Intelligent blind person audio walking stick control method based on image recognition

Abstract

The invention discloses an intelligent blind person audio walking stick control method based on image recognition, and relates to the field of intelligent equipment control. According to the method, ground type classification is carried out on each frame of an image sequence, a passable area is verified, and the passable ground coverage rate is calculated; the actual ponding area is converted by marking a ponding area and screening stable ponding and combining camera parameters; the method comprises the following steps: establishing a mapping relation between pixels and an actual elevation through parallax calculation, taking the average elevation of a passable ground as a reference, counting a ratio of a flatness related deviation value to an over-deviation region, and obtaining a ground flatness deviation value after comprehensive analysis; performing comprehensive processing on the passable ground coverage rate, the ponding area and the ground flatness deviation value to obtain a terrain index; the problems that an existing blind person auxiliary device cannot quantify the passable range, cannot distinguish stable accumulated water and interference information and cannot physically evaluate the ground flatness, so that passing misjudgment, accumulated water missing and false alarm and the like are caused are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent device control, and in particular to an intelligent audio cane control method for the blind based on image recognition. Background Technology

[0002] As blind people's demand for independent and safe travel increases, the safety, authenticity, and controllability of the ground environment have become factors restricting their travel. Existing assistive devices for the blind lack the ability to reliably assess the ground, leading to frequent safety risks such as accidentally entering impassable areas, slipping on puddles, and tripping over protrusions. These risks seriously affect their autonomy and efficiency in traveling and can easily cause personal injury.

[0003] However, existing image recognition-based intelligent audio cane control methods for the blind still have the following problems: First, existing assistive devices cannot quantitatively calculate the coverage of passable ground, resulting in misjudgment of the passable range. Traditional white canes rely on tactile feedback from the tip of the cane, which can only sense the current obstacle underfoot and cannot predict the proportion of passable area ahead. If there are obstacles such as construction barriers ahead, they can only be noticed when they reach the edge, which can easily lead to collisions. Furthermore, existing auxiliary equipment lacks a data credibility quantification and evaluation mechanism and has not established a system similar to confidence estimation, which cannot guarantee the reliability of data evaluation results. For example, it cannot evaluate the stability and validity of data through inter-frame standard deviation, effective frame ratio, etc., resulting in the reliability of ground evaluation results being unreliable.

[0004] To address this, an intelligent audio cane control method for the blind based on image recognition has been developed. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent audio cane control method for the blind based on image recognition, in order to solve the problems mentioned in the background art.

[0006] The objective of this invention can be achieved through the following technical solution: a smart audio cane control method for the blind based on image recognition, comprising: Active triggering: The user initiates an image acquisition command independently through the smart cane button, continuously acquiring images at a fixed rate to form an image sequence; Image analysis: For each image in the image sequence, ground information is extracted, including navigable ground coverage, water accumulation area, and ground flatness deviation. The ground information of each image is then comprehensively analyzed to obtain the corresponding landform index for the image sequence. Confidence assessment: A preset optimal landform index threshold is used as a reference. The landform index corresponding to the current image sequence is compared with the optimal landform index threshold. If it is greater than the optimal landform index threshold, the normal travel path is executed. If the value is below the optimal landform index threshold, a confidence estimate of the corresponding image sequence is obtained by extracting the inter-frame standard deviation and the effective frame ratio for comprehensive analysis. Audio prompts: The confidence level is determined based on the confidence estimate, and different audio prompts are executed according to different confidence levels.

[0007] In some embodiments, obtaining the passable ground coverage corresponding to the image sequence specifically involves: By combining a semantic recognition model, each frame of the image is classified into ground types to determine passable and impassable areas. Comparing the annotation results of three consecutive frames, pixel areas that are all determined to be passable are retained; the passable ground coverage rate of a single frame is the ratio of the number of passable ground pixels in that frame to the total number of pixels in the ground observation area. The average coverage rate of five consecutive frames in the image sequence is taken as the passable ground coverage rate of the sequence.

