An ai vision-based process guidance method and system for visually impaired students running

Abstract

The application provides an AI vision-based process guiding method and system for visually impaired students running, and relates to the technical field of image processing, which comprises the following steps: acquiring running environment visual data, obstacle position images and visually impaired student position images; constructing a running track space model according to the running environment visual data; extracting real-time positions of the visually impaired student and the obstacles in the running track coordinate system; determining the running environment passability of the visually impaired student; when the determination result is passable, planning a running path, otherwise, executing a safety avoidance strategy; planning a running path for the visually impaired student according to the real-time position of the visually impaired student and the real-time position of the obstacles, and obtaining an optimal running path; calculating the deviation state between the visually impaired student and the optimal running path in real time; adjusting the yaw of the running direction of the visually impaired student according to the deviation state to obtain a running following instruction; and guiding the visually impaired student to run according to the running following instruction.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for guiding visually impaired students during running based on AI vision. Background Technology

[0002] With the continuous development of campus sports activities and the concept of inclusive education, the demand for running and other sports training among visually impaired students is gradually increasing. Running requires a high level of spatial perception, directional judgment, and obstacle avoidance ability. Due to limited access to visual information, visually impaired students have difficulty accurately perceiving track boundaries, direction of travel, and changes in the surrounding environment, posing safety hazards such as deviating from the track and colliding with obstacles. Therefore, there is an urgent need for a technological means to provide real-time perception and guidance for visually impaired students in complex running environments to ensure their safety and improve their exercise experience.

[0003] However, existing technologies for assisting visually impaired individuals with movement often rely on simple wearable sensors, human companions, or fixed track devices. These typically provide only limited directional cues or post-run alerts, lacking a comprehensive understanding of the overall geometry of the track and the dynamic environment. Furthermore, existing vision-based methods often focus on single-frame target detection or simple track recognition, struggling to maintain stability under varying lighting conditions, occlusion, or the presence of dynamic obstacles. They also lack the ability to comprehensively assess student positional uncertainties and environmental risks, making it difficult to achieve running guidance that balances safety, continuity, and robustness. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method for guiding visually impaired students during running based on AI vision. This method addresses the limitations of existing technologies, which often rely on simple wearable sensors, human companions, or fixed track devices. These methods typically provide only limited directional cues or post-run alerts, lacking a comprehensive understanding of the overall geometry of the track and the dynamic environment. Furthermore, existing vision-based methods often focus on single-frame target detection or simple track recognition, making it difficult to maintain stability under conditions of changing lighting, occlusion, or the presence of dynamic obstacles. They also lack the ability to comprehensively assess the uncertainty of student posture and environmental risks, making it difficult to achieve a running guidance method that balances safety, continuity, and robustness.

[0005] A first aspect of this invention proposes a method for guiding visually impaired students during running based on AI vision, comprising:

[0006] S1: Acquire visual data of the running environment, images of obstacle locations, and images of the location of visually impaired students;

[0007] S2: Construct a track space model based on the visual data of the running environment;

[0008] S3: Based on the visually impaired student's position image, the obstacle's position image, and the track space model, extract the real-time positions of the visually impaired student and the obstacle in the track coordinate system to obtain the real-time positions of the visually impaired student and the obstacle.

[0009] S4: Based on the track space model, the real-time location of the visually impaired student, and the real-time location of the obstacle, determine the accessibility of the running environment for the visually impaired student.

[0010] S5: If the determination result is that it is passable, proceed to step S6; otherwise, execute the safety avoidance strategy.

[0011] S6: Based on the real-time location of the visually impaired student and the real-time location of the obstacle, plan the running path of the visually impaired student and determine the optimal running path;

[0012] S7: Calculate the deviation between the visually impaired student and the optimal running path in real time;

[0013] S8: Based on the deviation state, adjust the running direction of the visually impaired student to obtain a running follow instruction;

[0014] S9: Guide the visually impaired student's running process according to the running follow instruction.

[0015] A second aspect of the present invention provides a running guidance system for visually impaired students based on AI vision, comprising: a processor and a memory;

[0016] The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the AI ​​vision-based running guidance method for visually impaired students as described in the first aspect.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0018] In this embodiment of the invention, by acquiring visual data of the running environment, images of obstacle locations, and images of the visually impaired student's location, and constructing a track space model based on these, a comprehensive modeling of the track geometry, runnable area, and environmental elements is achieved. The structured track space model maps visual perception results uniformly to the track coordinate system, allowing the visually impaired student's location and obstacle locations to be expressed within the same spatial reference frame. This significantly improves the completeness of environmental understanding and the stability of positioning results. Under the constraints of the track space model, the drivability of the running environment is determined for the real-time locations of the visually impaired student and obstacles. Path planning is then executed only when the determination result indicates drivability. This achieves a hierarchical decision-making mechanism of "determine first, then plan," enabling early detection and intervention before the student deviates or collides. By identifying potential impassable conditions and triggering safety avoidance strategies when necessary, this invention effectively avoids the lag issues caused by relying solely on instantaneous position or simple distance thresholds for judgment, thereby improving the overall safety and robustness of the running process. Under passable conditions, by planning the optimal running path and calculating the deviation between the visually impaired student and the path in real time, and combining yaw adjustment to generate running follow instructions, continuous guidance and dynamic correction of the visually impaired student's running process are achieved. This invention can adaptively adjust the guidance instructions according to the student's real-time movement status, avoiding guidance failure or frequent correction problems caused by one-time path planning or static instructions, making the running guidance process smoother, more coherent, and in line with human movement characteristics, thereby improving the feasibility, comfort, and safety of running training for visually impaired students. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for guiding visually impaired students during running based on AI vision, as provided in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of a running guidance system for visually impaired students based on AI vision, provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] The following description, in conjunction with the accompanying drawings, details the AI-based vision-based running guidance method for visually impaired students provided by the present invention through specific embodiments and application scenarios.

[0024] Reference manual attached Figure 1 The diagram illustrates a flowchart of a method for guiding visually impaired students during running based on AI vision, provided by an embodiment of the present invention.

[0025] This invention provides a method for guiding visually impaired students during running based on AI vision, which may include the following steps:

[0026] S1: Acquire visual data of the running environment, images of obstacle locations, and images of the location of visually impaired students.

[0027] In one possible implementation, the visual data of the running environment specifically includes: the track boundary line, the track center line, the runnable area, the geometry of the straight sections, the geometry of the curves, the changes in track width, the spatial distribution of static obstacles around the track, and the spatial distribution of dynamic obstacles around the track.

[0028] Specifically, continuous visual data is collected during the running process using camera devices deployed on the sides or above the track, or worn by visually impaired students. This visual data includes at least environmental images reflecting the track structure, images showing the visually impaired student's position (including their anatomy), and obstacle position images representing the distribution of obstacles in the running environment. The environmental visual data characterizes the track boundaries, centerline, runnable area, curve morphology, and overall track structural features. The visually impaired student position images characterize their appearance, posture, and relative position changes within the scene. The obstacle position images characterize the spatial location, outline range, and displacement over time of static and dynamic obstacles located within or around the track, reflecting the distribution of obstacles relative to the track structure and the visually impaired student.

[0029] S2: Construct a running track spatial model based on visual data of the running environment.

