stabilizes at "closed" and meets the steady-state straight-line conditions, the safety angle difference threshold and minimum effective curvature threshold are updated. This ensures that the thresholds are adjusted within bounds under the constraints of sliding window statistical results, guaranteeing that subsequent riding risk assessments are always conducted under a unified reference benchmark and stable parameter conditions.

[0102] In one embodiment of the present invention, when abnormal input data is detected or the proportion of invalid samples within the window exceeds a preset proportion threshold, the threshold parameter update is immediately frozen and the freeze event is recorded. When no abnormal data is detected within multiple consecutive sampling periods and the proportion of invalid samples within the window is lower than the preset proportion threshold, the freeze state is lifted. The invalid sampling proportion is the ratio of sampling points with removed index markers to the total number of sampling points in the window. Through this freezing and unfreezing mechanism, the robustness and consistency of the threshold update process are ensured, and parameter offsets caused by incomplete or abnormal data are avoided.

[0103] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0104] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A riding risk identification system based on helmet vehicle end data processing, characterized in that, Comprise: The space-time reference alignment module is used for obtaining the helmet inertial navigation data of the helmet end, the vehicle end motion data of the vehicle end and the navigation auxiliary data, and time alignment is performed to establish three coordinate systems of the road surface, the vehicle body and the helmet, comprising: Step 11, the vehicle end clock is constructed to construct an equal interval unified time axis, the helmet inertial navigation data, the vehicle end motion data and the navigation auxiliary data are linearly interpolated and bound to the unified time axis index according to the time stamp; wherein the helmet inertial navigation data comprises a helmet yaw angular velocity; the vehicle end motion data comprises a vehicle body yaw rate, a vehicle body lateral acceleration and a vehicle speed; the navigation auxiliary data comprises a navigation heading and a navigation position; Step 12, on the unified time axis, the linear drift coefficient is set to zero, the time deviation of the helmet end and the vehicle end is enumerated according to the candidate set, for each time deviation, the absolute difference value of the vehicle body yaw rate and the helmet yaw angular velocity is calculated point by point according to the unified time axis, and the sum is accumulated, and the minimum sum result is taken as the time deviation initial value; the time deviation and the linear drift coefficient are searched jointly with the initial value, and the correction parameter pair is determined according to the minimum sum result, the lateral acceleration consistency residual is obtained by calculating the product of the vehicle speed and the vehicle body yaw rate minus the vehicle body lateral acceleration, and the unified time axis sampling points exceeding the threshold are marked and removed; Step 13, the road surface coordinate system, the vehicle body coordinate system and the helmet coordinate system are defined as right-handed coordinate systems, respectively fixing the forward, left and upward directions and the origin position, keeping the naming, axis direction and origin unchanged under the unified time axis index; The steady straight alignment module is used for calculating the heading difference and the roll difference when the vehicle is in steady straight, updating the heading bias and the roll bias, obtaining the corrected head heading and head roll, and outputting the steady straight marker, comprising: Step 21, on the unified time axis, the product of the vehicle speed and the vehicle body yaw rate is calculated to obtain the lateral acceleration; when the vehicle speed threshold, the yaw rate threshold and the lateral acceleration threshold are satisfied at the same time in the sliding window, the steady straight marker is set to valid, otherwise it is set to invalid; Step 22, on the sample with the steady straight marker being valid, the difference between the head heading and the vehicle body heading is calculated to obtain the heading difference, and the angle is normalized, and the difference between the head roll and the vehicle body roll is calculated to obtain the roll difference; the heading difference and the roll difference are taken as the observation of the heading bias and the roll bias; Step 23, in the sliding window with the steady straight marker being valid, the window average of the heading difference and the roll difference is calculated respectively, the weighted sum of the previous time heading bias and the window average heading difference is calculated to obtain the updated heading bias, and the weighted sum of the previous time roll bias and the window average roll difference is calculated to obtain the updated roll bias; the difference between the head heading and the heading bias is calculated to obtain the corrected head heading, and the difference between the head roll and the roll bias is calculated to obtain the corrected head roll, and the steady straight marker is outputted; The vehicle state integration module is used for time domain integration based on the vehicle end motion data to obtain the updated vehicle body heading, the updated vehicle body roll and the vehicle body instantaneous curvature; The intention response modeling module is configured to take the corrected head direction as an intention agent, take the updated vehicle body direction as a response agent, determine the vehicle body direction at the preview time according to the preview time and the vehicle body yaw rate, obtain a preview angle difference, and reserve an instant angle difference, and includes the following steps: Step 41: On the unified time axis, record the corrected head direction as an intention agent field, record the updated vehicle body direction as a response agent field, and establish a corresponding relationship between the field names and time indexes of the two fields in the unified time axis index; Step 42: At each time point, calculate the difference between the corrected head direction and the updated vehicle body direction to obtain an instant angle difference, add the product of the preview time and the vehicle body yaw rate to the updated vehicle body direction to obtain the vehicle body direction at the preview time, and perform angle normalization respectively; Step 43: At each time point, calculate the difference between the corrected head direction and the vehicle body direction at the preview time to obtain a preview angle difference, take the product of the vehicle speed and the preview time as a preview distance, and output the instant angle difference, the preview angle difference, the preview distance, and the vehicle body direction at the preview time; The stability margin evaluation module is configured to calculate an equivalent balance roll angle according to the vehicle speed and the vehicle body instantaneous curvature, obtain a posture stability margin according to the updated vehicle body roll, and combine the preview time and the vehicle speed to obtain a preview distance; The risk criterion hysteresis module is configured to establish a deterministic risk criterion, construct hysteresis control through entry count and exit count, and generate a risk state and a risk cause identifier; The event generation and reporting module is configured to generate an event package at a rising edge when the risk state changes from off to on, perform log writing and interface reporting, and perform bounded updating on threshold parameters when the steady straight line marker is valid and the risk state is off.