[0008] In some embodiments, obtaining the water accumulation area corresponding to the image sequence specifically involves: Pixel-level annotations were made for the water accumulation areas in each frame of the image; Extract the contour data of the water accumulation area labeled in 3 consecutive frames, including contour coordinates and pixel distribution. Calculate the intersection-union ratio, moment distance, and centroid Euclidean distance of the contours of adjacent frames. Count the number of contour breakpoints in a single frame. Preset the above index thresholds. If the intersection-union ratio of adjacent frames is higher than the threshold, and the moment distance, centroid Euclidean distance, and number of breakpoints are all lower than the threshold, it is determined to be stable water accumulation. Otherwise, it is non-water accumulation interference. The number of stable water accumulation pixels in each frame is counted, and the actual water accumulation area in a single frame is calculated by combining the camera parameters. The average area of ​​three consecutive frames in the image sequence is taken as the water accumulation area of ​​the sequence.

[0009] In some embodiments, obtaining the ground flatness deviation value corresponding to the image sequence specifically involves: Parallax calculation is performed on the preprocessed image sequence to establish a mapping relationship between pixels and actual elevation, and the actual elevation value of each pixel on the ground is obtained. Within the ground observation area of ​​the image sequence, select a passable ground area and calculate the average elevation of all pixels within that area as the baseline elevation; Iterate through all pixels in the ground observation area of ​​each frame image, calculate the absolute value of the difference between the actual elevation and the reference elevation of a single pixel, and take the maximum value as the ground flatness deviation value of a single frame; count the proportion of pixel areas whose absolute elevation deviation value is higher than a set threshold; take the average of the maximum flatness deviation value and the proportion of out-of-tolerance areas of 4 consecutive frames in the image sequence as the ground flatness deviation value corresponding to the image sequence. In some embodiments, obtaining the land index specifically involves: The passable ground coverage, water accumulation area, and ground flatness deviation value are normalized, and the water accumulation area and flatness deviation value are positively converted into the degree of no water accumulation and the degree of excellent flatness. Extend three line segments with an included angle of 120° from the origin of the coordinate system. Connect the endpoints of the line segments with these three indices as their lengths to form a closed triangle and obtain a three-dimensional evaluation map of the ground information. Calculate the actual area of ​​the triangle based on the normalized three indices, which is the landform index of the image sequence.

[0010] In some embodiments, obtaining the average standard deviation between frames of the image sequence specifically involves: Extract the passable ground coverage, water-free status, and flatness of the corresponding image sequence; calculate the standard deviation of the single value of consecutive frames within the sequence, and take the average of the three standard deviations as the average standard deviation between frames of the image sequence.

[0011] In some embodiments, obtaining the effective frame ratio and confidence estimate of the image sequence specifically involves: The total number of frames acquired in the image sequence is identified. After removing invalid frames, the number of valid frames is obtained. The ratio of the number of valid frames to the total number of frames is calculated to obtain the percentage of valid frames in the corresponding image sequence. After normalizing the inter-frame average standard deviation and effective frame ratio of the image sequence, the results are multiplied by the corresponding preset weight factors and then summed to obtain the confidence estimate of the corresponding image sequence.

[0012] In some embodiments, the execution of different audio prompts based on the comparison results specifically includes: When the confidence estimate is greater than the threshold, the system indicates that the current ground data confidence estimate is higher than the threshold, the data is highly reliable, and you should replan your route. When the confidence estimate is less than the threshold, the system will indicate that the current ground data confidence estimate is lower than the threshold and the data reliability is low. Please re-acquire images and then determine whether to proceed.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention classifies ground types for each frame of an image sequence, verifies passable areas, and calculates passable ground coverage. It marks waterlogged areas, filters stable waterlogged areas, and calculates the actual waterlogged area using camera parameters. It establishes a mapping relationship between pixels and actual elevation through parallax calculation, using the average elevation of passable ground as a benchmark, and statistically analyzes the flatness-related deviation values ​​and the proportion of areas exceeding the deviation. After comprehensive analysis, it obtains the ground flatness deviation value. Finally, it comprehensively processes the passable ground coverage, waterlogged area, and ground flatness deviation value to obtain a ground condition index. This invention solves the problems of existing assistive devices for the blind, such as the inability to quantify passable areas, distinguish stable waterlogged areas from interference information, and physically assess ground flatness, leading to misjudgments of passability and missed or false reports of waterlogging.

