A blind guiding path screening method based on intelligent sensors

Abstract

The application discloses a blind guiding path screening method based on intelligent sensors and belongs to the technical field of intelligent auxiliary travel. The application comprises a sensing data dynamic fusion noise reduction module based on residual correction and a hierarchical risk threshold adaptive iterative path screening module. Firstly, the application synchronously collects environment data through multiple sensors, and completes data preprocessing relying on criteria; then, the application adaptively adjusts the sensing data fusion weight of each sensor through a self-developed residual correction dynamic weight algorithm, effectively eliminates noise, corrects detection deviation, and outputs high-precision environment fusion data; on the basis, the application obtains three types of risk quantization indexes by disassembling the fusion data, adaptively solves the risk weight in combination with the visual barrier traffic safety priority, completes path risk hierarchical rating, constructs a comprehensive cost function of "risk priority, distance as auxiliary" to screen the optimal path, and sets a 200ms global iteration period to realize dynamic real-time correction of sensing data, risk level and blind guiding path.

Description

Technical Field

[0001] This invention relates to the fields of intelligent sensing and detection, path planning and barrier-free assisted travel, specifically a method for selecting guide routes for the visually impaired based on intelligent sensors. Background Technology

[0002] With the development of intelligent assisted travel technologies, sensor-based intelligent guide devices are gradually replacing traditional guide canes and guide dogs, becoming the core assistive tool for visually impaired users to travel independently. The accuracy and safety of route selection are the key factors determining the value of these guide devices. Currently, existing guide route selection technologies have the following problems: Existing multi-source sensor data fusion technologies are highly versatile but poorly adaptable. Traditional fixed-weight weighted average fusion algorithms and Kalman filtering algorithms have significant shortcomings in complex urban road conditions. Guided vehicle devices are equipped with lidar, ultrasonic sensors, and visual sensors responsible for collecting obstacle distance, road surface smoothness, and pedestrian / vehicle flow density data, respectively. Traditional algorithms assign fixed weights to all sensor data, failing to consider the differences in sensor detection accuracy under different road conditions. Furthermore, they cannot eliminate abnormal data caused by instantaneous environmental noise (such as light interference, airflow disturbances, and temporary obstructions), resulting in biased environmental perception data with a low signal-to-noise ratio. This fails to accurately reflect real-time road conditions and provides erroneous data support for subsequent route selection.

[0003] Existing path selection algorithms lack specificity; traditional Dijkstra's algorithm and A... General path planning algorithms, which use "shortest path and least travel time" as the sole selection objective, fail to adapt to the travel characteristics of visually impaired users. Visually impaired users cannot visually predict dynamic risks and have extremely low tolerance for road bumps, narrow passages, dense crowds, and rapidly moving obstacles. Traditional algorithms cannot stratify and quantify various travel risks; they can only select physically feasible paths and cannot avoid high-risk sections, easily leading guide devices to guide users into dangerous areas, posing significant safety hazards. Furthermore, traditional algorithms are static plans and cannot iteratively update path risk levels based on real-time sensor data, resulting in extremely poor dynamic adaptability.

[0004] In existing technologies, most solutions only perform simple optimizations on single sensor data or path algorithms, failing to achieve coordinated optimization of sensor data processing and path risk screening. This fails to fundamentally address the safety and accuracy issues of guide path selection in complex urban scenarios. Therefore, this invention proposes a guide path selection method based on intelligent sensors. Summary of the Invention

[0005] The purpose of this invention is to provide a method for selecting guideways for the visually impaired based on intelligent sensors, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for selecting guideways for the visually impaired based on intelligent sensors, comprising the following steps: S1. Real-time collection of travel environment data through intelligent sensor array, including obstacle distance data collected by lidar, road surface smoothness data collected by ultrasonic sensors, and pedestrian and vehicle flow density data collected by visual sensors, to complete multi-source data preprocessing; S2. The multi-source data preprocessed in step S1 is denoised and dynamically weighted by the sensor data dynamic fusion and denoising module based on residual correction, and high-precision unified environmental perception data is output. S3. The environmental perception data output in step S2 is used to perform risk stratification and iterative filtering algorithm of the hierarchical risk threshold adaptive iterative path filtering module to quantify the risk of the pre-selected access path, and the optimal safe blind path is selected by combining iterative optimization algorithm. S4. Update the sensor data in step S1 in real time, dynamically iterate the path risk level, and complete the real-time path correction output.

[0007] Preferably, the specific implementation steps of S2 are as follows: S21. Multi-source sensor data preprocessing: Outlier removal is performed on the multi-source data collected by the intelligent sensor group. The criteria remove abrupt noise points from single-frame data, eliminate invalid data caused by sensor jitter and environmental interference, and retain valid detection data sequences; specifically, the first... The effective data sequence of the sensor is ,in These correspond to LiDAR, ultrasonic sensors, and vision sensors, respectively. This refers to the number of sampling points in a single frame of data. S22. Data Reference Error Calculation: Calculate the real-time detection error of various sensor data, using the deviation between two adjacent frames of data as the reference error. The formula is as follows: ,in, For the first Type 1 sensor The baseline error of the frame data; For the first Valid frame detection data; This is the valid detection data from the previous frame. The smaller the baseline error, the more stable the current sensor detection state, and the higher the data reliability. S23. Residual Correction Dynamic Weight Calculation: A global residual factor is introduced to correct the weight allocation, abandoning the traditional fixed weight mode. The weight calculation formula is as follows: ,in For the first Type 1 sensor The dynamic weights of frame data satisfy ; For the first The historical residual mean of a sensor type characterizes the long-term detection bias of the sensor. This is the residual correction factor, and its value range is... The weighting is used to balance the influence of instantaneous error and historical residual; the preferred weighting in this invention is 0.35. S24. Weighted Fusion and Secondary Noise Reduction: Multi-source data fusion is performed based on dynamic weights. The fusion formula is as follows: ;in For the first The frames are ultimately fused with environmental data; For the preprocessed first The system collects valid sensor data; finally, it performs secondary noise reduction through sliding window mean filtering, outputting real-time environmental perception data with high signal-to-noise ratio and high accuracy, which is then transmitted to the hierarchical risk threshold adaptive iterative path filtering module in step S3.