2. The riding risk identification system based on helmet vehicle end data processing according to claim 1, characterized in that, Based on the vehicle end motion data, the updated vehicle body direction, the updated vehicle body roll, and the vehicle body instantaneous curvature are obtained by time domain integration, including the following steps: Step 31: On the unified time axis, exclude the sampling points marked by the rejection index, take the time interval between adjacent sampling points as the time step, and take the vehicle body direction of the first valid sample as the initial value; Step 32: At each time step, calculate the product of the vehicle body yaw rate and the time step, and add the updated vehicle body direction at the previous time to obtain the updated vehicle body direction at the next time, and perform angle normalization; Step 33: At each time step, calculate the ratio of the vehicle body yaw rate to the vehicle speed to obtain the vehicle body instantaneous curvature, replace the vehicle speed with a preset lower limit when the vehicle speed is lower than the preset lower limit, subtract the difference between the vehicle lateral acceleration and the product of the vehicle speed and the vehicle body yaw rate as the lateral consistency difference, and mark the suspicious sampling points on the unified time axis when the difference exceeds the threshold; Step 34: At each time step, calculate the product of the square of the vehicle speed and the vehicle body instantaneous curvature, divide by the gravitational acceleration, take the inverse tangent to obtain the principal value of the roll angle, determine the roll direction according to the sign of the vehicle body instantaneous curvature, and obtain the updated vehicle body roll.

3. The riding risk identification system based on helmet-car-end data processing according to claim 1, characterized in that, The stability margin evaluation module is configured to calculate an equivalent balance roll angle according to the vehicle speed and the vehicle body instantaneous curvature, obtain a posture stability margin according to the updated vehicle body roll, and combine the preview time and the vehicle speed to obtain a preview distance; Step 51, at each time point, square the vehicle speed and multiply it by the absolute value of the instantaneous curvature of the vehicle body, divide by the gravitational acceleration, and take the inverse tangent as the principal value of the equivalent balanced roll angle; Step 52, at each time point, subtract the absolute value of the updated vehicle body roll from the maximum allowed roll angle to obtain the attitude stability margin; Step 53, at each time point, take the product of the preview time and the vehicle speed as the preview distance, and take the ratio of the preview angle difference and the preview distance as the reachable curvature, and output the equivalent balanced roll angle, the attitude stability margin, the preview distance and the reachable curvature.

4. The cycling risk identification system based on helmet-car end data processing according to claim 1, characterized in that, A deterministic risk criterion is established, and a hysteresis control is constructed through entry counting and exit counting to generate a risk state and a risk cause identifier, including: Step 61, on a unified time axis, obtain the preview angle difference, the instantaneous curvature of the vehicle body, the vehicle speed, the reachable curvature and the attitude stability margin, and synchronously read the safety angle difference threshold, the minimum effective curvature threshold, the stability margin safety reduction amount and the gravitational acceleration constant; when any input is missing, the risk judgment is suspended; Step 62, establish a deterministic risk criterion including intention non-response risk, reverse response risk and unreachable risk, if any risk is established, the risk trigger signal is set to valid; Step 63, when the risk trigger signal is in the valid state for a preset time, the entry counter is incremented; when the risk trigger signal is continuously in the invalid state, the exit counter is incremented; when the counter cumulative value reaches the corresponding threshold, the opening and closing of the risk state are triggered; Step 64, when the risk state is turned on from off, the risk cause identifier is determined according to the trigger signal source, and the risk cause identifier is divided into three categories of intention non-response, reverse response and unreachable risk according to the preset priority.

5. The cycling risk identification system based on helmet-car end data processing according to claim 4, characterized in that, The deterministic risk criterion specifically includes: When the absolute value of the preview angle difference exceeds the safety angle difference threshold and the instantaneous curvature of the vehicle body is lower than the minimum effective curvature threshold, it is determined as intention non-response risk; When the preview angle difference and the instantaneous curvature of the vehicle body are opposite and the absolute value of the preview angle difference exceeds the safety angle difference threshold, it is determined as reverse response risk; Calculate the difference between the attitude stability margin and the stability margin safety reduction amount, and when the equivalent balanced roll angle calculated according to the vehicle speed and the reachable curvature is greater than the difference, it is determined as unreachable risk.

6. The cycling risk identification system based on helmet-car end data processing according to claim 1, characterized in that, An event package is generated at the rising edge when the risk state is turned on from off, the event package is written into a log file and reported through an interface, in the case that the risk state is stable as off and the straight driving flag is valid, the safety angle difference threshold and the minimum effective curvature threshold are executed for bounded adaptive update, wherein the absolute value of the instantaneous angle difference and the absolute value of the instantaneous curvature of the vehicle body are respectively taken as the median in the sliding time window, the candidate threshold is formed by weighting the sum of the current threshold by a fixed proportion, and the candidate threshold is limited in the preset allowed range through interval projection; the event package includes the current time index, the vehicle speed, the instantaneous angle difference, the preview angle difference, the instantaneous curvature of the vehicle body, the reachable curvature, the attitude stability margin and the risk cause identifier.

7. The cycling risk identification system based on helmet-car-end data processing according to claim 6, characterized in that, When the input data is detected to be abnormal or the proportion of invalid samples in the window exceeds the preset proportion threshold, the threshold parameter update is immediately frozen and a freezing event is recorded, and when no data abnormality is detected in a plurality of consecutive sampling periods and the proportion of invalid samples in the window is lower than the preset proportion threshold, the frozen state is released, wherein the proportion of invalid samples is the proportion of sampling points marked with the elimination index in the window sampling points.

Images