[0030] In this embodiment of the invention, by constructing a track space model based on visual data of the running environment, the originally scattered and instantaneous visual perception results can be transformed into a unified representation with clear geometric structure and spatial constraints. This allows information such as track boundaries, center lines, runnable areas, and width changes to be modeled as a whole within the same coordinate framework. This provides a stable, continuous, and reusable spatial foundation for the location mapping of visually impaired students, environmental risk assessment, and subsequent path planning, significantly improving the reliability and safety of the running guidance process.

[0031] In one possible implementation, S2 specifically includes:

[0032] S201: Perform temporal image stabilization and consistency enhancement operations on the visual data of the running environment to obtain stabilized and enhanced data.

[0033] In this embodiment of the invention, inter-frame motion analysis and temporal alignment are performed on continuously acquired visual data of the running environment to suppress image instability caused by camera shake, slight changes in viewing angle, or instantaneous occlusion. Based on this, regions with stable spatial structures in multiple frames are subjected to consistency enhancement processing to highlight long-term structural features such as track boundaries and center lines, while reducing random noise and short-term interference, resulting in stable image enhancement data with temporal stability and structural consistency.

[0034] S202: Construct the temporal structure probability field of the runway based on the image stabilization enhancement data.

[0035] Specifically, based on the stabilized image enhancement data, structured visual analysis of the runway scene is performed. A deep learning-based visual perception model is used to perform frame-by-frame inference on the stabilized image enhancement data, outputting the instantaneous probability distribution of the runway structure, including the runway boundary probability. Probability of runway center guide strip . Used to characterize pixels Confidence level of belonging to the runway boundary structure Used to characterize pixels The confidence scores for locations in the runway center area or near the main direction of travel are both determined by the visual perception model at the current moment. t The stabilized image enhancement data is directly output after semantic parsing. To enhance the continuity and stability of the runway structure in the temporal dimension, an adaptive temporal fusion mechanism is introduced to recursively update the structure probability of the current frame and the historical structure probability, constructing a runway temporal structure probability field. The update formula is as follows:

[0036]

[0037]

[0038] in, express t Real-time updates k Runway-like temporal structure probability, Indicates the temporal fusion weight coefficients. This indicates the spatial location in the runway image coordinate system. express t- Output at time 1 k Runway-like temporal structure probability, express t Output in real time k Class-instantaneous structure probability, This represents the Sigmoid function. This represents the weighting adjustment coefficient. express t Output the instantaneous probability of the runway center guide strip structure at all times. express t Output the instantaneous probability of the runway boundary structure at any given time. k Represents the structure category index, Indicates the runway boundary structure. This indicates the structure of the runway center guide strip.

[0039] It should be noted that those skilled in the art can set the magnitude of the time-series fusion weight coefficient and the weight adjustment coefficient according to actual needs, and this invention does not limit them.

[0040] At time, the structure categories are obtained. t Temporal probability Then, the temporal probabilities corresponding to different structural categories are combined in the same spatial coordinate system to form a runway temporal structure probability field composed of the boundary temporal probability field and the central guide zone temporal probability field.

[0041] S203: Calculate the dual boundary consistency index based on the runway temporal structure probability field.

[0042] Specifically, after obtaining the runway temporal structure probability field, based on the runway boundary temporal probability field... With the time-series probability field of the runway center guide strip Multiple continuous candidate boundary curves are extracted from the high-confidence boundary region, and then divided into left and right boundary sets according to the main runway direction determined by the central guiding strip. For any set of left and right boundary candidate curves... A dual boundary consistency index is constructed to measure whether the curve pair conforms to the geometric and semantic features of the left and right boundaries of a real runway. The dual boundary consistency index is as follows:

[0043]

[0044] in, Represents the dual boundary consistency index. These represent a candidate curve for the left boundary and a candidate curve for the right boundary, located on either side of the runway centerline. and These represent the arc lengths of the left and right boundary curves, respectively. This represents the average spacing between the left and right boundary curves in the normal direction. This represents the prior value of the runway width. and These represent the average curvature of the left and right boundary curves, respectively. This represents the weighting coefficient of the runway width consistency constraint term. This represents a infinitesimal element representing the arc length along the boundary curve. This represents the weighting coefficient of the curvature symmetry constraint term.

[0045] It should be noted that those skilled in the art can set the prior value of the runway width, the weight coefficient of the runway width consistency constraint term, and the weight coefficient of the curvature symmetry constraint term according to actual needs, and this invention does not limit these settings.

[0046] S204: Based on the dual boundary consistency index, the set of boundary curve pairs in the runway time-series structure probability field is filtered to obtain the set of boundary candidate curves.

[0047] Specifically, using the dual boundary consistency index as the evaluation criterion, the left and right boundary curve pairs extracted from the high-confidence boundary region in the runway temporal structure probability field are uniformly sorted and discriminated. By setting a consistency index threshold or selecting curve pairs with high consistency index ranking, curve pairs that simultaneously satisfy the runway geometric structure characteristics in terms of boundary probability confidence, runway width consistency, and left and right curvature symmetry are preferentially retained, while invalid boundary combinations with width deviation, curvature imbalance, or insufficient semantic confidence are eliminated, thereby obtaining a set of candidate boundary curves that can stably represent the runway left and right boundary structure.

[0048] S205: Perform regularized fitting on the set of candidate boundary curves to obtain the set of fitted curves.

[0049] Specifically, for the set of candidate boundary curves, the discrete points of each candidate curve are used as constraints. A joint optimization model containing data fitting terms and smoothing regularization terms is introduced to perform continuous and smooth fitting on the left and right boundary curves respectively, so as to eliminate the jitter and breakage caused by visual noise, local occlusion or boundary discontinuity, thereby obtaining a set of left and right boundary fitting curves that are continuous in space, smooth in shape and conform to the geometric characteristics of the runway.

[0050] S206: Generate the centerline and runway width field based on the set of fitted curves.

[0051] Specifically, based on the set of fitted curves, the left and right boundaries are paired one by one according to the corresponding arc length positions. The runway centerline is generated by calculating the dual boundary centerline, and the spatial distance between the left and right boundaries is calculated at the same time to form the runway width field. The centerline is used to characterize the main direction of travel and the overall orientation of the runway, and the runway width field is used to describe the width variation characteristics of the runway at different locations, thus providing a structured geometric basis for subsequent runway area modeling and path planning.

[0052] S207: Combine the centerline and runway width field to construct a runway spatial model.

[0053] Specifically, after obtaining the runway centerline and runway width field, the centerline is used as the geometric reference axis, and the runway area constraint is generated in the normal direction of the centerline in combination with the runway width field. The runway space is modeled as a whole to form a runway space model that includes the left and right boundaries of the runway, the centerline, the width distribution and the range of the runway area, so that the runway space model can completely depict the geometric structure and passage range of the runway.

[0054] S3: Based on the location images of visually impaired students, obstacle location images, and the runway space model, extract the real-time positions of visually impaired students and obstacles in the runway coordinate system to obtain the real-time positions of visually impaired students and obstacles.

[0055] In one possible implementation, S3 specifically includes:

[0056] S301: Timestamp align the location images of visually impaired students and obstacle locations to obtain a synchronized frame group.

[0057] S302: Perform target parsing on the synchronization frame group to obtain the candidate set of visually impaired students and the candidate set of obstacles.