[0008] Preferably, the residual correction coefficient in S23 The specific acquisition logic is as follows: S23.1 Calculate the global instantaneous error fluctuation mean, which characterizes the overall real-time detection stability of all sensors in the current frame, conforming to the dynamic road condition interference characteristics. The calculation formula is as follows: ,in, For the first The average instantaneous error fluctuation of the frame is determined by the real-time reference errors of the three types of sensors. The mean is obtained by solving; The larger the value, the stronger the instantaneous noise such as light interference, pedestrian obstruction, and airflow disturbance in the current road conditions, and the greater the fluctuation in real-time detection by the sensor. It is necessary to increase the correction weight of historical residuals to offset the interference of instantaneous abnormal data. S23.2 Calculate the global historical residual mean to characterize the long-term detection accuracy stability of the sensor and avoid the influence of inherent hardware deviations of the sensor. The calculation formula is as follows: ,in For the first The mean of global historical residuals is derived from the mean of historical residuals of the three types of sensors. The mean is obtained by solving; The larger the value, the greater the cumulative deviation of the sensor over a long period of time. It is necessary to weaken the weight of historical residuals to avoid the continuous accumulation of inherent errors that lead to distortion of fused data. S23.3 Adaptive solution for initial residual correction coefficient: Based on the balance between instantaneous error and historical residual, a proportional solution formula is constructed: ;in This represents the initial residual correction coefficient for the unconstrained input. The proportion-based solution formula uses the dynamic proportion of global historical residuals and global instantaneous errors as the basis for coefficient values, enabling parameters to adapt to changes in road conditions and sensor operating states. When instantaneous noise is strong, When it is too large, Synchronous increase enhances residual correction capability and suppresses instantaneous noise interference; when the sensor has a large long-term deviation, When it is too large, Adaptively reduce and weaken the cumulative effect of historical errors to ensure the accuracy of weight allocation; S23.4 Threshold constraint pruning, completing the final coefficient assignment: To avoid algorithm oscillations caused by parameter mutations under extreme road conditions, the optimal stable range is calibrated based on a large number of experiments. The initial coefficients are constrained and corrected using the following formula: .

[0009] Preferably, the specific implementation steps of S3 are as follows: S31. Quantitative Modeling of Multidimensional Risk Indicators: Based on the fused data output by the residual-corrected sensor data dynamic fusion and noise reduction module, three types of core access risk indicators are extracted and quantitatively assigned according to the access characteristics of visually impaired users. (1) Obstacle risk value Extracted from LiDAR fusion data, quantified based on obstacle distance and volume; the closer the obstacle, the larger its volume. The higher the value; (2) Road surface risk value Extracted from data fusion by ultrasonic sensors, and quantified based on road surface smoothness and the height of bumps and depressions; the worse the road surface smoothness, the more... The higher the value; (3) Risk value of pedestrian flow It is extracted from data fusion by visual sensors, and quantified according to crowd density and movement speed. The denser the crowd and the faster the movement speed, the better. The higher the value; S32. Risk Stratification Threshold Classification: Using a risk stratification mechanism, the overall risk of all paths is divided into three levels: low, medium, and high. Fixed risk threshold ranges are preset, combined with the tolerance of visually impaired users. Low-risk level: Overall risk value It is safe to pass; medium risk level: Exercise caution and monitor in real time; High-risk level: Passage is prohibited; The path comprehensive risk value is calculated using an improved weighted quantification formula: ;in , , For risk weighting coefficients, satisfying ; S33. Adaptive Iterative Path Selection: An iterative selection mechanism is adopted, prioritizing risk and supplementing it with distance. The iterative formula is as follows: ,in The path is optimized by comprehensively considering the cost; the smaller the cost, the better the path. This represents the current path length. The maximum length of all alternative paths; This is the distance balancing coefficient, with a value of 0.2. It is used to balance the weight of risk and path length, avoiding excessive pursuit of low risk that leads to excessively long paths. Iterative filtering logic: Traverse all pre-selected paths and input the fused data updated by the sensor data dynamic fusion and noise reduction module based on residual correction in real time to calculate the paths. The cost of the remaining paths is then updated iteratively to eliminate high-risk paths, and finally the path with the lowest overall cost is selected as the optimal guide path.

[0010] Preferably, the specific implementation details of the multidimensional risk indicator quantitative modeling in S31 are as follows: Quantitative modeling of multidimensional risk indicators of fused data from a dynamic fusion and denoising module based on residual correction: The final output of the dynamic fusion and denoising module based on residual correction of sensor data is the fused environmental data. The result is a weighted fusion of three types of sensor data, containing complete and accurate environmental information on current road conditions, obstacles, road surface, and pedestrian flow; firstly, the... By decomposing the components and separating the precise fusion components corresponding to the three types of sensors, the core risk indicators for visually impaired access are accurately quantified, completely eliminating the original sensor noise data. The decomposition and quantification formula is as follows: ;in , , These are the single-class precise fusion components of the LiDAR, ultrasonic sensor, and visual sensor data fused by the residual correction-based dynamic fusion and noise reduction module, respectively, satisfying the requirements of... ; , , The sensor weights are dynamically and adaptively solved by the sensor data dynamic fusion and noise reduction module based on residual correction. , , The effective sensor data is preprocessed by the residual correction-based dynamic fusion and noise reduction module; based on the above-mentioned precise fusion components, three types of risk indicators are quantified: (1) Obstacle risk value To meet the safety needs of visually impaired users, a lidar component with precise fusion output is generated based on a residual correction-based sensor data dynamic fusion and noise reduction module. The results were obtained through global normalization quantitative calculation. The system integrates two core sensor data points: real-time obstacle distance and obstacle size. The specific normalized calculation formula is as follows: ;in , These are the preset maximum and minimum effective detection thresholds for the lidar sensor, corresponding to a maximum detection distance of 10m and a minimum detection distance of 0m for the device. For the first The frame is the LiDAR fusion component value output by the sensor data dynamic fusion and noise reduction module based on residual correction. The computational logic aligns with the characteristics of guided passage for the visually impaired: the closer the obstacle, the larger its size, the more accurate the calculation. The smaller the value, the better the calculated result. The closer the risk value is to 1, the better; when there are no obstacles in the passage area, Take the maximum value. Approaching zero, it accurately achieves quantitative mapping of the degree of danger of obstacles, solving the problem of traditional technologies that only qualitatively judge obstacle risks and have poor accuracy; (2) Road surface risk value The ultrasonic sensor precisely fused components output by the residual correction-based sensor data dynamic fusion and noise reduction module. Normalized quantitative calculations yielded the following results. The specific calculation formula for accurately characterizing the height of road surface bumps and the degree of surface damage is as follows: ;in , These are the maximum and minimum road surface bump detection thresholds preset for the ultrasonic sensor, corresponding to the maximum road surface bump height that visually impaired users can tolerate and the absolute flat road surface reference value. For the first The frame is the ultrasonic fusion component value output by the residual-corrected sensor data dynamic fusion and noise reduction module; the calculation logic is: the more severe the road surface deformation and damage, the higher the value of the ultrasonic fusion component. The larger the value, the better the result after normalization. The higher the risk value, the better, especially when the road surface is smooth and undamaged. Approaching zero, it can accurately quantify the risks of road bumps and tripping, adapting to the travel characteristics of visually impaired users who cannot visually identify road defects.