[0014] This invention determines areas below the baseline area threshold by comparing the landform index with a corresponding preset baseline area threshold; it obtains the inter-frame average standard deviation by extracting the standard deviations of continuous frames for passable ground coverage, water-free conditions, and flatness, and averaging them; it identifies the total number of frames in the image sequence and calculates the percentage of valid frames by removing invalid frames; and it obtains a confidence estimate after comprehensive analysis. This invention solves the problems of existing auxiliary equipment lacking data credibility assessment and having unreliable risk judgments due to vague prompts. Attached Figure Description

[0015] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0016] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0017] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0018] Please see Figure 1 As shown, the image recognition-based intelligent audio cane control method for the blind includes: Active triggering: When a user encounters an unknown area or subjectively wants to understand the environment and ground conditions ahead, the user can initiate an image acquisition command autonomously through the smart cane button; It should be noted that the smart cane hardware uses anti-slip Braille collection buttons and a noise-canceling voice unit to ensure the reliability of the device triggering in various environments; at the same time, the multi-layer anti-mistouch design ensures that blind people can clearly perceive the triggering status and avoid misoperation, providing a foundation for subsequent image acquisition. Image acquisition and preprocessing: After the trigger command is executed, images are continuously acquired at a fixed rate through the camera integrated into the smart cane to form an image sequence; The image sequence is preprocessed, including but not limited to noise reduction, illumination compensation, and contour enhancement. At the same time, motion blur compensation technology is combined to correct the blur caused by fast-moving objects, ensuring that the contours and position information of moving objects in each frame are clearly distinguishable. Noise reduction targets sensor noise, environmental particles, and motion artifacts that may exist in the image sequence. It uses a spatiotemporal joint filtering algorithm to take the median gray value of the same pixel position within three consecutive frames, effectively suppressing instantaneous noise while preserving the edge details of moving objects. Illumination compensation addresses potential abrupt changes in illumination within image sequences by employing inter-frame brightness analysis for adaptive correction and dynamically adjusting pixel grayscale values ​​through gamma correction. For areas with uneven local illumination, the Retinex algorithm is used to separate the illumination component from the reflection component, enhancing details in dark areas and ensuring the consistency of grayscale features for the same object across different illumination frames. Contour enhancement performs threshold segmentation on the noise-reduced and illumination-compensated image to distinguish between foreground and background; optimizes region morphology through morphological operations; extracts contours using the Canny edge detection algorithm and completes them by combining contour features from adjacent frames to form continuous and complete target contours. Image analysis: For each image in the image sequence, ground information is extracted, including navigable ground coverage, water accumulation area, and ground flatness deviation. The ground information of each image is then comprehensively analyzed to obtain the corresponding landform index for the image sequence. For ground information, combined with a semantic recognition model, each frame of the image sequence is classified into ground types to determine passable ground and impassable areas (such as paved roads) and areas with obstacles and water accumulation. The ground area annotation results of three consecutive frames are compared, and the pixel areas that are all determined to be passable ground in all three frames are retained. The passable ground coverage rate of a single frame is calculated by the ratio of the total number of passable ground pixels in each frame to the total number of ground observation area pixels in the image. The average coverage rate of five consecutive frames in the image sequence is taken as the passable ground coverage rate corresponding to the image sequence. In addition, the passable ground coverage rate quantifies the passability of the ground in the corresponding area of ​​the image sequence. It is verified by three consecutive frames to filter out interference factors, reflecting the stability of the passable area and avoiding misjudgment of the feasibility of the path due to accidental factors. The higher the passable ground coverage rate, the greater the proportion of the ground in the area that is safe for users to pass through, and the lower the passability risk. Based on the image after illumination compensation to eliminate light and dark interference, the water accumulation area in each frame is marked at the pixel level by taking advantage of the low saturation and high reflectivity of the water accumulation. Extract the contour data of the water accumulation area obtained by previous annotation from three consecutive frames of images, including contour coordinates and pixel distribution. Calculate the intersection-union ratio and moment distance of the contours of two adjacent frames. Solve for the centroid coordinates of the two frames and calculate the Euclidean distance. Count the number of breakpoints in a single frame contour. The cross-over ratio, moment distance, centroid Euclidean distance, and single-frame contour