An intelligent bicycle navigation system and method

Abstract

This invention discloses a navigation system based on an intelligent bicycle, belonging to the field of navigation technology. It addresses the problems of path planning and safety control under dynamic road conditions. The system first constructs a digital situational awareness model of the vehicle by synchronously acquiring data from a multi-dimensional motion sensor group and wheel speed pulse sequences to calculate real-time attitude angles, speed decay, and lateral sway amplitude. A road condition analysis module uses computer vision algorithms to reconstruct a three-dimensional digital model of the road surface, identifying height abrupt changes and determining obstacle crossing costs. A cost fusion module uses a multi-modal data fusion algorithm to map vehicle condition parameters into energy factors and generates a navigation cost chain that has undergone local smoothing. A path generation module performs path search based on this cost chain and outputs the optimal steering command. This invention achieves a deep mapping between the physical driving environment and the digital cost space, significantly improving navigation accuracy under complex dynamic road conditions.

Description

Technical Field

[0001] This invention relates to the field of navigation technology, specifically to a smart bicycle navigation system and method. Background Technology

[0002] With the diversification of urban traffic environments and the expansion of the cycling population, cycling is extending from traditional commuting to long-distance leisure rides, suburban exploration, and mixed road conditions. Cycling is significantly more sensitive to route safety, controllability, and energy consumption than motorized travel; even slight changes in road surface, potholes, or slopes can significantly alter a bicycle's handling stability. Consequently, people expect navigation systems to not only provide the shortest route but also make more direct assessments of road accessibility, obstacle distribution, and travel costs. Navigation technology is thus evolving from single-location guidance towards dynamic environmental understanding and risk avoidance.

[0003] Current navigation technologies for bicycles still primarily rely on static recognition of elevation data, geographic road tags, or photographic images. These technologies have limited data sampling resolution and are significantly lagging in responding to fine-grained changes in road surface texture. When encountering short-scale obstacles such as gravel, potholes, or localized undulations, navigation results often fail to reflect the rider's actual handling load. Vision-based road surface recognition is prone to texture misjudgment in low light, backlight, or low-angle photography, leading to obstacle location drift. Solutions relying on external map elevation information are prone to unstable cost estimation on continuously changing road sections, causing abrupt changes in path planning or even creating unfeasible routes. Furthermore, existing methods typically fail to incorporate the vehicle's yaw, posture, and short-term speed changes into road condition assessments, resulting in a low correlation between navigation and the rider's actual handling experience, making it difficult to provide stable and controllable route suggestions in complex road conditions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent bicycle navigation system and method, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a smart bicycle navigation system comprising the following modules: a vehicle condition acquisition module, used to simultaneously acquire triaxial motion data from the onboard inertial measurement unit and wheel speed pulse sequences from the wheel encoder, and calculate the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle based on the temporal correspondence between the triaxial acceleration data and the wheel speed pulse sequence, generating a set of vehicle condition parameters representing the controllability of the current road segment; and a road condition analysis module, used to simultaneously call continuous road surface image frames captured by the onboard low-angle camera unit within the time window of the vehicle condition parameter set, calculate the local surface contour change through inter-frame pixel displacement, perform secondary screening of contact plane height abrupt change points, and combine the lateral sway amplitude from the vehicle condition parameter set with the analysis of the road condition parameters. The system uses dynamic amplitude and speed attenuation to determine the location, type, and crossing cost of obstacles on corresponding road segments, generating road condition description data. The cost fusion module maps the tilt angle and short-term speed attenuation from the vehicle condition parameter set to energy consumption factors based on the road condition description data. It then constructs a navigation cost sequence for continuous road segments through linear weight accumulation and locally smooths cost abrupt changes in the cost sequence to form a navigation cost chain. The path generation module performs segment-by-segment path search under the constraints of the navigation cost chain, calculates the unit travel cost increment for each candidate path segment, and blocks path extension when the cost increment exceeds a preset upper limit. This determines an effective navigation path that meets the bicycle handling stability threshold and outputs a sequence of steering commands for continuous path points.

[0006] Furthermore, the specific process of synchronously acquiring the three-axis motion data of the vehicle-mounted inertial measurement unit and the wheel speed pulse sequence of the wheel encoder is as follows: A time synchronization mechanism is established between the inertial measurement unit and the wheel encoder. Hardware interrupt signals ensure that the sampling clocks of the two sensors are synchronized. The raw acceleration and angular velocity data output by the inertial measurement unit are normalized to the coordinate system, transforming them from the sensor coordinate system to the bicycle's body coordinate system. Simultaneously, the pulse signal of the wheel encoder is periodically measured and pulses are counted. Combined with the known wheel circumference, the linear velocity and displacement of the bicycle are calculated in real time. The transformed three-axis motion data and the calculated wheel speed pulse sequence are packaged according to a unified timestamp to form a synchronized sensor data stream.

[0007] Furthermore, based on the temporal correspondence between triaxial acceleration data and wheel speed pulse sequences, the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle are calculated, and the specific process for generating a set of vehicle condition parameters representing the controllability of the current road segment is as follows: Based on synchronized sensor data streams, accelerometer and gyroscope data are fused using a complementary filtering algorithm to calculate the real-time attitude angle of the bicycle during travel. The changing trend of the wheel speed pulse sequence within the sliding time window is analyzed, and the differential characteristics of the speed are calculated to obtain the short-term speed decay, which characterizes the change in resistance. By monitoring the periodic changes in lateral acceleration in the bicycle's body coordinate system and combining it with attitude angle data, the lateral sway amplitude parameter is quantified. The real-time attitude angle, speed decay, and lateral sway amplitude parameters are normalized to construct a multi-dimensional set of vehicle condition parameters characterizing the handling stability of the bicycle on the current road segment.

[0008] Furthermore, within the time window of the vehicle condition parameter set, continuous road surface image frames captured by the vehicle-mounted low-angle camera unit are synchronously invoked. The local surface contour change is calculated through inter-frame pixel displacement, and the specific process for secondary screening of contact plane height abrupt change points is as follows: Based on the changing trend of speed attenuation in the vehicle condition parameter set, the sampling frequency and exposure time of the low-angle camera unit are dynamically adjusted. Feature points are extracted and matched on the captured continuous road surface image frames, and the pixel displacement vector between adjacent frames is calculated using the optical flow method. Based on the intrinsic and extrinsic parameters of the camera unit and the pixel displacement vector, the three-dimensional contour of the local surface is reconstructed using the triangulation principle. Height gradient analysis is performed on the reconstructed three-dimensional contour data to identify areas where the height change exceeds a set threshold, which are marked as contact plane height abrupt change points. Combining the distribution of height abrupt change points in multiple frames, noise interference is eliminated through cluster analysis, completing the secondary screening of height abrupt change points.