[0011] (3) Risk value of pedestrian flow The visual sensor's precisely fused component output from the dynamic fusion and noise reduction module based on residual correction of sensor data. Normalized quantitative calculations yielded the following results. By integrating real-time pedestrian density, pedestrian movement speed, and disorder multi-dimensional information, the data undergoes noise reduction and dynamic weight correction via a sensor data dynamic fusion and noise reduction module based on residual correction. The specific calculation formula is as follows: ;in , These are the maximum and minimum pedestrian risk detection thresholds preset by the visual sensor, corresponding to the road saturation pedestrian density and the baseline state of unmanned passage. For the first The frame represents the visual fusion component value output by the sensor data dynamic fusion and noise reduction module based on residual correction. The calculation logic is as follows: the denser the crowd in the scene, the faster the pedestrians move, and the more chaotic the motion trajectory, the better. The larger the value, the more it corresponds to The higher the risk value, the better; when there are no pedestrians on the road, Approaching zero, it can accurately quantify the risk of dynamic pedestrian collisions and congestion.

[0012] Preferably, the risk weighting coefficient in S32 , , The specific acquisition logic is as follows: Step 1, Weighted Basis Mapping: The residual-corrected sensor data dynamic fusion and noise reduction module outputs the dynamic fusion weights of the three types of sensors in real time. , , These represent the accuracy and reliability of data from LiDAR, ultrasonic sensors, and visual sensors under different road conditions. Higher reliability and smaller data errors correspond to larger weight values. Based on the core logic that "the higher the reliability of sensor data, the higher the reference value of the corresponding risk indicator," the dynamic sensor weights are used as the initial basis for the risk weights, and the mapping relationship is as follows: ,in , , As the initial base of uncorrected risk weights, the adaptive algorithm output of the sensor data dynamic fusion and noise reduction module based on residual correction is fully traceable, realizing data-level linkage and solving the technical defect of the disconnect between traditional risk weights and sensor detection accuracy. Step 2, Traffic Safety Priority Correction: Considering the traffic characteristics of visually impaired users, obstacle collision risk is considered a fatal safety risk, road bumps are considered a secondary injury risk, and pedestrian congestion is considered a minor disturbance risk. A fixed safety priority correction factor is introduced. , , The initial base is then weighted and corrected to strengthen the weighting of core security risks. The correction formula is as follows: ;in , , The weight parameters after priority correction are adjusted using differentiated correction factors to meet the core passage safety needs of visually impaired users: "prioritize obstacle avoidance, then smooth terrain, and finally avoid crowds." Step 3: Solve for the final weights using normalization constraints: To satisfy the mathematical constraints of weight normalization and ensure the overall risk value... Stable landing The corrected parameters are normalized within the specified interval to obtain the final dynamic risk weight, calculated using the following formula: , ;in For the first The real-time dynamic risk weight is dynamically updated based on the sensor data accuracy of the noise reduction module and the road condition scenario, according to the sensor data dynamic fusion based on residual correction.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention improves the traditional fixed-weight fusion algorithm by using a sensor data dynamic fusion and noise reduction module based on residual correction. Through dynamic weight adaptive adjustment and residual correction mechanism, it effectively eliminates the data distortion problem caused by environmental noise and sensor detection deviation, improves the signal-to-noise ratio of sensor data fusion, provides accurate and reliable underlying data support for path selection, and solves the core problem of low environmental perception accuracy in existing technologies.

[0014] This invention improves the traditional single-target path planning algorithm by using a hierarchical risk threshold adaptive iterative path selection module. It constructs a multi-dimensional risk hierarchical assessment system adapted to visually impaired users, prioritizing risk in the selection process. This system can accurately avoid high-risk road sections and dynamically update the path based on real-time sensor data. This solves the technical pain points of poor safety and weak dynamic adaptability of traditional guide paths, and significantly improves the travel safety of visually impaired users. Attached Figure Description

[0015] Fig. 1 This is a schematic diagram of the overall process of the present invention; Fig. 2 This is a schematic diagram illustrating the specific workflow of step S2 of the present invention; Fig. 3 This is a schematic diagram illustrating the specific workflow of step S3 of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figs. 1-3 This invention provides a technical solution: a method for selecting guideways for the visually impaired based on intelligent sensors, comprising the following steps: S1. Real-time collection of travel environment data through intelligent sensor arrays, including obstacle distance data collected by lidar, road surface smoothness data collected by ultrasonic sensors, and pedestrian and vehicle flow density data collected by visual sensors, to complete multi-source data preprocessing; the specific implementation steps are as follows: S11. Multi-source sensor synchronous timing acquisition. Set the acquisition frequency to match the iteration cycle, with a single acquisition cycle of 200ms. Simultaneously trigger three types of sensors to complete environmental data sampling, acquiring core traffic environment data for the scene: LiDAR collects real-time raw data on obstacle distance and obstacle size within a 0-10m detection range; ultrasonic sensors collect real-time raw data on road surface unevenness and smoothness deformation; and visual sensors collect real-time raw data on road pedestrian density, pedestrian movement speed, and motion disorder. Define the first... The raw data collected by the three types of sensors at each frame time is , , ,in The frame numbers are 1, 2, and 3, which correspond to the lidar, ultrasonic sensor, and vision sensor, respectively.