[0058] Specifically, based on the synchronous frame group, a visual analysis method is used to analyze the image content, identify target regions with characteristics of visually impaired students and target regions with characteristics of obstacles in the image, and form candidate sets of visually impaired students and candidate sets of obstacles respectively. The candidate sets include the spatial location, appearance features and corresponding confidence information of the target.

[0059] S303: Based on the runway space model, runway consistency gating is performed on the candidate set of visually impaired students and the candidate set of obstacles.

[0060] Specifically, using the runway space model, the candidate targets in the candidate sets of visually impaired students and obstacles are compared with the runway geometry to determine whether the candidate targets conform to the runway distribution characteristics in terms of spatial location. Candidate targets that are obviously located outside the runway range or do not match the runway structure are eliminated or downweighted, thereby retaining the visually impaired student candidate targets and obstacle candidate targets that are highly consistent with the runway space model.

[0061] S304: Extract the coordinate mapping representative points from the filtered candidate sets of visually impaired students and obstacle candidates respectively to obtain the representative points of visually impaired students and obstacle representatives.

[0062] Specifically, for the candidate sets of visually impaired students and obstacle candidates after the runway consistency screening, representative points that can stably reflect their contact position with the ground are selected from each candidate target as coordinate mapping benchmarks. Among them, the representative points of visually impaired students are preferentially selected from the weighted center of the bottom area of ​​the human body or key points, and the representative points of obstacles are selected from the lowest point or the center of the bottom edge in contact with the runway plane, so as to improve the accuracy and stability of subsequent coordinate mapping.

[0063] S305: Map the representative points of visually impaired students and the representative points of obstacles to the track coordinate system to obtain the initial track positions of visually impaired students and obstacles.

[0064] Specifically, based on the camera calibration relationship and the runway planar geometric relationship defined in the runway space model, the representative points of visually impaired students and the representative points of obstacles are transformed from the image coordinate system to the runway coordinate system to obtain the corresponding initial runway positions of visually impaired students and obstacles, thereby uniformly expressing the visual observation results under the runway space model.

[0065] S306: Project the initial track position of the visually impaired student and the initial track position of the obstacle onto the center line of the track to obtain the real-time position of the visually impaired student and the real-time position of the obstacle.

[0066] Specifically, in order to stably determine the relative positions of visually impaired students and obstacles on the runway even in the presence of visual noise, partial occlusion, or unclear runway structure, a probabilistically guided soft projection process is applied to the initial runway positions of visually impaired students and obstacles based on the runway centerline in the runway spatial model and the temporal probability of the center guidance strip in the runway temporal structure probability field. The optimal corresponding position parameters of the target in the runway direction are determined by jointly considering geometric distance and structural confidence. The calculation method is as follows:

[0067]

[0068]

[0069] in, This represents the target's real-time longitudinal position parameters along the runway direction. Indicates minimization. s Indicates the arc length parameter. Indicates the initial track position for visually impaired students or obstacles. This indicates that the runway centerline has an arc length parameter of... s The corresponding spatial location point at that time The coefficients represent probability-guided weightings, and log represents the logarithmic function. Represents a small positive number. This represents the lateral offset of the target relative to the runway centerline, and sign() represents the sign function. This indicates that the arc length parameter takes the optimal value. At that time, the spatial position point corresponding to the runway centerline, that is, the projection reference point of the target in the runway direction, Indicates the position of the runway centerline. The unit normal vector at point || represents the Euclidean norm of the vector.

[0070] It should be noted that those skilled in the art can set the probability-guided weight coefficient and the size of the tiny positive number according to actual needs, and this invention does not limit this.

[0071] Projection calculations are performed independently for visually impaired students and obstacles. The real-time position of the visually impaired student is calculated by substituting their initial track position into the formula above, and the real-time position of the obstacle is calculated by substituting the initial track position of each obstacle into the formula above. Both share the same track centerline reference frame but correspond to different target position parameters. This projection method introduces a center structure confidence constraint generated by AI vision on top of the traditional geometric nearest point projection, making the projection results more inclined towards the high-confidence track center region. This effectively suppresses position jumps caused by instantaneous detection errors, boundary breaks, or local occlusion, improving the temporal and spatial stability of the localization results for visually impaired students and obstacles.

[0072] S4: Based on the track space model, the real-time location of the visually impaired student, and the real-time location of obstacles, determine the accessibility of the running environment for the visually impaired student.

[0073] It should be noted that, unlike existing technical solutions that rely solely on track detection or instantaneous position determination of visually impaired students, this invention introduces a track spatial model and the parameterized pose of the visually impaired student in the track coordinate system. By constructing a local running environment determination unit and integrating the track geometric boundary crossing risk with the uncertainty of visually impaired student pose prediction, a continuous running environment accessibility determination index is formed. This determination method can identify potential impassable states during the running process before the visually impaired student crosses the boundary or collides, thereby significantly improving the safety and robustness of guiding visually impaired students in running, representing a substantial improvement over existing technologies.

[0074] In one possible implementation, S4 specifically includes:

[0075] S401: Construct a local decision unit based on the runway space model, the real-time location of the visually impaired student, and the real-time location of the obstacles.

[0076] Specifically, using the real-time position of the visually impaired student in the track coordinate system as the central reference point, and combining the track centerline direction, width distribution and runnable area range described in the track space model, a local judgment unit with a limited spatial range is constructed in front of the student's current direction of travel. At the same time, the real-time position of obstacles within this spatial range is included in the judgment unit to form a local running environment perception area.

[0077] S402: Calculate the left and right boundary margins based on the width field and the lateral offset of the visually impaired student in the runway space model.

[0078] Specifically, based on the runway width information corresponding to the current position provided by the runway space model, and combined with the lateral offset of the visually impaired student relative to the runway centerline, the remaining available space distance from the student's current position to the left and right boundaries of the runway is calculated respectively, so as to obtain the left and right boundary margins that characterize the safety margin of the student's current lateral position on the runway.

[0079] S403: Calculate the geometric risk index based on the left and right boundary margins.

[0080] Specifically, based on the remaining available space from the visually impaired student's current position to the left and right boundaries of the track, and considering the minimum value of the margin on the left and right boundaries and its symmetry, the safety level of the student in the lateral space of the track is quantitatively assessed. When the margin on either side decreases or the distribution of the margin on the left and right boundaries is significantly uneven, the corresponding geometric risk index increases. Through this method, the potential geometric constraint risks introduced by visually impaired students approaching the track boundaries or deviating from the center area of ​​the track are mapped into continuous geometric risk indicators.

[0081] S404: Calculate the spatiotemporal proximity risk index of the obstacle based on the real-time location of the obstacle and the real-time location of the visually impaired student.

[0082] Specifically, in the track coordinate system, the relative spatial positional relationship between visually impaired students and obstacles located within the local judgment unit is calculated. The analysis focuses on the longitudinal distance and lateral offset of obstacles relative to the student's current direction of travel, as well as whether the obstacles are within the student's potential path. Combined with the student's current movement state, the approach trend between obstacles and students in the spatial and temporal dimensions is assessed. Thus, the degree to which obstacles may interfere with or cause collisions to the student's running path is mapped into an obstacle spatiotemporal proximity risk index.