[0009] Furthermore, by combining the lateral sway amplitude and speed decay in the vehicle condition parameter set, the specific process for determining the obstacle location, obstacle type, and obstacle crossing cost of the corresponding road segment and generating road condition description data is as follows: The height abrupt change points obtained from the secondary screening are subjected to spatiotemporal correlation analysis with the vehicle condition parameter set to establish a mapping relationship between vehicle dynamics response and road surface geometric changes. By analyzing the changing trends of lateral sway amplitude and speed decay relative to the baseline value, abnormal vehicle response segments are identified, and the corresponding height abrupt change points within these segments are confirmed as effective obstacle locations. Based on the three-dimensional geometric features of the height abrupt change points at the effective obstacle locations and their spatial distribution patterns in consecutive frames, combined with the change amplitude and duration of speed decay, the cumulative impact of obstacles on driving resistance is assessed, and obstacle types are identified. Based on the obstacle type identification results and geometric feature parameters, the required passage safety boundary and expected kinetic energy loss for different types of obstacles are calculated, quantifying the comprehensive crossing cost of each obstacle. The obstacle location coordinates, obstacle type identifiers, and quantified obstacle crossing costs are structurally integrated to generate road condition description data.

[0010] Furthermore, based on the road condition description data, the tilt angle and short-term speed attenuation in the vehicle condition parameter set are mapped to energy consumption factors. The specific process of constructing a navigation cost sequence for continuous road segments through linear weight accumulation is as follows: Obstacle location information and crossing costs are extracted from the road condition description data. Within the same time window, the tilt angle and short-term speed attenuation in the vehicle condition parameter set are retrieved. Based on the changes in gravity components caused by the tilt angle and the changes in driving resistance reflected by the speed attenuation, an energy consumption factor reflecting the energy demand for a single segment is established. The energy consumption factor within each time window is linearly accumulated in the continuous order of road segments. The accumulated value is dynamically corrected in conjunction with the obstacle crossing cost, so that the cost can simultaneously reflect changes in slope, resistance, and obstacle crossing difficulty, resulting in an initial navigation cost sequence covering the entire target road segment. After the initial cost sequence is generated, the time consistency of the cost change rate is checked, and abnormal jump points caused by sensing errors are eliminated.

[0011] Furthermore, the specific process of locally smoothing cost abrupt change segments in the cost value sequence to form a navigation cost chain is as follows: Detect gradient changes in the navigation cost value sequence, and identify segments where the cost value change rate exceeds a set multiple of the average change rate as abrupt change segments to be processed; establish an adaptive smoothing window within the abrupt change segments, and use a local regression algorithm to fit the trend of the cost value, suppressing abnormal spikes caused by instantaneous measurement errors while maintaining the cost value characteristics of densely obstacle-prone areas; perform continuity optimization on the smoothed cost value sequence, and use interpolation to ensure smooth transitions in cost values ​​between adjacent segments, forming a navigation cost chain with spatial consistency.

[0012] Furthermore, the path generation module includes the following steps: Under the constraints of the navigation cost chain, starting from the starting position, a path search is performed on the feasible road segments, extending segment by segment. The endpoints of the current candidate path are mapped to the corresponding positions in the navigation cost chain, and the corresponding cost value is read as the path extension condition. During each path extension process, the unit travel cost increment of the candidate path segment is calculated based on the length, direction change, and local cost value in the navigation cost chain, and the cost increment is added to the current total path cost. When the unit travel cost increment of the candidate path segment exceeds the preset cost upper limit, the extension of the path branch is terminated. Among all candidate path branches, the path that meets the handling stability threshold and has the lowest total cost is selected. The path is parsed into a continuous path point sequence in chronological order, and a steering command sequence that can be directly used by the on-board execution unit is generated based on the direction change between the path points.

[0013] A navigation method based on intelligent bicycles includes the following steps: S1. Synchronously acquiring three-axis motion data from the onboard inertial measurement unit and wheel speed pulse sequences from the wheel encoder, and calculating the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle based on the temporal correspondence between the three-axis acceleration data and the wheel speed pulse sequence, generating a set of vehicle condition parameters for the controllability of the current road segment; S2. Synchronously calling continuous road surface image frames captured by the onboard low-angle camera unit within the time window of the vehicle condition parameter set, calculating the local surface contour change through inter-frame pixel displacement, performing secondary screening of contact plane height abrupt change points, and determining the lateral sway amplitude and speed decay in combination with the vehicle condition parameter set. S3. Based on the road condition description data, the tilt angle and short-term speed decay in the vehicle condition parameter set are mapped to energy consumption factors. A navigation cost sequence for continuous road segments is constructed by accumulating linear weights, and local smoothing is performed on cost abrupt segments in the cost sequence to form a navigation cost chain. S4. Under the constraints of the navigation cost chain, a path search is performed segment by segment. The unit travel cost increment is calculated for each candidate path segment, and the path extension is blocked when the cost increment exceeds the preset upper limit. An effective navigation path that meets the bicycle handling stability threshold is determined, and a sequence of steering instructions for continuous path points is output.

[0014] The present invention has the following beneficial effects:

[0015] (1) A smart bicycle navigation system that uses a vehicle condition acquisition module to synchronously correlate inertial measurement unit data with wheel speed pulses, enabling complementary verification of dynamic information from different physical quantities. The bicycle's attitude angle, lateral sway, and short-term speed decay can be quantified in real time, thus obtaining a set of vehicle condition parameters that reflect the degree of handling stability. After this parameter set is linked with the road condition analysis module, abrupt changes in the ground surface in the visual image can be screened a second time under the constraints of the vehicle's dynamic characteristics, significantly reducing misjudgments caused by changes in lighting, texture, or viewing angle, and making the identification of obstacle location, type, and crossing cost closer to the rider's real handling experience.

[0016] (2) A navigation method based on intelligent bicycles maps road condition description data and the vehicle's tilt angle and speed decay characteristics into an energy consumption factor, forming a navigation cost sequence for continuous road segments. Unstable transition segments are locally smoothed, ensuring the navigation cost chain remains continuous and interpretable even in complex road conditions. Under the constraints of the navigation cost chain, a segment-by-segment expansion search is performed, automatically interrupting the extension of uncontrollable road segments with sudden cost increases. This allows the obtained route to naturally avoid segments with excessive control load, ultimately outputting an effective navigation path that meets stable riding conditions, improving the feasibility and safety of navigation in scenarios such as gravel roads, undulating surfaces, and narrow roads.

[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0018] Figure 1 This is a flowchart of an intelligent bicycle navigation system according to the present invention.

[0019] Figure 2 This is a flowchart of a smart bicycle navigation method according to the present invention. Detailed Implementation

[0020] This application provides an intelligent bicycle navigation system and method that solves the technical problems of existing navigation systems being unable to assess road surface controllability based on the actual vehicle's dynamics and being unable to accurately identify minor obstacles, leading to route planning deviating from actual drivability.