[0018] S12. Coarse screening of anomalies in single-frame raw data. This method addresses pulse-like sudden noise caused by light interference, airflow disturbances, sensor jitter, and temporary occlusion in complex urban scenes. Statistical discrimination criteria are used to remove outliers from the raw data of a single frame, eliminating single-point invalid noise data and ensuring the validity of the single-frame data. The calculation formula is as follows: ,in For the first Sensor-like continuous The mean of the original frame data; For the first Sensor-like continuous The standard deviation of the original frame data; This refers to the number of sampling points per cycle, consistent with the number of sampling points in the data sequence of the sensor data dynamic fusion and noise reduction module based on residual correction. The data discrimination rule is: if the original data of a single frame... If the data is determined to be an outlier due to sudden noise, it will be removed directly; if it is within the range, it will be retained as valid original data.

[0019] S13. Time-series data regularization and sequence construction. The valid data retained after coarse screening undergoes time-series sorting and missing data completion, eliminating invalid null values ​​and out-of-order data. A regularized valid data sequence is reconstructed according to the acquisition time order, perfectly matching the input data of the residual-corrected sensor data dynamic fusion and noise reduction module. The final output is the [missing data]. Valid data sequence of sensor-like devices: ;in For the preprocessed first Type 1 sensor This sequence of valid sampled data serves as the input data source for the subsequent benchmark error calculation, residual correction, and dynamic weight fusion of the sensor data dynamic fusion and noise reduction module based on residual correction.

[0020] S14. Preprocessed data output. The normalized effective data sequences of the three types of sensors are synchronously transmitted to the sensor data dynamic fusion and noise reduction module based on residual correction. This provides accurate, noise-free, and time-aligned underlying data support for subsequent multi-source data noise reduction and residual correction dynamic fusion, solving the pre-processing technical problems of fusion distortion and risk misjudgment caused by unprocessed raw data, residual noise, and disordered timing in traditional solutions.

[0021] S2. The multi-source data preprocessed in step S1 is denoised and dynamically weighted using a sensor data dynamic fusion and denoising module based on residual correction to output high-precision unified environmental perception data. The specific implementation steps are as follows: S21. Multi-source sensor data preprocessing: Outlier removal is performed on the multi-source data collected by the intelligent sensor group. The criteria remove abrupt noise points from single-frame data, eliminate invalid data caused by sensor jitter and environmental interference, and retain valid detection data sequences; specifically, the first... The effective data sequence of the sensor is ,in These correspond to LiDAR, ultrasonic sensors, and vision sensors, respectively. This refers to the number of sampling points in a single frame of data. S22. Data Reference Error Calculation: Calculate the real-time detection error of various sensor data, using the deviation between two adjacent frames of data as the reference error. The formula is as follows: ,in, For the first Type 1 sensor The baseline error of the frame data; For the first Valid frame detection data; This is the valid detection data from the previous frame. The smaller the baseline error, the more stable the current sensor detection state, and the higher the data reliability. S23. Residual Correction Dynamic Weight Calculation: A global residual factor is introduced to correct the weight allocation, abandoning the traditional fixed weight mode. The weight calculation formula is as follows: ,in For the first Type 1 sensor The dynamic weights of frame data satisfy ; For the first The historical residual mean of a sensor type characterizes the long-term detection bias of the sensor. This is the residual correction factor, and its value range is... The weighting is used to balance the influence of instantaneous error and historical residual; the preferred weighting in this invention is 0.35. The core improvement of this algorithm lies in correcting the weight deviation of instantaneous errors through residual factors, avoiding weight failure caused by single instantaneous data fluctuations, and realizing real-time adaptive adjustment of weights according to road conditions and sensor status.