[0083] S405: Calculate the position error by subtraction based on the real-time and historical positions of visually impaired students.

[0084] S406: Based on the local decision unit, the position error is parametrically estimated using an uncertainty estimator to determine the pose uncertainty index.

[0085] Specifically, within a local decision unit centered on the current position of the visually impaired student, the position error calculated from the real-time and historical positions of the visually impaired student is used as the basic observation. Combined with the perceived reliability characteristics of the track structure within the local decision unit, the position error is parametrically analyzed through an uncertainty estimator. This maps the predictive reliability of the position error under the current running environment conditions into a pose uncertainty index, which reflects the degree of uncertainty in the pose prediction of the visually impaired student in the current environment.

[0086] The uncertainty estimator is a parametric estimation model used to jointly model positional errors and environmental structural reliability characteristics. It can employ a feature-mapping-based regression structure or a neural network structure to map the error to an uncertainty index. The inputs to the uncertainty estimator include the visually impaired student's positional error, as well as semantic uncertainty and structural change intensity indices calculated within local decision units. The output of the uncertainty estimator is a pose uncertainty index, which characterizes the level of uncertainty in the visually impaired student's pose prediction in the current running environment.

[0087] The training process of the uncertainty estimator is as follows:

[0088] During the offline training phase, a training dataset for the uncertainty estimator was constructed based on historical data collected from multiple running events of visually impaired students. This dataset involved short-term pose prediction of the historical position sequences of visually impaired students, followed by comparison of the predicted pose with the actual pose at the corresponding moment to obtain pose error samples. Simultaneously, within the local decision unit at the corresponding time, the semantic uncertainty index and structural change intensity index of the runway structure are calculated. The pose error and structural uncertainty characteristics are used as inputs to the estimator, and the mean parameter of the pose error distribution is output by the uncertainty estimator. With variance parameter During training, the estimator parameters are optimized by minimizing the negative log-likelihood loss of the pose error under the prediction distribution. The loss function is defined as:

[0089]

[0090] in, This represents the overall loss function value of the uncertainty estimator during the training phase. This represents the pose error training dataset. , Indicates the first i The true pose error corresponding to each training sample. Indicates the first i The local decision unit corresponding to each training sample Indicates the uncertainty estimator under the condition Below, regarding pose error The predicted mean, Indicates the uncertainty estimator under the condition Below, regarding pose error The prediction variance.

[0091] In one possible implementation, S406 specifically includes:

[0092] S4061: Based on local decision units, calculate semantic uncertainty index and structural change intensity index.

[0093] Specifically, within a local decision unit centered on the visually impaired student's current location, the temporal probability field of the runway structure is invoked to statistically analyze the probability distributions of the runway boundary structure and the central guiding strip structure. A semantic uncertainty index is calculated based on the dispersion of the structural probability distribution, and a structural change intensity index is calculated based on the spatial variation of the central guiding strip probability. This quantifies the semantic reliability and geometric stability of the runway structure at the current location. The calculation methods for the semantic uncertainty index and the structural change intensity index are as follows:

[0094]

[0095]

[0096] in, express t Semantic uncertainty index at time, expresst Local decision unit at time, Represents a very small positive number. express t Indicators of structural change intensity at any given time. Represents spatial gradient, This represents the L2 norm.

[0097] It should be noted that those skilled in the art can set the size of the extremely small positive number according to actual needs, and this invention does not limit it.

[0098] S4062: Calculate the pose error based on the real-time and historical positions of visually impaired students.

[0099] Specifically, in the track coordinate system, the real-time position of the visually impaired student at the current moment is compared with the short-term predicted position obtained from its historical position sequence, and the difference between the two is calculated as the pose error. The pose error includes at least the longitudinal error along the center line of the track, the lateral error perpendicular to the center line, and the corresponding orientation deviation, which is used to characterize the degree of deviation of the visually impaired student from the predicted motion state during the current running process.

[0100] S4063: Based on the semantic uncertainty index and the structural change intensity index, the pose error is parametrically estimated using an uncertainty estimator to obtain the pose error distribution parameters.

[0101] Specifically, pose error is used as the basic observation, and semantic uncertainty index and structural change intensity index calculated from the probabilistic features of the track structure within the local decision unit are used as environmental constraint inputs to modulate the credibility of pose error. An uncertainty estimator is introduced to parametrically model pose error and output distribution parameters that characterize the statistical properties of pose error, so that the pose error distribution can reflect the impact of changes in the running environment structure on the stability of motion prediction for visually impaired students.

[0102] S4064: Determine the pose uncertainty index based on the pose error distribution parameters.

[0103] Specifically, based on the pose error distribution parameters output by the estimator, a pose uncertainty index is constructed at the current time. This index characterizes the overall dispersion of the pose prediction error in the parameter space by performing a logarithmic mapping on the determinant of the pose error covariance matrix, and deviates from the reference shape parameter of the error distribution to characterize the anomaly of the pose error distribution. Thus, the scale and shape information of the pose error distribution are integrated to obtain a pose uncertainty index reflecting the reliability of the student's pose prediction at the current position. The pose uncertainty index is as follows:

[0104]

[0105] in, The pose uncertainty index at time t is represented. Represents the determinant operator. Represents the pose error vector of visually impaired students At any moment t The covariance matrix under the following conditions Let log represent the identity matrix, and let log represent the logarithmic function. express t The shape parameter of the distribution at time intervals, Represents a very small positive constant. Reference morphological parameters representing the distribution of pose error. This indicates the penalty weighting coefficient for deviation from the form.

[0106] It should be noted that those skilled in the art can set extremely small positive constants and the magnitude of the morphological deviation penalty weight coefficient according to actual needs, and this invention does not limit this.

[0107] S407: Calculate the comprehensive accessibility index based on geometric risk index, obstacle spatiotemporal proximity risk index, and pose uncertainty index.

[0108] Specifically, the geometric risk index calculated from the lateral spatial constraints of the track, the spatiotemporal proximity risk index of the obstacle calculated from the relative positional relationship between the obstacle and the visually impaired student, and the pose uncertainty index output by the uncertainty estimator are used as independent risk factors. A weighted fusion or mapping function is introduced to uniformly quantify and normalize the above risk factors, thereby constructing a comprehensive accessibility index to characterize the overall safety level of the current running environment.

[0109] S408: Determine the accessibility of the running environment for visually impaired students based on comprehensive accessibility indicators.

[0110] Specifically, the comprehensive accessibility index is compared with a preset accessibility judgment index. When the comprehensive accessibility index is within the safe range, the current running environment is determined to be accessible, allowing visually impaired students to continue running along the current running path. When the comprehensive accessibility index exceeds the judgment threshold, the current running environment is determined to be inaccessible, and corresponding safety prompts or avoidance strategies are triggered to reduce the potential risks to visually impaired students during running.

[0111] It should be noted that those skilled in the art can set the size of the preset accessibility judgment index according to actual needs, and this invention does not limit it.

[0112] S5: If the determination result is that it is passable, proceed to step S6; otherwise, execute the safety avoidance strategy.