[0021] The overall concept of the solution in this application embodiment is as follows:

[0022] During bicycle riding, the vehicle condition acquisition module simultaneously acquires triaxial motion data from the inertial measurement unit and wheel speed pulse sequences from the wheel encoder. By analyzing the temporal correlation between these two types of data, it calculates attitude angles, lateral sway amplitude, and short-term speed decay, forming a set of vehicle condition parameters characterizing the vehicle's handling stability. Under the temporal constraints of this parameter set, the road condition analysis module calls upon continuous image frames from the onboard low-angle camera unit. Based on inter-frame pixel displacement, it determines local surface contour changes, performs secondary screening of height abrupt changes, and identifies obstacle locations, obstacle types, and their crossing costs by combining the bicycle's dynamic response characteristics. Subsequently, the cost fusion module uses the vehicle condition parameters and obstacle information to construct a navigation cost sequence for continuous road segments and smooths abrupt changes to generate a navigation cost chain. Finally, under the constraints of the navigation cost chain, the path generation module performs segment-by-segment path expansion, filtering out uncontrollable road segments based on the unit travel cost increment, obtaining an effective navigation path that meets stable riding conditions. This achieves consistency between the navigation route and actual controllable riding ability, thereby improving the reliability and safety of navigation planning under complex road conditions.

[0023] Please see Figure 1This invention provides a technical solution: a smart bicycle navigation system comprising the following modules: a vehicle condition acquisition module, used to simultaneously acquire triaxial motion data from the onboard inertial measurement unit and wheel speed pulse sequences from the wheel encoder, and calculate the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle based on the temporal correspondence between the triaxial acceleration data and the wheel speed pulse sequence, generating a vehicle condition parameter set of the current road segment's controllability status; and a road condition analysis module, used to simultaneously call continuous road surface image frames captured by the onboard low-angle camera unit within the time window of the vehicle condition parameter set, calculate the local surface contour change through inter-frame pixel displacement, perform secondary screening of contact plane height abrupt change points, and combine the lateral sway amplitude in the vehicle condition parameter set with... The system comprises several modules: a speed attenuation module, a cost fusion module, and a path generation module. The former determines the location, type, and crossing cost of obstacles on the corresponding road segment, generating road condition description data. The latter maps the tilt angle and short-term speed attenuation from the vehicle condition parameter set to energy consumption factors based on the road condition description data. It then accumulates these factors using linear weights to construct a navigation cost sequence for continuous road segments and performs local smoothing on cost abrupt changes in the cost sequence to form a navigation cost chain. The latter performs a path search segment-by-segment under the constraints of the navigation cost chain, calculates the unit travel cost increment for each candidate path segment, and blocks path extension when the cost increment exceeds a preset upper limit. This determines an effective navigation path that meets the bicycle handling stability threshold and outputs a sequence of steering commands for continuous path points.

[0024] In this implementation scheme, the vehicle condition acquisition module is used to simultaneously acquire triaxial motion data output by the inertial measurement unit (IMU) and wheel speed pulse sequences generated by the wheel encoder during bicycle operation. An IMU is a combination of sensors used to measure motion states such as acceleration and angular velocity, reflecting real-time attitude changes when the vehicle experiences bumps, acceleration, or turning. The wheel speed pulse sequence is generated by the encoder through wheel rotation and characterizes the bicycle's instantaneous speed changes. This module calculates the bicycle's attitude angle, lateral sway amplitude, and short-term speed decay by analyzing the correspondence between triaxial acceleration and wheel speed pulses over time, thus forming a set of vehicle condition parameters reflecting the current road segment's controllability, providing a dynamic basis for subsequent obstacle recognition and path planning. The road condition analysis module synchronously calls up continuous road surface image frames acquired by the onboard low-angle camera unit within the time window corresponding to the vehicle condition parameter set, estimating the change in ground surface contour over time through inter-frame pixel displacement. Inter-frame pixel displacement refers to the change in the projected position of the same physical point between adjacent images, which can be used to calculate geometric features such as road surface height and slope. This module further screens detected height abrupt changes to eliminate misjudgments caused by non-realistic obstacles such as lighting and shadows. Then, based on the response patterns of lateral sway and speed decay in the vehicle condition parameter set, it determines the spatial location, type, and crossing cost of obstacles, ultimately generating structured road condition description data, providing both geometric and dynamic basis for navigation cost calculation. The cost fusion module maps attitude angles and short-term speed decay into quantifiable energy consumption factors based on the road condition description data and vehicle condition parameter set. Here, the energy consumption factor is not a direct calculation of physical energy, but a quantitative indicator used to express the difficulty of handling under different driving conditions. This module constructs a navigation cost sequence for continuous road segments by linearly weighting obstacle crossing costs, obstacle locations, and road segment morphology, and performs local smoothing on segments with abrupt changes in the sequence to eliminate sharp jumps caused by instantaneous attitude fluctuations. The smoothed navigation cost chain more stably reflects path continuity and riding controllability, serving as a constraint for path search. The path generation module uses the navigation cost chain as a constraint to perform a segment-by-segment extended search of candidate paths. The term "segment-by-segment extension" refers to dividing the path into continuous small segments and calculating the cost increment per unit distance in each segment based on the navigation cost chain to dynamically determine whether the segment meets the riding stability conditions. Once the cost increment exceeds a preset upper limit, the segment is considered to have potentially insurmountable obstacles or significant handling risks, thus blocking the further extension of the path. This module ultimately selects the path that meets the handling stability threshold and transforms it into a sequence of steering commands composed of continuous path points, allowing the navigation results to be directly used for vehicle control or user prompts, achieving consistency between the navigation path and actual riding ability.

[0025] Specifically, the process of synchronously acquiring the three-axis motion data of the vehicle-mounted inertial measurement unit and the wheel speed pulse sequence of the wheel encoder is as follows: A time synchronization mechanism is established between the inertial measurement unit and the wheel encoder. Hardware interrupt signals ensure that the sampling clocks of the two sensors are synchronized. The raw acceleration and angular velocity data output by the inertial measurement unit are normalized to the coordinate system, transforming them from the sensor coordinate system to the bicycle's body coordinate system. Simultaneously, the pulse signal of the wheel encoder is periodically measured and pulses are counted. Combined with the known wheel circumference, the linear velocity and displacement of the bicycle are calculated in real time. The transformed three-axis motion data and the calculated wheel speed pulse sequence are packaged according to a unified timestamp to form a synchronized sensor data stream.