[0022] Among them, residual correction coefficient The specific acquisition logic is as follows: S23.1 Calculate the global instantaneous error fluctuation mean, which characterizes the overall real-time detection stability of all sensors in the current frame, conforming to the dynamic road condition interference characteristics. The calculation formula is as follows: ,in, For the first The average instantaneous error fluctuation of the frame is determined by the real-time reference errors of the three types of sensors. The mean is obtained by solving; The larger the value, the stronger the instantaneous noise such as light interference, pedestrian obstruction, and airflow disturbance in the current road conditions, and the greater the fluctuation in real-time detection by the sensor. It is necessary to increase the correction weight of historical residuals to offset the interference of instantaneous abnormal data. S23.2 Calculate the global historical residual mean to characterize the long-term detection accuracy stability of the sensor and avoid the influence of inherent hardware deviations of the sensor. The calculation formula is as follows: ,in For the first The mean of global historical residuals is derived from the mean of historical residuals of the three types of sensors. The mean is obtained by solving; The larger the value, the greater the cumulative deviation of the sensor over a long period of time. It is necessary to weaken the weight of historical residuals to avoid the continuous accumulation of inherent errors that lead to distortion of fused data. S23.3 Adaptive solution for initial residual correction coefficient: Based on the balance between instantaneous error and historical residual, a proportional solution formula is constructed: ;in This represents the initial residual correction coefficient for the unconstrained input. The proportion-based solution formula uses the dynamic proportion of global historical residuals and global instantaneous errors as the basis for coefficient values, enabling parameters to adapt to changes in road conditions and sensor operating states. When instantaneous noise is strong, When it is too large, Synchronous increase enhances residual correction capability and suppresses instantaneous noise interference; when the sensor has a large long-term deviation, When it is too large, Adaptively reduce and weaken the cumulative effect of historical errors to ensure the accuracy of weight allocation; S23.4 Threshold constraint pruning, completing the final coefficient assignment: To avoid algorithm oscillations caused by parameter mutations under extreme road conditions, the optimal stable range is calibrated based on a large number of experiments. The initial coefficients are constrained and corrected using the following formula: ; Extensive testing under typical urban road conditions has verified that, in smooth traffic scenarios, the algorithm adaptively solves... The algorithm converges stably to 0.35, therefore, this invention sets 0.35 as the preferred fixed value while retaining the dynamic adaptive solution logic to adapt to extremely complex road conditions. This parameter acquisition method is compatible with the residual correction dynamic weight fusion algorithm of this invention, solving the technical problems of traditional fixed coefficient algorithms being unable to adapt to dynamic road conditions and having large weight matching deviations from the parameter iteration level, and greatly improving the accuracy and adaptability of multi-source sensor data fusion. S24. Weighted Fusion and Secondary Noise Reduction: Multi-source data fusion is performed based on dynamic weights. The fusion formula is as follows: ;in For the first The frames are ultimately fused with environmental data; For the preprocessed first The sensor-like effective data is then processed; finally, a second noise reduction is achieved through sliding window mean filtering, outputting real-time environmental perception data with high signal-to-noise ratio and high accuracy, which is then transmitted to the hierarchical risk threshold adaptive iterative path filtering module in step S3. S3. The environmental perception data output in step S2 is processed by the hierarchical risk quantification iterative filtering algorithm of the hierarchical risk threshold adaptive iterative path filtering module to perform risk hierarchical quantification on the pre-selected access paths, and the optimal safe guide path is selected by combining iterative optimization algorithm; the specific implementation steps are as follows: S31. Quantitative Modeling of Multidimensional Risk Indicators: Based on the fused data output by the residual-corrected sensor data dynamic fusion and noise reduction module, three types of core access risk indicators are extracted and quantitatively assigned according to the access characteristics of visually impaired users. The final output of the residual-corrected sensor data dynamic fusion and noise reduction module is the fused environmental data. The result is a weighted fusion of three types of sensor data, containing complete and accurate environmental information on current road conditions, obstacles, road surface, and pedestrian flow; firstly, the... By decomposing the components and separating the precise fusion components corresponding to the three types of sensors, the core risk indicators for visually impaired access are accurately quantified, completely eliminating the original sensor noise data. The decomposition and quantification formula is as follows: ;in , , These are the single-class precise fusion components of the LiDAR, ultrasonic sensor, and visual sensor data fused by the residual correction-based dynamic fusion and noise reduction module, respectively, satisfying the requirements of... ; , , The sensor weights are dynamically and adaptively solved by the sensor data dynamic fusion and noise reduction module based on residual correction. , , The effective sensor data is preprocessed by the residual correction-based dynamic fusion and noise reduction module; based on the above-mentioned precise fusion components, three types of risk indicators are quantified: (1) Obstacle risk value To meet the safety needs of visually impaired users, a lidar component with precise fusion output is generated based on a residual correction-based sensor data dynamic fusion and noise reduction module. The results were obtained through global normalization quantitative calculation. The system integrates two core sensor data points: real-time obstacle distance and obstacle size. The specific normalized calculation formula is as follows: ;in , These are the preset maximum and minimum effective detection thresholds for the lidar sensor, corresponding to a maximum detection distance of 10m and a minimum detection distance of 0m for the device. For the first The frame is the LiDAR fusion component value output by the sensor data dynamic fusion and noise reduction module based on residual correction. The computational logic aligns with the characteristics of guided passage for the visually impaired: the closer the obstacle, the larger its size, the more accurate the calculation. The smaller the value, the better the calculated result. The closer the risk value is to 1, the better; when there are no obstacles in the passage area, Take the maximum value. Approaching zero, it accurately achieves quantitative mapping of the degree of danger of obstacles, solving the problem of traditional technologies that only qualitatively judge obstacle risks and have poor accuracy; (2) Road surface risk value The ultrasonic sensor precisely fused components output by the residual correction-based sensor data dynamic fusion and noise reduction module. Normalized quantitative calculations yielded the following results. The specific calculation formula for accurately characterizing the height of road surface bumps and the degree of surface damage is as follows: ;in , These are the maximum and minimum road surface bump detection thresholds preset for the ultrasonic sensor, corresponding to the maximum road surface bump height that visually impaired users can tolerate and the absolute flat road surface reference value. For the first The frame is the ultrasonic fusion component value output by the residual-corrected sensor data dynamic fusion and noise reduction module; the calculation logic is: the more severe the road surface deformation and damage, the higher the value of the ultrasonic fusion component. The larger the value, the better the result after normalization. The higher the risk value, the better, especially when the road surface is smooth and undamaged. Approaching zero, it can accurately quantify the risks of road bumps and tripping, adapting to the travel characteristics of visually impaired users who cannot visually identify road defects.