[0113] Specifically, based on the determination of impassability and its corresponding risk type, the safety avoidance strategy includes at least one or more of deceleration avoidance, pause and waiting, and directional avoidance. First, when the impassability is mainly caused by the approach of a dynamic obstacle or insufficient minimum collision time, the system outputs voice commands to the visually impaired student via the running guidance terminal to decelerate or pause, guiding the student to reduce running speed or briefly stop moving to wait for the obstacle to move away or for the environment to return to safety. Second, when the impassability is mainly caused by the visually impaired student approaching the track boundary or the risk of crossing the boundary, the system generates temporary directional avoidance commands based on the track space model, guiding the student to make a slight directional adjustment towards the center of the track to avoid continuing to move into the danger zone. Furthermore, if the impassability persists, the system can maintain a waiting mode and periodically re-execute the running environment accessibility determination to restore normal path planning and running guidance procedures after changes in the environmental conditions.

[0114] In this embodiment of the invention, by introducing a branch control mechanism based on the feasibility determination result, the path planning process is only entered when the running environment meets the safety conditions, and a safety avoidance strategy is executed in a timely manner when there are potential risks. This can prevent visually impaired students from continuing to run in unsafe environments, thereby reducing misleading guidance and the occurrence of sudden dangerous situations, and improving the safety and reliability of the system operation.

[0115] S6: Based on the real-time location of the visually impaired student and the real-time location of the obstacles, plan the running path of the visually impaired student and determine the optimal running path.

[0116] In this embodiment of the invention, by planning the running path based on the real-time location of the visually impaired student and the real-time location of the obstacle, an optimal running path matching the student's current location and movement state can be generated under the premise of meeting the track space constraints and obstacle avoidance requirements. This makes the path planning result have good safety, continuity and adaptability, thereby providing a more stable and reasonable running guidance direction for the visually impaired student.

[0117] In one possible implementation, S6 specifically includes:

[0118] S601: Based on the real-time positions of the visually impaired student and the obstacle, estimate the velocities of the visually impaired student and the obstacle respectively, and obtain the motion state variables of the visually impaired student and the obstacle.

[0119] Specifically, at two adjacent moments and Below, obtain the position coordinates of visually impaired students in the track coordinate system or equivalent spatial coordinate system. and And the spatial coordinates of each dynamic obstacle at the corresponding moment. and And based on the position change and time interval between adjacent times. The velocity vectors of the visually impaired student and the obstacle at the current moment are obtained by difference calculation, thus representing their direction and speed of motion. The velocity vectors constitute the corresponding motion state variables, which are specifically:

[0120]

[0121]

[0122] in, This indicates that visually impaired students are always t The velocity vector, Indicates the first i A dynamic obstacle at any time t The velocity vector, express t The coordinates of visually impaired students in the horizontal direction (such as the horizontal direction of a running track). express t The coordinates of the visually impaired student in the vertical direction (height direction, which can be approximated as constant) at all times. express t The coordinates of the visually impaired student in the direction of movement (along the longitudinal direction of the track). express t Time of the first i The coordinates of a dynamic obstacle in the lateral direction. express t Time of the first i The coordinates of a dynamic obstacle in the total direction. express t Time of the first i The coordinates of a dynamic obstacle in the direction of travel. express t The coordinates of the visually impaired student in the horizontal direction at time 0. express t The vertical coordinates of the visually impaired student at time 0. express t The coordinates of the visually impaired student in the direction of movement at time 0. express t 0th minute i The coordinates of a dynamic obstacle in the lateral direction. express t 0th minute i The coordinates of a dynamic obstacle in the total direction. express t 0th minute i The coordinates of a dynamic obstacle in the direction of travel.

[0123] S602: Generate a set of candidate running paths based on the center line of the track and the real-time location of the visually impaired student.

[0124] Specifically, the centerline of the runway, as determined in the runway space model, is used as the reference path. Combining the current arc length position and lateral offset of the visually impaired student in the runway coordinate system, a planning interval is selected forward along the centerline under the constraint of the runway width field. Within the planning interval, different lateral offsets, curvature parameters, or smoothing control parameters are set to perform lateral offset and continuity processing on the centerline, thereby obtaining multiple candidate running paths covering different possible modes of movement.

[0125] S603: Calculate the minimum collision time based on the motion state of the visually impaired student and the motion state of the obstacle.

[0126] Specifically, at the current moment, obtain t Real-time position of visually impaired students in the track coordinate system and its velocity vector and obtain t Time of the first i Real-time position of each dynamic obstacle in the runway coordinate system and its velocity vector Based on the spatial relative distance and relative speed between the student and each obstacle, a minimum collision time estimation model is constructed. By minimizing the collision time estimates of all dynamic obstacles, the minimum collision time faced by the student at the current moment is obtained. This value is used to characterize the immediate safety level of the running environment and serves as the basis for subsequent path planning mode selection and constraint setting.

[0127] S604: Determine whether the minimum collision time is greater than the time threshold; if yes, proceed to step S605; otherwise, terminate the planning of the running path for the visually impaired student and return to step S4.

[0128] Specifically, by determining whether the minimum collision time between the visually impaired student and the obstacle is greater than a preset time threshold, the system assesses whether the current running environment has a sufficient time margin to continue path planning. When the minimum collision time is less than or equal to the time threshold, it indicates a potential emergency collision risk. In this case, the current running path planning process is terminated and the system returns to step S4, allowing the system to reassess the accessibility of the running environment and trigger corresponding safety avoidance strategies. This prevents the generation of unreliable or misleading running paths under high-risk conditions. When the minimum collision time is greater than the time threshold, it indicates that the current environment has sufficient reaction time and safety space, allowing the system to proceed to subsequent path planning steps. This improves the effectiveness of running guidance and the reliability of system operation while ensuring safety.

[0129] It should be noted that those skilled in the art can set the time threshold according to actual needs, and this invention does not limit it.

[0130] S605: Based on the runway centerline, the centerline reference heading sequence is obtained by calculating the tangential direction.

[0131] Specifically, based on the runway centerline determined in the runway spatial model, the centerline is discretized along its arc length, resulting in a series of centerline sampling points arranged in the order of travel. Subsequently, for adjacent centerline sampling points, their displacement direction in the planar coordinate system is calculated, and the local reference heading angle of the centerline at each sampling position is obtained through tangential direction calculation. This forms a centerline reference heading sequence that varies with arc length, used to characterize the runway's primary travel direction at different locations, and serving as the direction benchmark for subsequent path alignment consistency evaluation and path planning. The centerline reference heading sequence is as follows:

[0132]

[0133] in, Indicates the position of the runway centerline at the arc length. Reference heading angle at that location Indicates the first The centerline position corresponding to each discrete sampling point on the centerline. Represents the arctangent function in the four quadrants. Indicates the first +1 Centerline discrete sampling point corresponding to the centerline position This represents the lateral coordinate components of the runway centerline in a plane coordinate system. This represents the longitudinal coordinate components of the runway centerline in a plane coordinate system.

[0134] S606: Calculate the direction consistency cost based on the candidate path heading sequence and the centerline reference heading sequence.

[0135] Specifically, each candidate path is discretely sampled along its direction of travel, and the local direction of travel between adjacent sampling points is calculated to obtain the heading sequence of the candidate path. The heading sequence of the candidate path is then aligned point by point with the centerline reference heading sequence at the corresponding position. By calculating the angle difference between the two and introducing thresholding, the influence of small directional fluctuations on the evaluation results is suppressed, thereby obtaining the directional consistency cost that characterizes the degree of consistency between the candidate path and the main direction of travel of the runway. This cost is used to reflect whether the candidate path meets the requirements for forward and stable running guidance.