[0026] In this implementation scheme, a time synchronization mechanism is established and sampling is triggered: the inertial measurement unit (IMU) and the wheel encoder are connected to the same sampling control unit via a hardware interrupt line, and a unified sampling time base is configured. When the sampling time arrives, a hardware interrupt triggers both types of sensors to simultaneously output raw measurement data. This measure is used to eliminate data misalignment caused by asynchronous sampling and ensure the accuracy of subsequent time-based fusion calculations. Sensor data preprocessing and coordinate system normalization: the raw acceleration and angular velocity vectors output by the IMU are subjected to zero-bias correction and temperature drift compensation (bias is measured during system initialization or periodic calibration), and the acceleration vector in the sensor coordinate system is converted to the bicycle body coordinate system based on the calibration matrix at the time of installation. Parameter description: This represents the three-axis acceleration vector in the bicycle's body coordinate system; This represents the rotation matrix from the sensor to the main body obtained from the installation calibration; This represents the three-axis acceleration vector (in the sensor coordinate system) of the IMU's raw output. Calibration matrix. Obtained through a static calibration procedure, which includes multi-pose sampling and least-squares fitting to eliminate installation errors and project subsequent measurements onto a unified reference frame. Encoder pulse to linear velocity / displacement conversion: The pulse count of the wheel encoder is periodically measured and accumulated within each synchronous sampling window, and the linear velocity within the current window is calculated using the following formula: Parameter description: The estimated linear velocity within this window; The number of pulses accumulated by the encoder within this time window; The actual rolling length of the wheel corresponding to each pulse (determined by the wheel circumference and the number of pulses per encoder revolution). The duration of this synchronous sampling window. Displacement is measured within a continuous window. The cumulative values ​​are obtained. During periodic measurement, pulse loss or jitter is detected in parallel for anomaly labeling. Multimodal data stream packaging: The three-axis acceleration vector and angular velocity vector after coordinate transformation, along with linear velocity, displacement, and other parameters derived from the encoder, are packaged according to a unified timestamp to form a strictly time-aligned sensor data record (multimodal sensor data stream). This data stream serves as the direct input for subsequent attitude calculation and road condition analysis. The time alignment strategy is completed at the sampling level, and sequence numbers and verification information are attached at the communication level to ensure data integrity.

[0027] Specifically, the process of calculating the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle based on the temporal correspondence between triaxial acceleration data and wheel speed pulse sequences, and generating a set of vehicle condition parameters representing the controllability of the current road segment, is as follows: Based on synchronized sensor data streams, accelerometer and gyroscope data are fused using a complementary filtering algorithm to calculate the real-time attitude angle of the bicycle during travel. The changing trend of the wheel speed pulse sequence within the sliding time window is analyzed, and the differential characteristics of the speed are calculated to obtain the short-term speed decay, which characterizes the change in resistance. By monitoring the periodic changes in lateral acceleration in the bicycle's body coordinate system and combining it with attitude angle data, the lateral sway amplitude parameter is quantified. The real-time attitude angle, speed decay, and lateral sway amplitude parameters are normalized to construct a multi-dimensional set of vehicle condition parameters characterizing the handling stability of the bicycle on the current road segment.

[0028] In this implementation scheme, the real-time attitude angle calculation (complementary filtering) aims to combine the characteristics of a gyroscope (fast short-term response but prone to integral drift) with the characteristics of an accelerometer (long-term stable indication of the direction of gravity) to obtain a smooth and drift-resistant attitude angle estimate, which serves as the basic input for determining control stability. The core calculation (complementary filtering) is expressed as follows: Parameter description: : The estimated attitude angle at time k (angle about a certain axis); : The estimated attitude angle at time k-1; : The angular velocity component output by the IMU at time k (corresponding to the attitude angle axis); : Sampling time interval; The instantaneous acceleration reference angle is calculated by projecting the acceleration vector onto the direction of gravity. Complementary filter weighting coefficients. The determination method is as follows: During the system initialization phase, noise statistics are performed on short-time data windows of stationary or uniform straight-line motion, and the variance of the gyroscope angular velocity sequence is estimated respectively. variance of the angle sequence derived from the accelerometer And calculate the self-adaptive calculation according to the following formula. : Parameter description: : Variance estimation of gyroscope output angular velocity; Variance estimation of the angle derived from acceleration. When gyroscope noise is relatively low, the confidence level of the integral term of angular velocity is increased; when acceleration measurement is more stable, the confidence level of the acceleration reference angle is increased, thus adaptively balancing the contributions of the two sensors. Extraction of short-time speed decay: The purpose is to quantify the magnitude and rate of speed decrease within a short timescale to reflect the impact of sudden increases in road resistance or deceleration events (such as encountering obstacles, steep slopes, changes in friction) on driving energy consumption and handling. Calculation steps and formulas: Calculate the speed difference over a time series with a sliding window length of N and take the positive decay component: Parameter description: D: represents the short-term velocity decay within the sliding window; : The estimated linear velocity at time k; : Estimated linear velocity at the start of the window; N: Number of sampling points within the sliding window; Sampling time interval. Explanation: This definition ensures that when speed decreases, D is positive, indicating the rate of deceleration; zero indicates no decay. The sliding window N can be empirically set based on typical riding gait and environmental changes. Quantification of lateral sway amplitude: The purpose is to measure the vehicle's lateral instability performance, providing a basis for identifying lateral vibrations caused by narrow road deviation, obstacle avoidance swaying, or uneven road surfaces. Calculation steps and formula (root mean square value): Parameter description: S: Represents the root mean square value of the lateral sway amplitude; : The lateral acceleration component at the m-th sampling point in the body coordinate system; M: The number of sample points used for calculation. Explanation: The RMS index is used because it can simultaneously reflect amplitude and frequency, distinguishing between a single short impact and a sustained oscillation; in actual operation, spectral analysis can be performed in parallel to identify periodic oscillation components and use them for further feature extraction. Parameter normalization and vehicle condition parameter set formation: The above... The original quantities such as D and S are mapped to a unified scale to form a multi-dimensional vehicle condition parameter set. This is used for subsequent coupling and judgment with visual measurement results. Normalization adopts the historical baseline or field calibration percentile method: the 95th percentile value is calculated for a baseline data segment without special events. and median and normalize it according to the following formula: Parameter description: c: The component used to construct the vehicle condition parameter set after normalization; x: The original quantity to be normalized (can be...) (or D or S, etc.); Historical median baseline; Historical 95th percentile baseline. Explanation: The percentile method, rather than the extreme value method, is used because it is robust to outliers and provides a comparable scale for each dimension in subsequent cost calculations. The normalized components are combined to form a multi-dimensional vehicle condition parameter set. The data is then transmitted along with the timestamp to the traffic condition analysis module for spatiotemporal registration.

[0029] Specifically, within the time window of the vehicle condition parameter set, continuous road surface image frames captured by the vehicle-mounted low-angle camera unit are synchronously invoked. The local surface contour change is calculated through inter-frame pixel displacement, and the secondary screening of contact plane height abrupt change points is performed as follows: Based on the changing trend of speed attenuation in the vehicle condition parameter set, the sampling frequency and exposure time of the low-angle camera unit are dynamically adjusted. Feature points are extracted and matched on the captured continuous road surface image frames, and the pixel displacement vector between adjacent frames is calculated using the optical flow method. Based on the intrinsic and extrinsic parameters of the camera unit and the pixel displacement vector, the three-dimensional contour of the local surface is reconstructed using the triangulation principle. Height gradient analysis is performed on the reconstructed three-dimensional contour data to identify areas where the height change exceeds a set threshold, which are marked as contact plane height abrupt change points. Combining the distribution of height abrupt change points in multiple frames, noise interference is eliminated through cluster analysis, completing the secondary screening of height abrupt change points.