[0023] (3) Risk value of pedestrian flow The visual sensor's precisely fused component output from the dynamic fusion and noise reduction module based on residual correction of sensor data. Normalized quantitative calculations yielded the following results. By integrating real-time pedestrian density, pedestrian movement speed, and disorder multi-dimensional information, the data undergoes noise reduction and dynamic weight correction via a sensor data dynamic fusion and noise reduction module based on residual correction. The specific calculation formula is as follows: ;in , These are the maximum and minimum pedestrian risk detection thresholds preset by the visual sensor, corresponding to the road saturation pedestrian density and the baseline state of unmanned passage. For the first The frame represents the visual fusion component value output by the sensor data dynamic fusion and noise reduction module based on residual correction. The calculation logic is as follows: the denser the crowd in the scene, the faster the pedestrians move, and the more chaotic the motion trajectory, the better. The larger the value, the more it corresponds to The higher the risk value, the better; when there are no pedestrians on the road, Approaching zero, it can accurately quantify the risk of dynamic pedestrian collisions and congestion; S32. Risk Stratification Threshold Classification: Using a risk stratification mechanism, the overall risk of all paths is divided into three levels: low, medium, and high. Fixed risk threshold ranges are preset, combined with the tolerance of visually impaired users. Low-risk level: Overall risk value It is safe to pass; medium risk level: Exercise caution and monitor in real time; High-risk level: Passage is prohibited; The path comprehensive risk value is calculated using an improved weighted quantification formula: ;in , , For risk weighting coefficients, satisfying ; Risk weighting coefficient , , The specific acquisition logic is as follows: Step 1, Weighted Basis Mapping: The residual-corrected sensor data dynamic fusion and noise reduction module outputs the dynamic fusion weights of the three types of sensors in real time. , , These represent the accuracy and reliability of data from LiDAR, ultrasonic sensors, and visual sensors under different road conditions. Higher reliability and smaller data errors correspond to larger weight values. Based on the core logic that "the higher the reliability of sensor data, the higher the reference value of the corresponding risk indicator," the dynamic sensor weights are used as the initial basis for the risk weights, and the mapping relationship is as follows: ,in , , As the initial base of uncorrected risk weights, the adaptive algorithm output of the sensor data dynamic fusion and noise reduction module based on residual correction is fully traceable, realizing data-level linkage and solving the technical defect of the disconnect between traditional risk weights and sensor detection accuracy. Step 2, Traffic Safety Priority Correction: Considering the traffic characteristics of visually impaired users, obstacle collision risk is considered a fatal safety risk, road bumps are considered a secondary injury risk, and pedestrian congestion is considered a minor disturbance risk. A fixed safety priority correction factor is introduced. , , The initial base is then weighted and corrected to strengthen the weighting of core security risks. The correction formula is as follows: ;in , , The weight parameters after priority correction are adjusted using differentiated correction factors to meet the core passage safety needs of visually impaired users: "prioritize obstacle avoidance, then smooth terrain, and finally avoid crowds." Step 3: Solve for the final weights using normalization constraints: To satisfy the mathematical constraints of weight normalization and ensure the overall risk value... Stable landing The corrected parameters are normalized within the specified interval to obtain the final dynamic risk weight, calculated using the following formula: , ;in For the first The real-time dynamic risk weight is dynamically updated based on the sensor data accuracy of the noise reduction module and the road condition scenario, according to the sensor data dynamic fusion based on residual correction. Step 4: Steady-state optimal parameter calibration: Extensive testing under normal urban road conditions verified that the detection accuracy of the three types of sensors is stable in typical traffic scenarios. The dynamic weights of the sensor data dynamic fusion and noise reduction module based on residual correction tend to a fixed value. The steady-state optimal risk weight coefficient is obtained through iterative solution. , , This value represents the optimal steady-state result after the adaptive algorithm converges. Under extremely complex road conditions, the weights can be dynamically adjusted in real time, balancing stability in general scenarios with adaptability to specific scenarios. S33. Adaptive Iterative Path Selection: An iterative selection mechanism is adopted, prioritizing risk and supplementing it with distance. The iterative formula is as follows: ,in The path is optimized by comprehensively considering the cost; the smaller the cost, the better the path. This represents the current path length. The maximum length of all alternative paths; The distance balancing coefficient, with a value of 0.2, is used to balance the weights of risk and path length, preventing excessively long paths due to an overemphasis on low risk. The two path length parameters are the basic geometric parameters for path cost calculation, and their specific acquisition methods are as follows: (1) Current path length How to obtain: This invention pre-generates multiple connected alternative navigation paths based on the navigation start point, end point, and real-time passable area. Each path consists of a continuous, ordered sequence of navigation coordinate points. Let a single alternative path contain... There are 1 navigation coordinate node, and the coordinates of adjacent nodes are 1. , ,in The length of the current complete path is calculated by accumulating Euclidean distances. The calculation formula is: In each 200ms global iteration cycle, the system verifies the passable area of ​​the current path in real time. If road conditions are obstructed or risk areas change, causing minor adjustments to path nodes, the system will synchronously recalculate the data. This ensures that the path length parameter is completely matched with real-time traffic conditions.

[0024] (2) The maximum length of all alternative paths How to obtain: Let the set of currently valid alternative paths in the system be . , This represents the total number of valid candidate paths for the current frame. , , These are the real-time path lengths of each candidate path. By iterating through all valid candidate path lengths, the maximum value is selected as the current frame's length. The solution formula is: ; As a globally unified parameter for a single frame, all candidate paths share this parameter within the same iteration cycle, thereby normalizing the path length, eliminating the interference of absolute path length differences on cost calculation, and standardizing and rationalizing the weight balance logic between risk value and path length. Iterative filtering logic: Traverse all pre-selected access paths and call the latest updated fusion component from Module 1 in real time. , , Synchronously update each path , , Overall risk value Comprehensive cost of the path The process involves rigorously eliminating high-risk paths, iteratively updating the cost of the remaining paths, and ultimately selecting the path with the lowest overall cost as the optimal navigation path. This step implements a linked iterative logic of "real-time data update of the sensor data dynamic fusion and noise reduction module based on residual correction → real-time risk refresh of the risk selection module based on the adaptive iterative path selection with hierarchical risk thresholds → real-time path optimization."

[0025] S4. Update the sensor data from step S1 in real time, dynamically iterate the path risk level, and complete the real-time path correction output; the specific implementation steps are as follows: S41, Global Synchronization Iteration Cycle Trigger. Based on the sensor acquisition and algorithm computation performance described above, a unified 200ms global iteration cycle is set. This cycle corresponds perfectly to the S1 sensor acquisition cycle and the data update cycle of the residual-corrected sensor data dynamic fusion and noise reduction module, balancing dynamic road condition response sensitivity with equipment computing power load. The system generates a global iteration trigger command every 200ms, synchronously driving the S1 sensor acquisition preprocessing flow, the residual correction fusion and noise reduction algorithm of the residual-corrected sensor data dynamic fusion and noise reduction module, and the risk iteration screening algorithm of the hierarchical risk threshold adaptive iterative path screening module to update synchronously. This ensures complete alignment of all data sequences and avoids path determination deviations caused by asynchronous updates.

[0026] S42. The sensor data dynamic fusion and noise reduction module based on residual correction iteratively updates all high-precision sensor data. After the iteration command is triggered, the multi-source sensor data preprocessing, baseline error calculation, adaptive solution of residual correction coefficients, dynamic weight allocation, and multi-source data fusion full-process improved algorithm are re-executed, and the current data is updated in real time. All core parameters of the frame: reference errors of three types of sensors Global instantaneous error Global historical residuals Adaptive residual correction coefficient Sensor dynamic fusion weights , , The final update yields the latest globally merged data. and three types of independent precise fusion components , , It provides the latest, noise-free, and residual-corrected accurate data source for path risk iteration, completely eliminating the problem of reusing outdated data in traditional algorithms.