breakpoint thresholds are preset respectively. If the cross-over ratio of the contours of adjacent frames is higher than the threshold, the moment distance is lower than the threshold, the centroid Euclidean distance is lower than the threshold, and the single-frame contour breakpoint is lower than the threshold, then it is determined to be stable water accumulation; otherwise, it is determined to be non-water accumulation interference. The number of pixels in the stable water accumulation area in each frame is counted, and the actual water accumulation area in a single frame is calculated based on the camera parameters; the average water accumulation area of ​​3 consecutive frames in the image sequence is calculated as the water accumulation area corresponding to the image sequence. In addition, if the outline of a certain area remains stable in consecutive frames and there is no obvious positional movement (for example, water accumulation on a road does not move with the image frames), it is judged as stable water accumulation. If the outline of the area is broken in consecutive frames, or moves with the image frame (e.g., the reflection on the water surface sways due to a vehicle passing by, or the shadow moves with the light), it is determined to be non-water accumulation interference. The water accumulation area can quantify the scale and distribution intensity of real and stable water accumulation in the corresponding area of ​​the image sequence. By eliminating the interference of light and dark and the shape changes through illumination compensation, stable water accumulation is screened out. The interference factors of uneven illumination and temporary reflective debris are eliminated, ensuring that the water accumulation area reflects the real water accumulation situation that continuously affects the ground condition, rather than instantaneous or false interference. Parallax calculation is performed on the preprocessed image sequence to establish a mapping relationship between pixels and actual elevation, and the actual elevation value of each pixel on the ground is obtained. Within the ground observation area of ​​the image sequence, select a passable ground area and calculate the average elevation of all pixels within that area as the baseline elevation; Traverse all pixels in the ground observation area of ​​each frame image, calculate the absolute value of the difference between the actual elevation and the reference elevation of a single pixel, and take the maximum value as the ground flatness deviation value of a single frame; count the proportion of pixel areas whose absolute elevation deviation value is higher than a set threshold; the set threshold is 2cm; take the average of the maximum flatness deviation value and the proportion of out-of-tolerance areas of 4 consecutive frames in the image sequence as the ground flatness deviation value corresponding to the image sequence. To further explain, the ground flatness deviation reflects the actual impact of ground elevation fluctuations on the safety of blind people's passage. By calculating parallax, pixel information is converted into physical elevation, and the average elevation of passable ground is used as a benchmark to ensure that the data fits the actual passage scenario of blind people and avoids visual judgment errors. The maximum flatness deviation can directly identify extreme defects such as ground protrusions and depressions that are easy to trip over, while the average percentage of out-of-tolerance areas reflects the prevalence of unevenness problems. After normalizing the passable ground coverage, water accumulation area, and ground flatness deviation values ​​in the ground information, the water accumulation area and ground flatness deviation values ​​are positively converted to obtain the degree of no water accumulation and the degree of good flatness. The normalization process is to normalize the passable ground coverage, water accumulation area, and ground flatness deviation values ​​to the same numerical range. To clarify, the positive conversion involves transforming the water accumulation area and the ground flatness deviation value, converting the two negative indicators—where a larger value indicates a worse ground condition—into positive indicators where a larger value indicates a better ground condition; this aligns with the direction of the passable ground coverage rate. Starting from the origin of the coordinate system, three straight line segments are extended, with the included angle between the three straight line segments being 120°. The extension lengths of the three straight line segments are respectively determined by the passability coverage, water-free status, and flatness of the ground information. After the three straight line segments are extended, the extension endpoints of the three straight line segments are connected in sequence to form a closed triangle, which is used as a three-dimensional evaluation map of the ground information. Based on the normalized passable ground coverage, water-free status, and flatness, the actual area of ​​the triangle corresponding to the current image sequence is calculated using an area formula adapted to the 120° angle characteristic, which is the landform index. In addition, the 3D assessment map transforms the three separate ground indicators—accessibility, water accumulation hazard, and tripping risk—into a visual and quantifiable overall ground condition assessment vehicle. The lengths of the three line segments visually represent the strengths and weaknesses of each indicator, while the 120° equal angle ensures fair weighting of the three indicators. The area of ​​the triangle serves as the key link between indicator data and risk level; the closer the area is to the baseline optimal value, the better the overall ground condition, and vice versa. Confidence assessment: Using the preset optimal landform