[0030] In this implementation scheme, adaptive acquisition parameter settings are used: the frame rate and exposure time of the low-angle camera unit are adjusted according to the changing trend of short-term speed decay in the vehicle condition parameter set. Speed ​​changes affect imaging blur and exposure requirements; at low speeds, the frame rate can be reduced and the exposure increased to improve texture details, while at high speeds, the frame rate can be increased and the exposure shortened to reduce motion blur. The normalized speed decay component in the vehicle condition parameter set is read, and the frame rate and exposure are selected according to the mapping table. This ensures the robustness of subsequent feature point matching and optical flow calculation, reducing optical measurement errors. Feature point extraction and optical flow calculation: Feature point extraction (such as corner points or FAST / SIFT type points) is performed on consecutive image frames, and dense or sparse optical flow methods are used to calculate the pixel displacement vector field between adjacent frames. Pixel displacement reflects the parallax of the camera relative to the road surface, and parallax is a fundamental quantity for 3D reconstruction. A set of pixel displacement vectors is obtained on each pair of adjacent frames for subsequent 3D reconstruction. 3D reconstruction based on motion baseline: The vehicle's travel displacement between adjacent frames is used as the parallax baseline, and the pixel displacement is triangulated to recover the depth of the ground points. The depth estimation of a single pixel can be expressed as: Parameter description: Z: Estimated line-of-sight depth from the viewpoint to the camera unit; F: Equivalent focal length of the camera unit (in pixels); B: Body coordinate displacement of the camera unit in the vehicle's direction of motion between adjacent frames (obtained by wheel speed accumulation and attitude integration); d: Pixel parallax of the viewpoint in adjacent frames (lateral or longitudinal displacement components are selected according to the installation angle). The motion baseline B determined by wheel speed and attitude data can replace external baseline measurements, thereby achieving motion-based depth recovery at a single camera end; the formula inversely calculates depth through pixel parallax, providing a basic quantity for subsequent height calculation. Derivation of ground height from depth and performance of height gradient analysis (key calculation): Projecting the 3D point cloud onto a ground reference plane, calculating the height difference between adjacent points along the travel direction or lateral direction, and approximating the height gradient with a finite difference: Parameter description: G: Local height gradient; ΔX: Estimated ground height of the p-th reconstructed point; ΔX: Estimated distance between adjacent reconstructed points in real space (ground projection). Height gradients can highlight local abrupt changes, facilitating the detection of significant height variations that may affect the contact surface. Preparation for initial and secondary screening of height abrupt changes: The height gradient G is compared with a preset gradient threshold to mark initial screening abrupt changes; for each initial screening point, its pixel position, corresponding depth Z, neighborhood gradient, and timestamp are recorded for use in secondary screening. Initial screening quickly filters out most smooth areas, focusing calculations and judgments on potential obstacle locations, reducing subsequent misjudgments and improving computational efficiency.

[0031] Specifically, the process of determining the obstacle location, obstacle type, and obstacle crossing cost for the corresponding road segment by combining the lateral sway amplitude and speed decay amount in the vehicle condition parameter set, and generating road condition description data is as follows: The height abrupt change points obtained from the secondary screening are subjected to spatiotemporal correlation analysis with the vehicle condition parameter set to establish a mapping relationship between vehicle dynamics response and road surface geometric changes. By analyzing the changing trends of lateral sway amplitude and speed decay amount relative to the baseline value, abnormal vehicle response segments are identified, and the corresponding height abrupt change points within these segments are confirmed as valid obstacle locations. Based on the three-dimensional geometric features of the height abrupt change points at the valid obstacle locations and their spatial distribution patterns in consecutive frames, combined with the change amplitude and duration of speed decay amount, the cumulative impact of obstacles on driving resistance is assessed, and obstacle types are identified. Based on the obstacle type identification results and geometric feature parameters, the required passage safety boundary and expected kinetic energy loss for different types of obstacles are calculated, quantifying the comprehensive crossing cost of each obstacle. The obstacle location coordinates, obstacle type identifiers, and quantified obstacle crossing costs are structurally integrated to generate road condition description data.

[0032] In this implementation scheme, spatiotemporal registration and response segment identification are performed as follows: Height abrupt changes after secondary screening are mapped to the time axis of the vehicle condition parameter set according to their timestamps. Vehicle condition response windows adjacent to the time of each height abrupt change are retrieved, and the deviation of lateral sway amplitude and speed decay relative to the baseline within these windows is analyzed. Real, physically significant obstacles are often accompanied by vehicle dynamic responses; spatiotemporal registration couples visual measurements with dynamic responses, improving the reliability of the judgment. A time window is constructed for each height abrupt change, and the deviation of the response indicators is calculated. Abnormal response segment determination: Response skewness is defined as a combined indicator for determining abnormal response segments. Parameter description: Re: Response skewness; : The average lateral swing amplitude observed within the corresponding time window; The baseline value of lateral sway observed by the vehicle in the smooth reference section; : The average short-term velocity decay observed within the corresponding time window; : Baseline value of velocity decay; Lateral response weighting coefficient; : Speed ​​response weighting coefficient. If If the response exceeds the set threshold, it is marked as an abnormal response segment; weighting coefficient The selection of weights is described below, and they can adjust the relative influence of lateral and longitudinal responses on the judgment. Weighting coefficient selection: Vehicle response samples under various obstacles (such as bumps, depressions, and lateral gravel) are collected through calibration experiments, and determined using linear regression or least squares fitting. Make The response threshold is most strongly correlated with the severity of manually labeled obstacles; response threshold selection: calculated based on training samples. The statistical distribution is used, and an empirical quantile (e.g., the 90th digit) is selected as the initial judgment threshold, which is then fine-tuned during field operation based on the false alarm / false alarm ratio. This allows for a reproducible engineering acquisition process for the threshold and weights, facilitating review and implementation. Obstacle location confirmation and geometric feature extraction: After determining an abnormal response segment, the set of height abrupt change points within that time window is mapped to the vehicle trajectory coordinate system and aggregated to determine the spatial center location and geometric boundary of the obstacle; for each obstacle, geometric features such as maximum height are extracted. Average slope Lateral extension length These geometric parameters provide direct quantitative basis for subsequent type determination and cost calculation. Obstacle type identification: A rule-based determination is adopted based on geometric features and spatial distribution: if... Exceeding a certain height threshold and Small, classified as "protruding"; if the surrounding height is low and Larger obstacles are classified as "depression type"; if the height change extends in a band and has a high lateral proportion, it is classified as "lateral interference type". These classification rules can be calibrated using experimental samples to ensure engineering reproducibility. Crossing cost quantification: For each confirmed obstacle, a comprehensive crossing cost is calculated as the core quantity of the road condition description data, which can be approximated using a linear weighted form as follows: Parameter description: : Cost indicators for overcoming obstacles; : The maximum height of the obstacle; : The lateral extension length of the obstacle; The average slope at the obstacle; : Highly influential weight; : Horizontal extension affects weight; Slope influence weight. This linear approximation maps geometric influences to costs that can be used for planning; weight. The series can be determined by applying linear regression through small-scale real-vehicle tests (e.g., measuring real speed loss or handling load on a variety of representative obstacles), thereby enabling... It is statistically relevant to the actual cost of passage. Safety boundary and expected kinetic energy loss estimation: Based on obstacle geometry and vehicle response, a safety boundary for path determination can be calculated. Compared with the expected speed loss index : Parameter description: : Estimated minimum safe lateral clearance width; Estimation of the vehicle's inherent width related to bicycle handling / riding eccentricity; : The influence coefficient of height on safe width; The influence coefficient of the vehicle's inherent width. The expected speed loss can be calculated using observed values. Alternatively, a weighted geometric approximation can be used: Parameter description: : Expected speed loss indicator; A calibration coefficient that converts the crossing cost into a velocity loss. Used by the path generation module to determine if there is a passable space. Used as a speed-related factor in cost calculation during cost fusion. Output of structured road condition description data: spatial coordinates of obstacles, identified obstacle types, and crossing costs. Safety boundary and expected speed loss The fields are encapsulated into structured records and sent to the cost fusion module along with the timestamp. The structured output allows subsequent modules to read it directly and use it for cost chain construction and path filtering.