[0027] S43. The hierarchical risk threshold adaptive iterative path selection module dynamically refreshes the risk parameters of the entire path. Based on the residual correction-based sensor data dynamic fusion and noise reduction module updated in S42, it fuses component data and performs a full iterative update of the risk parameters for each of the pre-selected travel paths, relying entirely on the normalization algorithm and adaptive weighting algorithm described above. First, the basic risk value of a single path is updated in real time using three types of risk normalization formulas: obstacle risk value. Road surface risk value Crowd risk value ; Secondly, the real-time dynamic weighting of the sensor data dynamic fusion and noise reduction module based on residual correction is reused. After weighted basis mapping, safety priority correction, and normalization solution, the dynamic risk weights are updated in real time. , , ; Finally, through the comprehensive risk formula Iteratively update the real-time comprehensive risk value of each alternative path. It accurately maps the real-time traffic risks of the current road conditions.

[0028] S44. Path risk level determination and comprehensive cost iterative calculation. Based on a preset risk stratification threshold: low risk. Medium risk High risk The risk level of all candidate paths is reassessed, and all high-risk paths in the current frame are directly eliminated to avoid dangerous road sections. Simultaneously, the path length parameter is combined with a comprehensive cost formula. The comprehensive screening cost of iteratively updating the remaining valid paths In the formula, the distance balance coefficient The principle of keeping things constant ensures a unified balance between risk and path length.

[0029] S45. Risk Level Jump Detection and Dynamic Path Correction Output. Compare the path risk level and comprehensive cost value of the current iteration frame with the previous iteration frame, and execute a differentiated path output strategy to avoid frequent path jitter. (1) Steady-state scenario: If the risk level of all effective paths does not change and the overall cost is low. If the fluctuation is less than the preset small threshold, the road conditions are determined to be stable, the current optimal guide path remains unchanged, and the passage of visually impaired users is guaranteed to be stable; (2) Sudden change scenario: If the path risk level increases, the original optimal path jumps to the medium / high risk level, or the comprehensive cost value is selected through iterative screening. Smaller, lower-risk paths immediately trigger path replanning, filtering the current frame. The path corresponding to the minimum value is used as the latest optimal guide path, and real-time correction output is completed.

[0030] This invention discloses a method for selecting navigation routes for the visually impaired based on intelligent sensors, belonging to the field of intelligent assisted travel technology. The invention includes a sensor data dynamic fusion and noise reduction module based on residual correction and a hierarchical risk threshold adaptive iterative route selection module. This invention first collects environmental data synchronously through multiple sensors, relying on… The criteria complete data preprocessing; then, through a self-developed residual correction dynamic weight algorithm, the fusion weights of each sensor data are adaptively adjusted to effectively remove noise, correct detection biases, and output high-precision environmental fusion data; on this basis, the fusion data is decomposed to obtain three types of risk quantification indicators, and the risk weights are adaptively solved in combination with the priority of visually impaired passage safety to complete the path risk stratification rating, construct a comprehensive cost function of "risk first, distance second" to screen the optimal path, and set a global iteration cycle of 200ms to realize dynamic real-time correction of sensor data, risk level, and guide path.

[0031] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for selecting guideways for the visually impaired based on intelligent sensors, characterized in that, Includes the following steps: S1. Real-time collection of travel environment data through intelligent sensor array, including obstacle distance data collected by lidar, road surface smoothness data collected by ultrasonic sensors, and pedestrian and vehicle flow density data collected by visual sensors, to complete multi-source data preprocessing; S2. The multi-source data preprocessed in step S1 is denoised and dynamically weighted by the sensor data dynamic fusion and denoising module based on residual correction, and high-precision unified environmental perception data is output. S3. The environmental perception data output in step S2 is used to perform risk stratification and iterative filtering algorithm of the hierarchical risk threshold adaptive iterative path filtering module to quantify the risk of the pre-selected access path, and the optimal safe blind path is selected by combining iterative optimization algorithm. S4. Update the sensor data in step S1 in real time, dynamically iterate the path risk level, and complete the real-time path correction output.

2. The method for selecting a guide path for the visually impaired based on intelligent sensors according to claim 1, characterized in that: The specific implementation steps of S2 are as follows: S21. Multi-source sensor data preprocessing: Outlier removal is performed on the multi-source data collected by the intelligent sensor group. The criteria remove abrupt noise points from single-frame data, eliminate invalid data caused by sensor jitter and environmental interference, and retain valid detection data sequences; specifically, the first... The effective data sequence of the sensor is ,in These correspond to LiDAR, ultrasonic sensors, and vision sensors, respectively. This refers to the number of sampling points in a single frame of data. S22. Data Reference Error Calculation: Calculate the real-time detection error of various sensor data, using the deviation between two adjacent frames of data as the reference error. The formula is as follows: ,in, For the first Type 1 sensor The baseline error of the frame data; For the first Valid frame detection data; This is valid detection data from the previous frame; S23. Residual Correction Dynamic Weight Calculation: A global residual factor is introduced to correct the weight allocation, abandoning the traditional fixed weight mode. The weight calculation formula is as follows: ,in For the first Type 1 sensor The dynamic weights of frame data satisfy ; For the first The historical residual mean of a sensor type characterizes the long-term detection bias of the sensor. This is the residual correction factor, and its value range is... This is used to balance the weighting effects of instantaneous errors and historical residuals; S24. Weighted Fusion and Secondary Noise Reduction: Multi-source data fusion is performed based on dynamic weights. The fusion formula is as follows: ;in For the first The frames are ultimately fused with environmental data; For the preprocessed first The system collects valid sensor data; finally, it performs secondary noise reduction through sliding window mean filtering, outputting real-time environmental perception data with high signal-to-noise ratio and high accuracy, which is then transmitted to the hierarchical risk threshold adaptive iterative path filtering module in step S3.

3. The method for selecting a guide path for the visually impaired based on intelligent sensors according to claim 2, characterized in that: The residual correction coefficient in S23 The specific acquisition logic is as follows: S23.1 Calculate the global instantaneous error fluctuation mean, which characterizes the overall real-time detection stability of all sensors in the current frame, conforming to the dynamic road condition interference characteristics. The calculation formula is as follows: ,in, For the first The average instantaneous error fluctuation of the frame is determined by the real-time reference errors of the three types of sensors. The mean is obtained by solving; S23.2 Calculate the global historical residual mean to characterize the long-term detection accuracy stability of the sensor and avoid the influence of inherent hardware deviations of the sensor. The calculation formula is as follows: ,in For the first The mean of global historical residuals is derived from the mean of historical residuals of the three types of sensors. The mean is obtained by solving; S23.3 Adaptive solution for initial residual correction coefficient: Based on the balance between instantaneous error and historical residual, a proportional solution formula is constructed: ;in This is the correction factor for the unconstrained initial residuals; S23.4 Threshold constraint pruning, completing the final coefficient assignment: The initial coefficients are constrained and corrected, as shown in the following formula: .

