A method for collecting image data of a blind guide system based on edge computing

Abstract

The present invention relates to the field of image processing technology, and in particular to an image data acquisition method for a blind guide system based on edge computing, comprising: acquiring image data and depth data; dividing the image data to obtain a plurality of regions; obtaining information richness based on the information intensity of pixels in the region and the grayscale difference of adjacent regions; performing region matching on adjacent frames to obtain the corresponding region of the region in the adjacent frames; obtaining the degree of image information loss based on the information richness difference between the region and the corresponding region and the density of optical flow points; obtaining the blind guide information loss coefficient in combination with the light invariance of the depth data corresponding to the region of the image data; processing each pixel according to the blind guide information loss coefficient to obtain a preprocessed image of the image data for edge computing of the blind guide device. The present invention aims to solve the problem that when edge computing is performed on the collected image, fuzzy data occupies computing resources of edge computing, so as to reduce the amount of computing and improve the computing effectiveness.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an image data acquisition method for a blind guide system based on edge computing. Background Art

[0002] The image data collection of the guide system realized by edge computing technology has significant advantages such as strong real-time performance, fast response speed, and high data processing efficiency. It can effectively improve the intelligence level of the guide system, provide more accurate and reliable environmental perception capabilities for the visually impaired, and greatly enhance their travel safety and independence, thereby significantly improving the quality of life and promoting social care and support for the disabled.

[0003] However, due to the limited computing resources of edge computing technology, each processed image needs to contain rich and non-redundant information in order to maximize processing efficiency. However, the actual acquisition process is affected by factors such as external light interference and user movement, which often leads to insufficient information about a single object in the collected image, and the environmental perception effect obtained after processing is weak, which in turn reduces the overall monitoring capability of the system, thereby affecting the system's accurate perception of the surrounding environment and failing to provide users with reliable navigation assistance. Summary of the invention

[0004] The present invention provides an image data acquisition method for a blind guide system based on edge computing to solve the existing problems.

[0005] The present invention provides an image data collection method for a blind guide system based on edge computing, which adopts the following technical solutions:

[0006] An embodiment of the present invention provides a method for collecting image data of a blind guide system based on edge computing, the method comprising the following steps:

[0007] Obtain image data and depth data of the same frame collected by the blind guide system;

[0008] Dividing the image data to obtain a plurality of image data regions;

[0009] Based on the information intensity of the pixels in the region of the image data and the grayscale difference with the adjacent region, the information richness of the region of the image data is obtained;

[0010] Performing region matching on adjacent frames to obtain the corresponding region of the image data region in the adjacent frames, as well as the optical flow points of the image data region and the corresponding region;

[0011] The degree of image information loss of the region of the image data is obtained according to the difference in information richness between the region of the image data and the corresponding region of the adjacent frame, and the density of optical flow points in the region matching process;

[0012] Based on the image information loss degree of the image data region and the light invariance of the depth data corresponding to the image data region, a guide blind information loss coefficient of the image data region is obtained;

[0013] Each pixel point in the area of ​​the image data is processed according to the guide information loss coefficient to obtain a preprocessed image of the image data, and the preprocessed image is used for edge computing of the guide device.

[0014] Preferably, the specific method of acquiring the area of ​​the image data is:

[0015] A superpixel segmentation algorithm is used to divide each frame of image data to obtain several regions in each frame of image data.

[0016] Preferably, the specific method of obtaining the information richness is:

[0017] The texture clarity of the i-th region in the j-th frame image data is recorded as , the information richness of the i-th region of the j-th frame image data The calculation method is:

[0018]

[0019] in, is the adjacency information difference coefficient of the i-th region of the j-th frame image data, is a linear normalization function.

[0020] Preferably, the specific method of obtaining the adjacency information difference coefficient is:

[0021] The region adjacent to the i-th region in the j-th frame image data is recorded as the adjacent region of the i-th region in the j-th frame image data, and the adjacency information difference coefficient of the i-th region in the j-th frame image data is The calculation method is:

[0022]

[0023] in, is the number of adjacent regions of the i-th region of the j-th frame image data, is the first region in the jth frame image data The mean gray value of adjacent regions, is the mean gray value of the i-th region in the j-th frame image data; is the absolute value function.