[0033] Specifically, based on road condition description data, the tilt angle and short-term speed attenuation in the vehicle condition parameter set are mapped to energy consumption factors. The process of constructing a navigation cost sequence for continuous road segments through linear weight accumulation is as follows: Obstacle location information and crossing costs are extracted from the road condition description data, and the tilt angle and short-term speed attenuation in the vehicle condition parameter set are retrieved within the same time window. Based on the changes in gravity components caused by the tilt angle and the changes in driving resistance reflected by the speed attenuation, an energy consumption factor reflecting the energy demand of a single segment is established. The energy consumption factor within each time window is linearly accumulated in the continuous order of road segments, and the accumulated value is dynamically corrected in combination with the obstacle crossing cost, so that the cost can simultaneously reflect changes in slope, changes in resistance, and difficulty of obstacle passage, resulting in an initial sequence of navigation costs covering the entire target road segment. After the initial sequence of costs is generated, the time consistency of the cost change rate is checked, and abnormal jump points caused by sensing errors are eliminated.

[0034] In this implementation plan, in this step, the obstacle location and crossing cost are first extracted from the road condition description data; simultaneously, the tilt angle within the corresponding time window is obtained from the vehicle condition parameter set. With short-time speed decay .in, The coordinates of the obstacle's position in three-dimensional space; Additional energy expenditure caused by obstacle crossing; The vehicle at the time point The instantaneous tilt angle; D: speed decay, reflecting changes in local resistance. Based on physical energy analysis, the energy consumption of a single segment of travel... Represented as ;in, Gravity influence weighting coefficient, used to quantify the contribution of tilt angle to kinetic energy loss; : Velocity decay weighting coefficient, used to convert drag changes into an energy index; The obstacle weighting coefficient is determined based on the obstacle type and location; each term in the formula reflects the energy demand for a single segment of travel, comprehensively considering changes in gradient, resistance, and obstacle difficulty. Subsequently, the energy consumption factors for all time windows are calculated. The initial navigation cost sequence is obtained by linearly accumulating the values ​​in the order of consecutive road segments. During the accumulation process, if a section has dense obstacles or a sudden change in slope, the weights are dynamically adjusted. Adjustments were made to ensure that the cost value reflected the actual driving difficulty. Finally, time consistency checks were performed on points with excessive jumps in the sequence to remove abnormal spikes caused by sensing errors.

[0035] Specifically, the process of locally smoothing cost abrupt changes in the cost value sequence to form a navigation cost chain is as follows: Gradient changes in the navigation cost value sequence are detected, and segments where the cost value change rate exceeds a set multiple of the average change rate are identified as abrupt changes to be processed; an adaptive smoothing window is established within the abrupt change segments, and the cost value is fitted to the trend using a local regression algorithm to suppress abnormal spikes caused by instantaneous measurement errors, while maintaining the cost value characteristics of densely obstacle-prone areas; the smoothed cost value sequence is continuously optimized, and interpolation is used to ensure smooth transitions in cost value between adjacent segments, forming a navigation cost chain with spatial consistency.

[0036] In this implementation plan, the specific process of locally smoothing cost mutation segments in the cost value sequence to form the navigation cost chain is as follows: This step mainly targets the initial navigation cost sequence. Local mutation segments are smoothed to ensure the continuity and operability of the path search. First, the cost gradient is calculated. Gradient identification Exceeding the set multiple of the average rate of change The segment is designated as the mutation segment, in which It can be determined based on statistical analysis; formula parameters: : Cost changes at two consecutive points; Set a threshold factor. For each mutation segment, establish an adaptive smoothing window. The local regression algorithm was used to fit the trend of cost change: ;in, The cost after smoothing; The local regression weight coefficients are dynamically adjusted based on the point's position within the window, giving greater weight to data closer to the window center. This approach suppresses spikes caused by instantaneous measurement errors while preserving the cost characteristics of obstacle-dense areas. Subsequently, interpolation optimization is performed on the entire sequence to ensure that the smoothed navigation cost chain is spatially continuous and consistent, facilitating subsequent path planning.

[0037] Specifically, the path generation module includes the following steps: Under the constraints of the navigation cost chain, starting from the starting position, a path search is performed on the feasible road segments, extending segment by segment. The endpoints of the current candidate path are mapped to the corresponding positions in the navigation cost chain, and the corresponding cost value is read as the path extension condition. During each path extension process, the unit travel cost increment of the candidate path segment is calculated based on the length, direction change, and local cost value in the navigation cost chain, and the cost increment is added to the current total path cost. When the unit travel cost increment of the candidate path segment exceeds the preset cost upper limit, the extension of the path branch is terminated. Among all candidate path branches, the path that meets the handling stability threshold and has the lowest total cost is selected. The path is parsed into a continuous path point sequence in chronological order, and a steering command sequence that can be directly used by the on-board execution unit is generated based on the direction change between the path points.

[0038] In this implementation scheme, the specific process of the path generation module is as follows: In this step, the path generation module performs a segment-by-segment extension path search based on the smoothed navigation cost chain. First, the starting position is mapped to the navigation cost chain, and the corresponding cost value is read as the path extension condition. For each candidate path segment, the unit travel cost increment is calculated. : ;in, : Path segment length; The change in direction of the path segment reflects the turning radius of the vehicle; : Navigation cost corresponding to the path segment; function The path length, direction change, and navigation cost are combined into a unit travel cost to assess path feasibility. During path extension, when... Exceeding the preset limit If the branch extension fails, terminate the extension to avoid traversing difficult or unstable sections. After extending all candidate paths, select the one with the lowest total cost that satisfies the handling stability threshold. The path is parsed into a continuous sequence of path points. It generates a sequence of steering commands that can be directly used by the onboard execution unit based on changes in the direction of the path points, thereby achieving safe and stable navigation control.