[0136] S607: Based on the directional consistency cost, perform preliminary sorting and filtering on the candidate running path set to obtain a subset of candidate paths.

[0137] Specifically, based on the calculated directional consistency cost, each candidate path in the candidate running path set is quantitatively evaluated and sorted in ascending order of directional consistency cost. On this basis, combined with a preset directional consistency threshold or sorting ratio rule, candidate paths whose direction deviates significantly from the main heading of the runway centerline are eliminated, retaining only candidate paths whose heading maintains a high degree of consistency with the runway centerline reference heading. This forms a subset of candidate paths, reducing the search space for subsequent path feasibility screening and optimization calculations, and improving the stability and directional consistency of the path planning results.

[0138] S608: Based on the runway width field and boundary constraints, the feasibility of candidate path subsets is screened through runnable area determination and boundary violation penalty to obtain a set of feasible paths;

[0139] Specifically, for each candidate path subset, the runway width field and boundary safety constraints provided in the runway space model are used to perform point-by-point detection along the direction of travel for each candidate path, determining whether each sampling point on the path is within the runnable area defined by the runway width field. Candidate paths that are completely within the runnable area are deemed to meet the basic feasibility requirements and are retained. For candidate paths that are partially close to or exceed the runway boundary, a boundary violation penalty is applied based on the degree of boundary violation, or they are directly determined as infeasible paths and eliminated. This process filters out a set of feasible paths that simultaneously meet the runway geometric constraints and safety margin requirements.

[0140] S609: Set the objective function and constraints.

[0141] The objective function is specifically designed to minimize the path offset between the candidate running path and the visually impaired student's current position, the overall curvature change of the path, and the directional deviation between the path's direction of travel and the runway centerline reference heading. Constraints include at least the runway runway area constraint, boundary safety margin constraint, path forward travel constraint, and motion continuity constraint. The runway runway area constraint ensures the running path remains within the passable area defined by the runway width field. The boundary safety margin constraint prevents the path from approaching the runway's left and right boundaries excessively. The path forward travel constraint limits the overall path's extension along the main direction of travel on the runway centerline, preventing reverse or unreasonable backtracking. The motion continuity constraint limits the directional change between adjacent path segments.

[0142] S610: Under the constraints of the conditions, the optimal running path is obtained by solving the objective function through an optimization algorithm.

[0143] Optionally, the optimization algorithm is specifically a genetic algorithm. Using a genetic algorithm, firstly, each path in the set of feasible paths is encoded to form the initial population of the genetic algorithm, where each individual corresponds to a candidate running path. Then, a fitness function is constructed based on the objective function, and the fitness of each path individual in the population is evaluated, so that the fitness value can comprehensively reflect the path's safety, directional consistency, smoothness, and obstacle avoidance effect. Based on this, a selection operator selects path individuals with higher fitness from the current population as parent individuals, and a crossover operator combines the path parameters or control variables of the parent paths to generate new offspring paths. Simultaneously, a mutation operator randomly perturbs the local parameters of some path individuals to enhance the diversity of the search space and avoid getting trapped in local optima. The above selection, crossover, and mutation processes are iteratively executed until a preset number of iterations or fitness convergence condition is met. Finally, the path individual with the best fitness is selected from the population as the optimal running path output. Genetic algorithms are a mature existing technology, and will not be elaborated further in this invention.

[0144] S7: Real-time calculation of the deviation between visually impaired students and the optimal running path.

[0145] In this embodiment of the invention, by calculating the deviation between the visually impaired student and the optimal running path in real time, the difference between the student's actual movement trajectory and the planned path can be continuously monitored, and the cumulative trend of lateral offset or directional deviation can be identified in a timely manner, thereby providing accurate and continuous status basis for subsequent yaw adjustment and improving the responsiveness and stability of the running guidance process.

[0146] In one possible implementation, S7 specifically includes:

[0147] S701: Based on the real-time location of the visually impaired student, determine the path reference location corresponding to the visually impaired student through path nearest point matching.

[0148] Specifically, based on the student's real-time position in the track coordinate system, the student's position is spatially matched with the centerline of the running path. By searching for the path point with the smallest Euclidean distance to the student's current position within the arc length parameter domain of the path centerline, the path reference position corresponding to the student at the current moment is determined, which is used to establish a one-to-one correspondence between the student's position and the running path.

[0149] S702: Calculate the lateral deviation of the visually impaired student based on the student's real-time location and path reference location.

[0150] Specifically, after determining the path reference position, the path normal direction coordinate axis is constructed based on the path tangential direction at the path reference position, and the student's current position is projected onto this normal direction. By calculating the difference between the student's coordinates in the normal direction and the normal coordinates of the path centerline, the lateral deviation of the student relative to the running path centerline is obtained.

[0151] S703: Calculate the instantaneous heading angle of the visually impaired student based on the time series of the student's real-time location.

[0152] Specifically, based on the real-time position data of visually impaired students in the track coordinate system collected continuously, the student position points at adjacent moments are selected, and the instantaneous heading angle of the visually impaired student at the current moment is determined by calculating the displacement direction between the current position and the previous position, which is used to characterize the student's current actual running direction.

[0153] S704: Calculate the heading deviation based on the optimal running path and instantaneous heading angle.

[0154] Specifically, after determining the path reference position for the visually impaired student, the tangential direction of the optimal running path at that path reference position is obtained, and the path tangential direction is compared with the instantaneous heading angle of the visually impaired student. By calculating the angle difference between the two, the heading deviation of the visually impaired student relative to the optimal running path is obtained, which is used to characterize the degree of deviation between the student's actual movement direction and the ideal path direction.

[0155] S705: Perform time-series consistency processing on the lateral deviation and heading deviation to obtain a stable deviation estimate.

[0156] Specifically, for deviation jitter caused by instantaneous position changes, visual perception noise, or gait fluctuations, temporal consistency processing is performed on lateral deviation and heading deviation respectively. The deviation is smoothed by moving average, exponential weighted filtering, or recursive update, so as to obtain continuous and stable lateral deviation and heading deviation estimates in the time dimension.

[0157] S706: Combining the runway width field and lateral deviation of the runway space model, determine the boundary approach risk index for visually impaired students.

[0158] Specifically, after determining the path reference position for the visually impaired student, the runway width information at that path reference position is obtained based on the runway width field in the runway spatial model, and half of that width is taken as the runway half-width. Combined with stable lateral deviation estimation, the remaining safety margin between the visually impaired student's current position and the left and right boundaries of the runway is calculated, and the safety margin is compared with the safety threshold. A boundary approach risk index is generated through normalization mapping or smooth function mapping.

[0159] It should be noted that those skilled in the art can set the size of the safety threshold and the preset range according to actual needs, and the present invention does not limit this.

[0160] S707: Combining stable deviation estimation and boundary proximity risk indicators, the deviation state between visually impaired students and the optimal running path is calculated through state assembly.