index threshold as a reference, the landform index corresponding to the current image sequence is compared with the optimal landform index threshold; if it is greater than the optimal landform index threshold, the normal travel path is executed. If the value is below the optimal landform index threshold, a confidence estimate of the corresponding image sequence is obtained by extracting the inter-frame standard deviation and the effective frame ratio for comprehensive analysis. Specifically: Extract the passable ground coverage, water-free status, and flatness of the corresponding image sequence; calculate the standard deviation of the single value of consecutive frames within the sequence, and take the average of the three standard deviations as the average standard deviation between frames of the image sequence. In addition, the average standard deviation between frames can measure the stability of ground indicators such as passable ground coverage, water-free status, and flatness in the image sequence, avoiding misjudgments caused by fluctuations in indicators in a single frame; the smaller the value, the smoother the change of indicators within consecutive frames, while the larger the value, the more drastic the fluctuation of indicators between frames, which may be due to sensor noise or temporary occlusion causing data distortion, and the reliability of the indicators should be carefully considered. The total number of frames acquired in the image sequence is identified. After removing invalid frames, the number of valid frames is obtained. The ratio of the number of valid frames to the total number of frames is calculated to obtain the percentage of valid frames in the corresponding image sequence. In addition, invalid frames refer to image frames where the image is blurred and the ground area cannot be identified, where sensor noise causes the indicator deviation to exceed the preset threshold (such as the difference between the passable coverage of a single frame and the adjacent frame > 20%), and where sudden changes in illumination cause misjudgment of the ground type. The effective frame ratio can assess the proportion of usable data in an image sequence and eliminate the interference of invalid frames on the calculation of indicators. The higher the ratio, the more clearly the ground area can be identified in most of the collected frames and the indicators have no abnormal deviations, reflecting a good image acquisition environment. The lower the ratio, the more unusable a large number of frames are due to quality issues, which may be due to harsh environment or temporary equipment failure, resulting in a lack of sufficient reliable data to support the calculation of indicators. After normalizing the inter-frame average standard deviation and effective frame ratio of the image sequence, the results are multiplied by the corresponding preset weight factors and then summed to obtain the confidence estimate of the corresponding image sequence. In addition, the confidence estimate can quantify the overall credibility of the process from the calculation results of ground indicators to the determination of risk level; the higher the estimate, the more stable the inter-frame indicators are and the more effective the frames are, reflecting the reliability of the current risk level determination; the lower the estimate, the more stable the indicators are or the less effective the data is, and the current risk level may be misjudged. Audio prompts: A preset confidence threshold is set, and the confidence value of the corresponding image sequence is compared with the confidence threshold; different audio prompts are executed based on the comparison results. Specifically: When the confidence estimate is greater than the threshold, the current ground data confidence estimate is higher than the threshold, the data is highly reliable, please replan the route; When the confidence estimate is less than the threshold, the current ground data confidence estimate is lower than the threshold, the data credibility is low, please re-collect images and then determine whether it is passable; The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for controlling an intelligent blind audio cane based on image recognition, characterized in that, The method comprises the following steps: Active triggering: the user initiates an image acquisition instruction through a smart walking stick, and continuously acquires images at a fixed rate to form an image sequence; Image analysis: ground information is extracted from each image in the image sequence, wherein the ground information includes passable ground coverage, water area, and ground flatness deviation value; and the ground information of each image is analyzed to obtain a ground state index corresponding to the image sequence; Confidence evaluation: an optimal ground state index threshold is preset, and the ground state index corresponding to the current image sequence is compared with the optimal ground state index threshold; if the ground state index is greater than the optimal ground state index threshold, a normal travel path is executed; If the ground state index is lower than the optimal ground state index threshold, a confidence value corresponding to the image sequence is obtained through comprehensive analysis of the inter-frame standard deviation and the effective frame proportion; Audio prompt: an evaluation threshold corresponding to the confidence value is preset, and the confidence value corresponding to the image sequence is compared with the evaluation threshold, and different audio prompts are executed based on the comparison result. 