4. The method for selecting a guide path for the visually impaired based on intelligent sensors according to claim 1, characterized in that: The specific implementation steps of S3 are as follows: S31. Quantitative Modeling of Multidimensional Risk Indicators: Based on the fused data output by the residual-corrected sensor data dynamic fusion and noise reduction module, three types of core access risk indicators are extracted and quantitatively assigned according to the access characteristics of visually impaired users. (1) Obstacle risk value Extracted from LiDAR fusion data and quantified based on obstacle distance and volume; (2) Road surface risk value Extracted from data fusion by ultrasonic sensors, and quantified based on road surface smoothness and the height of bumps and depressions; (3) Risk value of pedestrian flow Extracted from data fusion by visual sensors, and quantified based on pedestrian density and movement speed; S32. Risk Stratification Threshold Classification: Using a risk stratification mechanism, the overall risk of all paths is divided into three levels: low, medium, and high. Fixed risk threshold ranges are preset, combined with the tolerance of visually impaired users. Low-risk level: Overall risk value It is safe to pass; medium risk level: Exercise caution and monitor in real time; High-risk level: Passage is prohibited; The path comprehensive risk value is calculated using an improved weighted quantification formula: ;in , , For risk weighting coefficients, satisfying ; S33. Adaptive Iterative Path Selection: An iterative selection mechanism is adopted, prioritizing risk and supplementing it with distance. The iterative formula is as follows: ,in The path is optimized by comprehensively considering the cost; the smaller the cost, the better the path. This represents the current path length. The maximum length of all alternative paths; This is the distance balance coefficient.

5. The method for selecting a guide path for the visually impaired based on intelligent sensors according to claim 4, characterized in that: The specific implementation details of the multidimensional risk indicator quantitative modeling in S31 are as follows: Quantitative modeling of multidimensional risk indicators of fused data from a dynamic fusion and denoising module based on residual correction: The final output of the dynamic fusion and denoising module based on residual correction of sensor data is the fused environmental data. The result is a weighted fusion of three types of sensor data, containing complete and accurate environmental information on current road conditions, obstacles, road surface, and pedestrian flow; firstly, the... By decomposing the components and separating the precise fusion components corresponding to the three types of sensors, the core risk indicators for visually impaired access are accurately quantified, completely eliminating the original sensor noise data. The decomposition and quantification formula is as follows: ;in , , These are the single-class precise fusion components of the LiDAR, ultrasonic sensor, and visual sensor data fused by the residual correction-based dynamic fusion and noise reduction module, respectively, satisfying the requirements of... ; , , The sensor weights are dynamically and adaptively solved by the sensor data dynamic fusion and noise reduction module based on residual correction. , , Effective sensor data after preprocessing by the residual correction-based sensor data dynamic fusion and noise reduction module; Based on the above precise fusion components, the three types of risk indicators were quantified: (1) Obstacle risk value The precisely fused components of the lidar output from the sensor data dynamic fusion and noise reduction module based on residual correction. The results were obtained through global normalization quantitative calculation. The system integrates two core sensor data points: real-time obstacle distance and obstacle size. The specific normalized calculation formula is as follows: ;in , These are the preset maximum and minimum effective detection thresholds for the lidar sensor; For the first The frame is the LiDAR fusion component value output by the sensor data dynamic fusion and noise reduction module based on residual correction; (2) Road surface risk value The ultrasonic sensor precisely fused components output by the residual correction-based sensor data dynamic fusion and noise reduction module. Normalized quantitative calculations yielded the following results. The specific calculation formula for accurately characterizing the height of road surface bumps and the degree of surface damage is as follows: ;in , These are the maximum and minimum road surface bump detection thresholds preset for the ultrasonic sensor, corresponding to the maximum road surface bump height that visually impaired users can tolerate and the absolute flat road surface reference value. For the first The frame is the ultrasonic fusion component value output by the residual-corrected sensor data dynamic fusion and noise reduction module; (3) Risk value of pedestrian flow The visual sensor's precisely fused component output from the dynamic fusion and noise reduction module based on residual correction of sensor data. Normalized quantitative calculations yielded the following results. By integrating real-time pedestrian density, pedestrian movement speed, and disorder multi-dimensional information, the data undergoes noise reduction and dynamic weight correction via a sensor data dynamic fusion and noise reduction module based on residual correction. The specific calculation formula is as follows: ;in , These are the maximum and minimum pedestrian risk detection thresholds preset by the visual sensor, corresponding to the road saturation pedestrian density and the baseline state of unmanned passage. For the first The frame is the visual fusion component value output by the sensor data dynamic fusion and noise reduction module based on residual correction.

6. The method for selecting a guide path for the visually impaired based on intelligent sensors according to claim 1, characterized in that: Risk weighting coefficient in S32 , , The specific acquisition logic is as follows: Step 1, Weighted Basis Mapping: The residual-corrected sensor data dynamic fusion and noise reduction module outputs the dynamic fusion weights of the three types of sensors in real time. , , These represent the accuracy and reliability of data from lidar, ultrasonic sensors, and visual sensors under different road conditions. Higher reliability and smaller data errors correspond to larger weight values. The sensor dynamic weights are used as the initial basis for the risk weights, and the mapping relationship is as follows: ,in , , The initial basis for uncorrected risk weights; Step 2, Traffic Safety Priority Correction: Considering the traffic characteristics of visually impaired users, obstacle collision risk is considered a fatal safety risk, road bumps are considered a secondary injury risk, and pedestrian congestion is considered a minor disturbance risk. A fixed safety priority correction factor is introduced. , , The initial basis is then weighted and corrected using the following formula: ;in , , The weight parameters after priority correction are adjusted using a differential correction factor. Step 3: Normalize the constraints to obtain the final weights: Normalize the corrected parameters to obtain the final dynamic risk weights. The calculation formula is as follows: , ;in For the first Real-time dynamic risk weights for each frame.

Images