[0039] Please see Figure 2A navigation method based on intelligent bicycles includes the following steps: S1. Synchronously acquiring three-axis motion data from the onboard inertial measurement unit and wheel speed pulse sequences from the wheel encoder, and calculating the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle based on the temporal correspondence between the three-axis acceleration data and the wheel speed pulse sequence, generating a set of vehicle condition parameters for the controllability of the current road segment; S2. Synchronously calling continuous road surface image frames captured by the onboard low-angle camera unit within the time window of the vehicle condition parameter set, calculating the local surface contour change through inter-frame pixel displacement, performing secondary screening of contact plane height abrupt change points, and determining the lateral sway amplitude and speed decay in the vehicle condition parameter set. S3. Based on the road condition description data, the tilt angle and short-term speed decay in the vehicle condition parameter set are mapped to energy consumption factors. A navigation cost sequence for continuous road segments is constructed by accumulating linear weights, and local smoothing is performed on cost abrupt segments in the cost sequence to form a navigation cost chain. S4. Under the constraints of the navigation cost chain, a path search is performed segment by segment. The unit travel cost increment is calculated for each candidate path segment, and the path extension is blocked when the cost increment exceeds the preset upper limit. An effective navigation path that meets the bicycle handling stability threshold is determined, and a sequence of steering instructions for continuous path points is output.

[0040] In this implementation scheme, S1: In this step, triaxial acceleration and angular velocity data from the onboard inertial measurement unit and wheel speed pulse sequences from the wheel encoder are synchronously acquired to form a synchronous sensing data stream with a unified timestamp. Subsequently, based on the temporal correspondence between the triaxial acceleration and the wheel speed pulse sequence, the real-time attitude angle of the bicycle is calculated using fusion algorithms such as complementary filtering, and short-term speed decay and lateral sway amplitude are calculated to generate a multi-dimensional vehicle condition parameter set. This step can reflect the controllability status of the vehicle in a specific road segment in real time, providing basic data for subsequent road condition analysis and navigation decisions. S2: In this step, using the time window of the vehicle condition parameter set, the onboard low-angle camera unit is called to capture continuous road surface image frames. The local three-dimensional contour of the ground surface is reconstructed through inter-frame pixel displacement, and secondary screening is performed on contact plane height abrupt change points. Combining the lateral sway amplitude and speed decay, the obstacle location, obstacle type, and crossing cost are determined, generating structured road condition description data. This step can correlate the vehicle's dynamic response with road surface geometric features, achieving accurate quantification of obstacles on the driving path and providing reliable input for navigation cost assessment. S3: Based on road condition description data, the tilt angle and short-term speed decay in the vehicle condition parameter set are mapped to energy consumption factors, and a navigation cost sequence for continuous road segments is formed through linear weight accumulation. After identifying abrupt change segments in the cost sequence, local smoothing is used to eliminate peaks caused by measurement errors, while maintaining the cost characteristics of obstacle-dense areas, ultimately forming a spatially continuous navigation cost chain that reflects the actual driving difficulty. This step achieves a quantitative mapping between vehicle dynamic characteristics and driving energy demand, providing a measurable cost basis for path planning. S4: Under the constraints of the navigation cost chain, a path search is performed segment by segment for feasible road segments starting from the starting point. Each candidate path is evaluated based on the unit travel cost increment, and the extension terminates when the increment exceeds a preset upper limit. Finally, the path that meets the handling stability threshold and has the optimal total cost is selected, parsed into a continuous path point sequence, and a steering command sequence that can be directly used by the onboard execution unit is generated based on the changes in the direction of the path points. This step tightly integrates navigation cost with path search, achieving safe, smooth, and executable path planning control.

[0041] In summary, this application has at least the following effects:

[0042] A method and system based on intelligent bicycle navigation is proposed. By real-time acquisition of data from the onboard inertial measurement unit and wheel encoders, the system accurately calculates the bicycle's attitude angle, speed decay, and lateral sway amplitude, enabling dynamic monitoring of the vehicle's controllability. By combining road surface images and vehicle condition parameters captured by a low-angle camera unit, the system accurately identifies obstacle locations, types, and crossing costs, forming structured road condition description data. Furthermore, the vehicle condition parameters are mapped to energy consumption factors, constructing a continuous and smooth navigation cost chain to effectively quantify gradient changes, driving resistance, and obstacle crossing difficulty. Based on this, a segmented path search is performed to select the path that meets the handling stability threshold and has the optimal total cost, generating continuous path points and steering command sequences to achieve safe and efficient intelligent bicycle navigation control.

[0043] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0048] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A smart bicycle navigation system, characterized in that, Includes the following modules: The vehicle condition acquisition module is used to synchronously acquire the three-axis motion data of the vehicle-mounted inertial measurement unit and the wheel speed pulse sequence of the wheel encoder, and calculate the real-time attitude angle, short-term speed decay and lateral sway amplitude of the bicycle based on the time sequence correspondence between the three-axis acceleration data and the wheel speed pulse sequence, and generate a set of vehicle condition parameters for the controllability of the current road section. The road condition analysis module is used to synchronously call the continuous road surface image frames captured by the vehicle-mounted low-angle camera unit within the time window of the vehicle condition parameter set. It calculates the local surface contour change through inter-frame pixel displacement, performs secondary screening of contact plane height change points, and combines the lateral sway amplitude and speed attenuation in the vehicle condition parameter set to determine the obstacle location, obstacle type and obstacle crossing cost of the corresponding road segment, and generates road condition description data. The cost fusion module is used to map the tilt angle and short-term speed attenuation in the vehicle condition parameter set into energy consumption factors based on road condition description data. It constructs a navigation cost sequence for continuous road segments through linear weight accumulation and performs local smoothing on cost abrupt segments in the cost sequence to form a navigation cost chain. The path generation module is used to perform a segment-by-segment path search under the constraints of the navigation cost chain. It calculates the unit travel cost increment for each candidate path segment and blocks path extension when the cost increment exceeds a preset upper limit. It determines the effective navigation path that meets the bicycle handling stability threshold and outputs a sequence of turning instructions for consecutive path points.

2. The intelligent bicycle navigation system according to claim 1, characterized in that: The specific process of synchronously acquiring the three-axis motion data of the on-board inertial measurement unit and the wheel speed pulse sequence of the wheel encoder is as follows: A time synchronization mechanism between the inertial measurement unit and the wheel encoder is established. The sampling clock of the two sensors is synchronized through a hardware interrupt signal. The raw acceleration and angular velocity data output by the inertial measurement unit are normalized to the coordinate system and transformed from the sensor coordinate system to the bicycle body coordinate system. Simultaneously, the pulse signal of the wheel encoder is periodically measured and pulses are counted, and the linear velocity and displacement of the bicycle are calculated in real time based on the known wheel circumference. The converted triaxial motion data and the calculated wheel speed pulse sequence are packaged together with a unified timestamp to form a synchronized sensor data stream.