[0161] Specifically, the stable lateral deviation estimate and stable heading deviation estimate are combined with the boundary approach risk index. Through state assembly, a deviation state is constructed to comprehensively characterize the degree of deviation of the current position, the deviation of the movement direction, and the level of safety risk of the visually impaired student. The deviation state serves as the input state for the subsequent yaw adjustment and running follow command generation, which is used to realize real-time guidance and safety control of the visually impaired student's running process.

[0162] S8: Based on the deviation status, adjust the running direction of the visually impaired student to obtain the running follow instruction.

[0163] In this embodiment of the invention, by adjusting the running direction of a visually impaired student based on the deviation from the optimal running path and generating corresponding running follow instructions, complex path deviation information can be transformed into intuitive and executable guidance instructions, enabling visually impaired students to correct their direction of travel in a timely manner during running, reducing the accumulation of deviations, and improving the continuity, safety, and controllability of the running process.

[0164] In one possible implementation, S8 specifically includes:

[0165] S801: Based on the deviation state, a deviation prediction model is constructed by discretizing the small-angle error dynamics.

[0166] Specifically, under the assumption of small-angle deviation, the lateral deviation and heading deviation of visually impaired students are discretized and modeled to construct a predictive model to describe the evolution of the deviation state over time. This deviation prediction model is used to characterize the changing trends of lateral and heading deviations at future times under different yaw angle increment control parameters. The deviation prediction model is as follows:

[0167]

[0168]

[0169] in, Indicates the first The lateral deviation of a visually impaired student from the centerline of the running path at discrete moments is used to characterize the magnitude and direction of the student's current deviation from the path. Indicate the first The lateral deviation of a visually impaired student from the centerline of the running path at a discrete time. Indicates the first The deviation of a visually impaired student's course at discrete moments. This represents the scalar value of a visually impaired student's running speed at the current moment. Indicates time interval, Indicates the first The deviation of a visually impaired student's course at discrete moments. Indicates the first The yaw angle increment control quantity generated at each discrete moment. Indicates the first The geometric curvature of the running path at the path reference position corresponding to the visually impaired student at each discrete time.

[0170] S802: Set the parameters of the optimization cost function based on the deviation prediction model.

[0171] Specifically, based on the state variables of the deviation prediction model and the requirement to introduce boundary approach risk indicators, weight coefficients are set for lateral deviation, heading deviation, boundary approach risk indicators, yaw angle increment control, and their rate of change.

[0172] S803: Calculate the optimal yaw angle increment control quantity based on the deviation prediction model and optimized cost function parameters.

[0173] Specifically, within the set prediction time domain, with deviation from the prediction model as a constraint and the aforementioned cost function parameters as optimization weights, the constrained optimization problem is solved to obtain the sequence of yaw angle increment control quantities that minimize the cost function. The optimal yaw angle increment control quantity corresponding to the current moment is selected to guide visually impaired students in adjusting their running direction. The optimization objective function is expressed as follows:

[0174]

[0175] Where min represents minimization. The weighting coefficient represents the lateral deviation. The weighting coefficient represents the deviation from the heading. The weighting coefficients representing the risk indicators close to the boundary. Indicates the yaw angle increment control quantity The weighting coefficients, Indicates the rate of change of yaw angle increment The weighting coefficients, This indicates that in predicting the first... The lateral deviation of a visually impaired student from the centerline of the optimal running path at discrete moments. This indicates that in predicting the first... The deviation of a visually impaired student's course at discrete moments. Indicates the first The incremental yaw angle control applied at each discrete moment. This indicates that in predicting the first... At any discrete moment, the boundaries of visually impaired students approach risk indicators. N This indicates the length of the prediction time domain.

[0176] S804: Perform dead zone and hysteresis processing on the optimal yaw angle increment control to obtain the running follow command.

[0177] Specifically, the optimal yaw angle increment control is engineered by setting a dead zone to suppress minor ineffective adjustments, using a hysteresis mechanism to avoid frequent changes in running direction, and limiting the amplitude of the yaw angle increment, thereby generating running follow instructions that conform to human movement characteristics, which can be used to guide visually impaired students in adjusting their running direction in real time.

[0178] S9: Guide visually impaired students during their running process according to running instructions.

[0179] For example, once a running follow instruction is generated, it is output through a running guidance terminal worn by the visually impaired student to provide real-time guidance during their run. The running guidance terminal may include bone conduction headphones, a voice broadcast module, or other accessible interactive devices to provide clear and continuous voice guidance information to the visually impaired student without obstructing their perception of the external environment. The running follow instruction is converted into corresponding directional adjustment prompts, deviation warnings, or safety reminders based on the visually impaired student's current deviation status, and is broadcast in real-time via voice to guide the student in making minor adjustments to their running direction.

[0180] Specifically, when a visually impaired student is steadily running along the optimal running path, the system can output voice prompts to maintain direction or provide encouragement. When the system detects that the visually impaired student is gradually deviating from the optimal running path or approaching the track boundary, the system generates corresponding correction instructions based on the yaw adjustment results, such as prompting to adjust the running direction to the left or right. When an obstacle or potential safety risk is detected ahead, the system promptly outputs safety prompts related to avoidance or slowing down. Through these methods, visually impaired students can complete the running process without the need for accompaniment by relying on real-time, gentle, and understandable guidance information.

[0181] By using voice interaction to guide visually impaired students in running, a closed-loop control system is achieved, encompassing visual perception, intelligent decision-making, and human-executable guidance. This not only effectively reduces safety risks for visually impaired students during running but also enhances their autonomy, continuity, and participation experience, enabling them to independently complete running training under safe and controllable conditions.

[0182] Reference manual attached Figure 2The diagram shows a structural schematic of a running guidance system for visually impaired students based on AI vision, provided by an embodiment of the present invention.

[0183] This invention provides an AI vision-based running guidance system 20 for visually impaired students, comprising: a processor 201 and a memory 202;