[0024] Preferably, the specific method of obtaining the texture clarity is:

[0025] The gradient of each pixel in the ith region of the jth frame image data is obtained based on the Laplace operator, which is recorded as the information intensity of each pixel in the ith region of the jth frame image data;

[0026] The variance of the information intensity of all pixels in the i-th region of the j-th frame image data is recorded as the texture clarity of the i-th region in the j-th frame image data.

[0027] Preferably, the specific method of obtaining the corresponding area is:

[0028] An optical flow matching model is constructed for all the collected frame image data, and the pixel points in all the frame image data are matched to obtain the corresponding area of ​​each area in each frame image data in the previous frame image data.

[0029] Preferably, the specific method of obtaining the degree of image information loss is:

[0030] The information richness of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data is recorded as ;

[0031] The number of optical flow points in the i-th region of the j-th frame image data is recorded as , the number of optical flow points of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data is recorded as ;

[0032] The degree of image information loss in the i-th region of the j-th frame image data The calculation method is:

[0033]

[0034] in, is the number of pixels in the corresponding area of ​​the i-th area of ​​the j-th frame image data, is the number of pixels in the i-th region of the j-th frame image data, To take the absolute value function, is the information richness of the i-th region of the j-th frame image data.

[0035] Preferably, the specific method of obtaining the blind guide information loss coefficient is:

[0036] The blind guide information loss coefficient of the i-th region of the j-th frame image data The calculation method is:

[0037]

[0038] in, is the image information loss degree of the i-th region of the j-th frame image data, is the depth data variance of the i-th region of the j-th frame image data, is the depth data variance of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data; To take the absolute value function, is the linear normalization function, Preset hyperparameters.

[0039] Preferably, the specific method of obtaining the preprocessed image of the image data is:

[0040] Convolution calculation is performed according to the convolution kernel size of all pixels in all areas of the j-th frame image data to obtain output grayscale values ​​of all pixels, and the output grayscale values ​​of all pixels constitute a preprocessed image of the j-th frame image data.

[0041] Preferably, the specific method of obtaining the convolution kernel size is:

[0042] Preset global convolution kernel ;

[0043] The convolution kernel size of the nth pixel in the ith region of the jth frame image data The calculation method is:

[0044]

[0045] in, It is expressed as the guide information loss coefficient of the i-th region of the j-th frame image data.

[0046] The beneficial effects of the technical solution of the present invention are as follows: the present invention obtains image data and depth data of the same frame collected by a guide system and then divides them to obtain several image data regions; based on the information intensity of pixel points in the region of the image data and the grayscale difference with the adjacent region, the information richness of the region of the image data is obtained, reflecting the information content of each region in each frame of image data; region matching is performed on adjacent frames to obtain the corresponding region of the region of the image data in the adjacent frames, as well as the optical flow points of the region of the image data and the corresponding region; according to the difference in information richness between the region of the image data and the corresponding region of the adjacent frames, as well as the density of optical flow points during the region matching process, the degree of image information loss of the region of the image data is obtained, and the information change of the object corresponding to the same region in the image data of different frames is quantified, thereby reflecting the image data. The information richness of the image data is lost; based on the image information loss degree of the image data area and the light invariance of the depth data corresponding to the image data area, the guide information loss coefficient of the image data area is obtained, and the variability and invariance of the image data and depth data under light conditions are used to eliminate the offset of the image information loss degree caused by the movement of the object, so that the guide information loss coefficient is more accurate; each pixel in the image data area is processed according to the guide information loss coefficient to obtain a preprocessed image of the image data, and the preprocessed image is used for edge computing of the guide device, so as to preprocess the image collected by the guide device with a high-precision guide information loss coefficient, reduce the problem of low-quality image data occupying computing resources for edge computing, and improve the accuracy of edge computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 The present invention is a flowchart of the steps of a method for collecting image data of a guide system based on edge computing. DETAILED DESCRIPTION

[0049] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the image data acquisition method of the guide system based on edge computing proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0050] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0051] The following is a detailed description of a method for collecting image data for a guide system based on edge computing provided by the present invention in conjunction with the accompanying drawings.

[0052] See also Figure 1 , which shows a flowchart of a method for collecting image data of a guide system based on edge computing provided by an embodiment of the present invention, the method comprising the following steps:

[0053] Step S001: Acquire image data and depth data of the same frame collected by the blind guide system.

[0054] It should be noted that the blind guide system based on edge computing usually transmits the collected external environment data directly to its own edge computing module for real-time processing, and provides feedback to the blind based on the processed results. However, the edge computing module itself has fewer computing resources, and continuous data processing at a high frame rate will consume too many data computing resources, thereby reducing its real-time processing capabilities. Therefore, inertial sensors are usually integrated inside the device to adjust the acquisition frame rate.