2.The image recognition based intelligent blind audio walking stick control method according to claim 1, characterized in that, The passable ground coverage corresponding to the image sequence is obtained in the following manner: The semantic recognition model is combined to classify the ground type of each frame of image, and the passable ground and the impassable area are determined; The pixel regions determined as passable in three consecutive frames are retained; the single-frame passable ground coverage is the ratio of the number of passable ground pixels in the frame to the total ground observation area pixel number; the average value of the coverage of five consecutive frames in the image sequence is taken as the passable ground coverage of the sequence. 3.The image recognition based intelligent blind audio walking stick control method according to claim 1, characterized in that, The water area corresponding to the image sequence is obtained in the following manner: The water area of each frame of image is pixel-level labeled; The contour data of the labeled water area of three consecutive frames, including contour coordinates and pixel distribution, are extracted; the intersection-over-union, moment distance and centroid Euclidean distance of adjacent frames are calculated; and the number of single-frame contour break points is counted; The threshold values of the above indexes are preset; if the intersection-over-union of adjacent frames is higher than the threshold value, the moment distance, the centroid Euclidean distance and the number of break points are all less than the threshold value, it is determined that the water is stable, otherwise it is determined that the water is not stable; The number of stable water pixels of each frame is counted; the actual water area of a single frame is converted according to the camera parameters; and the average value of the areas of three consecutive frames in the image sequence is taken as the water area of the sequence. 4.The image recognition based intelligent blind audio walking stick control method according to claim 1, characterized in that, The ground flatness deviation value corresponding to the image sequence is obtained in the following manner: The disparity of the preprocessed image sequence is calculated to establish a pixel-to-actual elevation mapping relationship, and the actual elevation value of each pixel point of the ground is obtained; In the ground observation area of the image sequence, the passable ground area is selected, and the average elevation of all pixel points in the area is calculated as the reference elevation; The absolute value of the actual elevation of a single pixel point minus the reference elevation is calculated by traversing all pixel points of the ground observation area in each frame of image; the maximum value is taken as the single-frame ground flatness deviation value; the proportion of the pixel region with an elevation deviation absolute value higher than a set threshold value is counted; and the maximum value of the flatness deviation values of four consecutive frames in the image sequence and the average value of the out-of-tolerance region proportions are taken as the ground flatness deviation value corresponding to the image sequence. 5.The image recognition based intelligent blind audio walking stick control method according to claim 1, characterized in that, The ground state index is obtained in the following manner: The passable ground coverage, the water accumulation area and the ground flatness deviation value are normalized, and the water accumulation area and the flatness deviation value are positively converted into the water-free degree and the flatness good degree. Three line segments with an included angle of 120° are extended from the origin of the coordinate system, and a closed triangle is formed by connecting the end points of the line segments to obtain a three-dimensional evaluation map of the ground information. 6.The image recognition based intelligent blind audio walking stick control method according to claim 1, characterized in that, The average standard deviation between frames of the image sequence is obtained, specifically as follows: The passable ground coverage, the water-free degree and the flatness good degree corresponding to the image sequence are extracted, and the standard deviations of the single values of the consecutive frames in the sequence are calculated to take the average of the three standard deviations as the average standard deviation between frames of the image sequence. 7.The image recognition based intelligent blind audio walking stick control method according to claim 6, characterized in that, The effective frame proportion and the confidence estimate value of the image sequence are obtained, specifically as follows: The total number of frames collected by the image sequence is identified, and the effective frame number is obtained after eliminating the invalid frames. The effective frame proportion of the corresponding image sequence is calculated by comparing the effective frame number with the total frame number. After the average standard deviation between frames of the image sequence and the effective frame proportion are normalized, the confidence estimate value of the corresponding image sequence is obtained by multiplying the normalized values by the corresponding preset weight factors and summing them up. 8.The image recognition based intelligent blind audio walking stick control method according to claim 1, characterized in that, The different audio prompts are executed based on the comparison result, specifically as follows: When the confidence estimate value is greater than the threshold value, it is prompted that the current ground data confidence estimate value is higher than the threshold value, the data reliability is high, and the path needs to be re-planned. When the confidence estimate value is less than the threshold value, it is prompted that the current ground data confidence estimate value is lower than the threshold value, the data reliability is low, and the image needs to be re-collected before determining whether it is passable.

Images