[0184] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described AI vision-based method for guiding visually impaired students during running, and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0185] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0186] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0187] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0188] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0189] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0190] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0191] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0192] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0193] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0194] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0195] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for guiding visually impaired students during running based on AI vision, characterized in that, include: S1: Acquire visual data of the running environment, images of obstacle locations, and images of the location of visually impaired students; S2: Construct a track space model based on the visual data of the running environment; Specifically, S2 includes: S201: Perform temporal image stabilization and consistency enhancement operations on the visual data of the running environment to obtain stabilized and enhanced data; S202: Construct a runway temporal structure probability field based on the stabilized image enhancement data; Specifically, the runway temporal structure probability field is as follows: in, express t Real-time updates k Runway-like temporal structure probability, Indicates the temporal fusion weight coefficients. This indicates the spatial location in the runway image coordinate system. express t- Output at time 1 k Runway-like temporal structure probability, express t Output in real time k Class-instantaneous structure probability, This represents the Sigmoid function. This represents the weighting adjustment coefficient. express t Output the instantaneous probability of the runway center guide strip structure at all times. express t Output the instantaneous probability of the runway boundary structure at any given time. k Represents the structure category index, Indicates the runway boundary structure. This indicates the structure of the runway center guide strip; S203: Calculate the dual boundary consistency index based on the runway temporal structure probability field; Specifically, the dual boundary consistency index is: in, Represents the dual boundary consistency index. These represent a candidate curve for the left boundary and a candidate curve for the right boundary, located on either side of the runway centerline. and These represent the arc lengths of the left and right boundary curves, respectively. This represents the average spacing between the left and right boundary curves in the normal direction. This represents the prior value of the runway width. and These represent the average curvature of the left and right boundary curves, respectively. This represents the weighting coefficient of the runway width consistency constraint term. This represents a infinitesimal element representing the arc length along the boundary curve. The weighting coefficients represent the curvature symmetry constraint terms. express t The temporal probability field of the runway boundary at time t; S204: Based on the dual boundary consistency index, the set of boundary curve pairs in the runway time-series structure probability field is filtered to obtain a set of candidate boundary curves. S205: Perform regularized fitting on the set of candidate boundary curves to obtain a set of fitted curves; S206: Generate the centerline and runway width field based on the set of fitted curves; S207: Construct the runway spatial model by combining the centerline and the runway width field; S3: Based on the visually impaired student's position image, the obstacle's position image, and the track space model, extract the real-time positions of the visually impaired student and the obstacle in the track coordinate system to obtain the real-time positions of the visually impaired student and the obstacle. S4: Based on the track space model, the real-time location of the visually impaired student, and the real-time location of the obstacle, determine the accessibility of the running environment for the visually impaired student. S5: If the determination result is that it is passable, proceed to step S6; otherwise, execute the safety avoidance strategy. S6: Based on the real-time location of the visually impaired student and the real-time location of the obstacle, plan the running path of the visually impaired student and determine the optimal running path; S7: Calculate the deviation between the visually impaired student and the optimal running path in real time; S8: Based on the deviation state, adjust the running direction of the visually impaired student to obtain a running follow instruction; S9: Guide the visually impaired student's running process according to the running follow instruction.

2. The method for guiding visually impaired students during running based on AI vision according to claim 1, characterized in that, The visual data of the running environment specifically includes: track boundary lines, track center lines, runnable area, straight track geometry, curve geometry, track width variations, spatial distribution of static obstacles around the track and dynamic obstacles around the track.

3. The method for guiding visually impaired students during running based on AI vision according to claim 1, characterized in that, S3 specifically includes: S301: Timestamp align the location images of the visually impaired student and the location images of the obstacle to obtain a synchronization frame group; S302: Perform target parsing on the synchronization frame group to obtain a candidate set of visually impaired students and a candidate set of obstacles; S303: Based on the runway space model, perform runway consistency gating screening on the candidate set of visually impaired students and the candidate set of obstacles; S304: Extract the coordinate mapping representative points from the filtered candidate sets of visually impaired students and obstacle candidates respectively to obtain the representative points of visually impaired students and the representative points of obstacles; S305: Map the visually impaired student representative point and the obstacle representative point to the track coordinate system to obtain the initial track position of the visually impaired student and the initial track position of the obstacle; S306: Project the initial track position of the visually impaired student and the initial track position of the obstacle onto the track centerline respectively to obtain the real-time position of the visually impaired student and the real-time position of the obstacle.

4. The method for guiding visually impaired students during running based on AI vision according to claim 1, characterized in that, S4 specifically includes: S401: Construct a local decision unit based on the runway space model, the real-time position of the visually impaired student, and the real-time position of the obstacle. S402: Calculate the left boundary margin and the right boundary margin based on the width field and the lateral offset of the visually impaired student in the runway space model; S403: Calculate the geometric risk index based on the left boundary margin and the right boundary margin; S404: Calculate the spatiotemporal proximity risk index of the obstacle based on the real-time position of the obstacle and the real-time position of the visually impaired student; S405: Calculate the position error by subtraction based on the real-time position and historical position of the visually impaired student; S406: Based on the local determination unit, the position error is parametrically estimated using an uncertainty estimator to determine the pose uncertainty index; S407: Calculate the comprehensive passability index based on the geometric risk index, the obstacle spatiotemporal proximity risk index, and the pose uncertainty index; S408: Based on the comprehensive accessibility index, determine the accessibility of the running environment for the visually impaired student.

5. The method for guiding visually impaired students during running based on AI vision according to claim 4, characterized in that, Specifically, S406 includes: S4061: Based on the local decision unit, calculate the semantic uncertainty index and the structural change intensity index; S4062: Calculate the pose error based on the real-time position and the historical position of the visually impaired student; S4063: Based on the semantic uncertainty index and the structural change intensity index, the pose error is parametrically estimated using the uncertainty estimator to obtain the pose error distribution parameters; S4064: Determine the pose uncertainty index based on the pose error distribution parameters.

6. The method for guiding visually impaired students during running based on AI vision according to claim 1, characterized in that, S6 specifically includes: S601: Based on the real-time position of the visually impaired student and the real-time position of the obstacle, the velocity of the visually impaired student and the obstacle are estimated respectively to obtain the motion state quantity of the visually impaired student and the motion state quantity of the obstacle. S602: Generate a set of candidate running paths based on the center line of the track and the real-time location of the visually impaired student; S603: Calculate the minimum collision time based on the motion state of the visually impaired student and the motion state of the obstacle; S604: Determine whether the minimum collision time is greater than the time threshold; if yes, proceed to step S605; otherwise, terminate the planning of the running path for the visually impaired student and return to step S4. S605: Based on the runway centerline, the centerline reference heading sequence is calculated using the tangential direction; S606: Calculate the direction consistency cost based on the candidate path heading sequence and the centerline reference heading sequence; S607: Based on the directional consistency cost, perform preliminary sorting and filtering on the candidate running path set to obtain a subset of candidate paths; S608: Based on the runway width field and boundary constraints, the feasibility of the candidate path subset is screened through runnable area determination and boundary violation penalty to obtain a set of feasible paths; S609: Based on the set of feasible paths, set the objective function and constraints; S610: Under the constraints of the above constraints, the optimal running path is determined by solving the objective function using an optimization algorithm.

7. The method for guiding visually impaired students during running based on AI vision according to claim 1, characterized in that, Specifically, S7 includes: S701: Based on the real-time location of the visually impaired student, determine the path reference location corresponding to the visually impaired student through path nearest point matching; S702: Calculate the lateral deviation of the visually impaired student based on the real-time position of the visually impaired student and the path reference position; S703: Calculate the instantaneous heading angle of the visually impaired student based on the time sequence of the student's real-time location; S704: Calculate the heading deviation based on the optimal running path and the instantaneous heading angle; S705: Perform time-series consistency processing on the lateral deviation and the heading deviation to obtain a stable deviation estimate; S706: Combine the runway width field of the runway space model and the lateral deviation to determine the boundary approach risk index of the visually impaired student; S707: Combining the stable deviation estimate and the boundary approach risk index, calculate the deviation state between the visually impaired student and the optimal running path through state assembly.

8. The method for guiding visually impaired students during running based on AI vision according to claim 1, characterized in that, S8 specifically includes: S801: Based on the deviation state, a deviation prediction model is constructed by discretizing the small-angle error dynamics; S802: Based on the deviation prediction model, set the parameters of the optimization cost function; S803: Calculate the optimal yaw angle increment control amount based on the deviation prediction model and the optimized cost function parameters; S804: Perform dead zone and hysteresis processing on the optimal yaw angle increment control quantity to obtain the running follow command.

9. A running guidance system for visually impaired students based on AI vision, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the AI ​​vision-based running guidance method for visually impaired students as described in any one of claims 1 to 8.

Images