[0055] Preferably, an inertial sensor is installed in the guide device, and the user's walking motion every second is collected based on the inertial sensor, and compared with the motion file in the BVH format stored in the guide system. The walking state corresponding to the motion file in the BVH format includes a stable state and a jittery motion state. After comparison, the state most similar to the motion file is used as the user's walking state every second. When the user's walking state every second is a stable state, the collection frequency is set to 20FPS. When the user's walking state every second is a jittery motion state, the collection frequency is set to 5FPS.

[0056] Furthermore, the guide device is equipped with an image acquisition sensor and a laser radar scanning device, which perform synchronous acquisition at the user's acquisition frequency of one second. The image acquisition sensor collects images of all frames within one second and obtains each frame of image data of the user after grayscale conversion; at the same time, the depth data corresponding to each frame of image data is collected, and the depth data is the distance between each pixel point in the image data and the guide device, in meters.

[0057] Step S002: Divide the image data into several regions of image data, and obtain the information richness of the region of image data based on the information intensity of the pixels in the region of image data and the grayscale difference with the adjacent region.

[0058] It should be noted that the guide system based on edge computing captures external environmental information in real time through an integrated acquisition module, and then accurately transmits the captured information to the data processing module for in-depth calculation and analysis. The calculation results are quickly transmitted to the response module to efficiently provide the navigation assistance services required by the user. However, in view of the limited computing resources inherent in the edge computing environment and the strict requirements for real-time performance in practical applications, the information density of the amount of data processed by a single calculation needs to be as high as possible. In actual operation, due to the complexity and variability of the outdoor environment and the potential interference of factors such as user movement, the image quality of each acquisition often shows significant differences. Therefore, in order to improve the data quality and the overall monitoring efficiency of the system, this embodiment uses a strategy of deep fusion of the image data acquired once with the laser radar scanning data. Thereby effectively identifying and processing the interference area in the image, thereby achieving a significant improvement in data quality.

[0059] It should be further explained that, since the acquisition device is in motion during the acquisition process, and when there are large differences in the external ambient light, the acquired image under a single frame will be blurred to varying degrees, resulting in the edges of objects in the acquired image data being relatively smooth, that is, the information density is low. Direct data processing will result in a waste of computing resources. Therefore, in order to improve the information density of the image data, this embodiment divides different areas of the blurred image data into several areas, so as to achieve different degrees of blur based on different blurred areas, and obtain the information richness of each area based on the difference between the area and the surrounding area and the expression of detailed texture.

[0060] Preferably, each frame of image data is divided using a superpixel segmentation algorithm to obtain a number of regions in each frame of image data, each region corresponding to an object in the image data.

[0061] Preferably, the information richness of each region in each frame of image data is obtained through the texture details inside each region in each frame of image data and the texture difference with adjacent regions; the specific steps are as follows:

[0062] The gradient of each pixel in the ith region of the jth frame image data is obtained based on the Laplace operator, which is recorded as the information intensity of each pixel in the ith region of the jth frame image data;

[0063] The variance of the information intensity of all pixels in the ith region of the jth frame image data is recorded as the texture clarity of the ith region in the jth frame image data. In this embodiment, the texture clarity is calculated using the Laplace operator, and the distribution range of the gradient values ​​obtained by the Laplace operator can reflect the clarity of the edges corresponding to the texture details in the image. When the texture details are blurred by motion, the gradient values ​​obtained by the Laplace operator will be reduced. When the distribution range of the gradient values ​​obtained by the Laplace operator is small, it means that the texture inside the region is not clear enough and may be affected by motion.

[0064] Furthermore, the region adjacent to the ith region in the jth frame image data is recorded as the adjacent region of the ith region in the jth frame image data, and the adjacency information difference coefficient of the ith region in the jth frame image data is The calculation method is:

[0065]

[0066] in, is the number of adjacent regions of the i-th region of the j-th frame image data, is the first region in the jth frame image data The mean gray value of adjacent regions, is the mean gray value of the i-th region in the j-th frame image data; is the absolute value function.

[0067] It represents the gray value difference between the ith region and the adjacent regions of the jth frame image data. The larger the difference, the more likely the region is to be an object rather than a background region. Therefore, the region can better reflect the image information.