3. The intelligent bicycle navigation system according to claim 2, characterized in that: The specific process of calculating the real-time attitude angle, short-term speed decay, and lateral sway amplitude of the bicycle based on the temporal correspondence between triaxial acceleration data and wheel speed pulse sequences, and generating a set of vehicle condition parameters for the current road segment's controllability state, is as follows: Based on the synchronous sensor data stream, the accelerometer and gyroscope data are fused through a complementary filtering algorithm to calculate the real-time attitude angle of the bicycle during the process of moving, analyze the changing trend of the wheel speed pulse sequence within the sliding time window, calculate the differential characteristics of the speed, and obtain the short-time speed decay amount that characterizes the change of resistance. By monitoring the periodic variation characteristics of lateral acceleration in the bicycle's body coordinate system and combining it with attitude angle data, the lateral sway amplitude parameter is quantified. The real-time attitude angle, speed decay, and lateral sway amplitude parameters are normalized to construct a multi-dimensional set of vehicle condition parameters characterizing the handling stability of the bicycle on the current road section.

4. The intelligent bicycle navigation system according to claim 1, characterized in that: Within the time window of the vehicle condition parameter set, continuous road surface image frames captured by the vehicle-mounted low-angle camera unit are synchronously invoked. The local surface contour changes are calculated through inter-frame pixel displacement, and secondary screening is performed on contact plane height abrupt change points. The specific process is as follows: Based on the changing trend of the concentrated speed decay of vehicle condition parameters, the sampling frequency and exposure time of the low-angle camera unit are dynamically adjusted, feature points are extracted and matched on the captured continuous road image frames, and the pixel displacement vector between adjacent frames is calculated by optical flow method. Based on the intrinsic and extrinsic parameters of the camera unit and the pixel displacement vector, the three-dimensional contour of the local surface is reconstructed through the principle of triangulation. The reconstructed three-dimensional contour data is subjected to height gradient analysis to identify areas where the height change exceeds a set threshold and mark them as contact plane height abrupt change points. By combining the distribution of height abrupt change points in multiple frames of images, cluster analysis is used to eliminate noise interference and complete the secondary screening of height abrupt change points.

5. A smart bicycle navigation system according to claim 4, characterized in that: The specific process of generating road condition description data by combining the lateral sway amplitude and speed decay of vehicle condition parameters to determine the obstacle location, obstacle type, and obstacle crossing cost for the corresponding road segment is as follows: The height mutation points obtained from the secondary screening are subjected to spatiotemporal correlation analysis with the vehicle condition parameter set to establish a mapping relationship between vehicle dynamic response and road surface geometric changes. By analyzing the changing trends of lateral sway amplitude and speed attenuation relative to the baseline value, abnormal response sections of the vehicle body are identified, and the corresponding height mutation points in these sections are confirmed as effective obstacle locations. Based on the three-dimensional geometric features of height abrupt change points at effective obstacle locations and their spatial distribution patterns in consecutive frames, combined with the magnitude and duration of speed decay, the cumulative impact of obstacles on driving resistance is assessed, and obstacle types are identified. Based on the obstacle type identification results and geometric feature parameters, the required safety boundary and expected kinetic energy loss for different types of obstacles are calculated respectively, and the comprehensive crossing cost of each obstacle is quantified. The obstacle location coordinates, obstacle type identifiers, and quantified obstacle crossing costs are structurally integrated to generate road condition description data.

6. A smart bicycle navigation system according to claim 1, characterized in that: Based on road condition description data, the tilt angle and short-term speed attenuation in the vehicle condition parameter set are mapped to energy consumption factors. The specific process of constructing a navigation cost sequence for continuous road segments through linear weight accumulation is as follows: Obstacle location information and crossing cost are extracted from road condition description data. Within the same time window, the tilt angle and short-term speed reduction are retrieved from the vehicle condition parameter set. Based on the changes in gravity component caused by tilt angle and the changes in driving resistance reflected by speed reduction, an energy consumption factor reflecting the energy demand of a single segment of driving is established. The energy consumption factor within each time window is linearly accumulated in the continuous sequence of road segments. The accumulated value is dynamically corrected in combination with the obstacle crossing cost, so that the cost value can simultaneously reflect the slope change, resistance change and obstacle crossing difficulty, and obtain the initial sequence of navigation cost value covering the entire target road segment. After the initial sequence of the substitution value is generated, the time consistency of the substitution value change rate is checked to eliminate abnormal jump points caused by sensing errors.

7. A smart bicycle navigation system according to claim 6, characterized in that: The specific process of locally smoothing cost mutation segments in the cost value sequence to form a navigation cost chain is as follows: Detect gradient changes in the navigation cost value sequence and identify segments whose cost value change rate exceeds a set multiple of the average change rate as abrupt segments to be processed. An adaptive smoothing window is established within the abrupt change zone. The cost value is fitted to a trend using a local regression algorithm to suppress abnormal spikes caused by instantaneous measurement errors, while maintaining the cost value characteristics of densely obstacle-prone areas. The smoothed cost sequence is continuously optimized, and interpolation is used to ensure a smooth transition of cost between adjacent segments, forming a navigation cost chain with spatial consistency.

8. A smart bicycle navigation system according to claim 1, characterized in that: The path generation module includes the following steps: Under the constraints of the navigation cost chain, a path search is performed segment by segment on feasible road segments starting from the starting position. The endpoints of the current candidate path are mapped to the corresponding positions in the navigation cost chain, and the corresponding cost value is read as the path extension condition. During each path extension process, the unit travel cost increment of the candidate path segment is calculated based on the length, direction change, and local cost value in the navigation cost chain, and the cost increment is added to the current total path cost. When the unit travel cost increment of a candidate path segment exceeds the preset cost limit, the extension of that path branch is terminated. The path that meets the handling stability threshold and has the lowest total cost among all candidate path branches is selected. This path is then parsed into a continuous sequence of path points in chronological order, and a steering command sequence that can be directly used by the on-board execution unit is generated based on the directional changes between the path points.

9. A method for intelligent bicycle navigation, applied to an intelligent bicycle navigation system as described in any one of claims 1-8, characterized in that, Includes the following steps: S1. Synchronously acquire the three-axis motion data of the vehicle-mounted inertial measurement unit and the wheel speed pulse sequence of the wheel encoder, and calculate the real-time attitude angle, short-term speed decay and lateral sway amplitude of the bicycle according to the time sequence correspondence between the three-axis acceleration data and the wheel speed pulse sequence, and generate a set of vehicle condition parameters for the controllability of the current road section. S2. Within the time window of the vehicle condition parameter set, continuously captured road surface image frames by the vehicle-mounted low-angle camera unit are synchronously called. The local surface contour change is calculated by inter-frame pixel displacement. The contact plane height change point is screened for a second time. Combined with the lateral swing amplitude and speed attenuation in the vehicle condition parameter set, the obstacle location, obstacle type and obstacle crossing cost of the corresponding road segment are determined, and road condition description data is generated. S3. Based on the road condition description data, the tilt angle and short-term speed attenuation in the vehicle condition parameter set are mapped to energy consumption factors. A navigation cost sequence for continuous road segments is constructed through linear weight accumulation, and local smoothing is performed on cost abrupt segments in the cost sequence to form a navigation cost chain. S4. Perform a segment-by-segment path search under the constraints of the navigation cost chain, calculate the unit travel cost increment for each candidate path segment, and block path extension when the cost increment exceeds the preset upper limit, determine the effective navigation path that meets the bicycle handling stability threshold, and output the turning instruction sequence of continuous path points.

Images