[0068] Furthermore, the texture clarity of the i-th region in the j-th frame image data is recorded as , the information richness of the i-th region of the j-th frame image data The calculation method is:

[0069]

[0070] in, is the adjacency information difference coefficient of the i-th region of the j-th frame image data, is a linear normalization function.

[0071] Step S003, performing region matching on adjacent frames to obtain the corresponding region of the image data region in the adjacent frames, and obtaining the degree of image information loss of the image data region based on the difference in information richness between the image data region and the corresponding region of the adjacent frames, and the density of optical flow points in the region matching process, and obtaining the guide blind information loss coefficient of the image data region based on the degree of image information loss of the image data region and the light invariance of the depth data corresponding to the image data region.

[0072] It should be noted that the process of collecting image data by the guide system is in a dynamic scene. Then the area in the current frame image data may be affected by the motion state in other frames, resulting in motion blur of the edge and content of the captured object, causing loss of image information in the area corresponding to the object, and the blurring of the edges will cause deformation of the object in the image data. Therefore, this embodiment first needs to obtain the corresponding areas of each area in continuous frames through the optical flow matching model, so as to evaluate the image loss of each image data through the detailed texture differences of the corresponding areas after matching in adjacent frames.

[0073] Preferably, an optical flow matching model is constructed for all the collected frame image data, and the pixel points in all the frame image data are matched to obtain the corresponding area of ​​each area in each frame image data in the previous frame image data, as well as the number of matching optical flow points of the area in each frame image data and the number of matching optical flow points of the corresponding area. It should be noted that the optical flow matching model described in this embodiment is a well-known model obtained based on the optical flow method, and specifically matches the mean grayscale value, group characteristics, and regional characteristics of the pixel points in the neighborhood of each pixel point. This embodiment will not repeat the optical flow matching model.

[0074] It should be noted that this embodiment only analyzes the areas that can be matched. If a region of the current frame image data cannot obtain a corresponding region in the previous frame image data, it means that the region may not be the region corresponding to the object or has disappeared in the image data acquisition range of the guide system. Therefore, when evaluating the image loss of each image data, the area that cannot be matched is not used as an analysis reference.

[0075] It should be noted that, under adjacent frame image data, when the information richness difference between a region and the corresponding region of other frame image data is too large, it means that the degree of information loss within the current region is greater. Therefore, this embodiment combines the regional matching optical flow points under adjacent frames and the regional differences of the laser radar scanning data under adjacent frames to obtain the guide information loss coefficient in a single area.

[0076] Preferably, based on the information richness difference of the corresponding area in each frame of image data in the previous frame of image data and the optical flow point density difference between the area and the corresponding area, the image information loss degree of each area in each frame of image data is obtained, and the specific operation is:

[0077] Get the information richness of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data, denoted as ;

[0078] The number of optical flow points in the i-th region of the j-th frame image data is recorded as , the number of optical flow points of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data is recorded as ;

[0079] The degree of image information loss in the i-th region of the j-th frame image data The calculation method is:

[0080]

[0081] in, is the number of pixels in the corresponding area of ​​the i-th area of ​​the j-th frame image data, is the number of pixels in the i-th region of the j-th frame image data, is the absolute value function.

[0082] is the density of the optical flow point of the i-th region of the j-th frame image data relative to the i-th region, then is the density of optical flow points in adjacent frames. The smaller the density, the more information loss of the area corresponding to the same object in the j-th frame image data and the j-1-th frame image data, and the fewer the number of optical flow points belonging to the feature points, the greater the degree of image information loss; It is the difference in information richness in adjacent frames. The larger the value, the greater the change in information richness of image data in adjacent frames. It is more likely that the image is affected by motion blur and causes loss of image information, so the degree of image information loss is greater.

[0083] It should be noted that the evaluation is performed based on the difference in depth data between the current frame image data area and the laser radar scanning points in its corresponding area, and the depth data can reflect the distance relationship between the current guide device and the corresponding objects in each area. Since the depth data is less or not affected by light, that is, the depth data is invariant, then when the depth data of the existing laser radar scanning points changes less overall in the adjacent frame image data, it means that the guide information loss of this area in the current frame image data should consider the degree of image information loss. Therefore, based on the above logic, the guide information loss coefficient of each area in each frame image is obtained.

[0084] The blind guide information loss coefficient of the i-th region of the j-th frame image data The calculation method is:

[0085]

[0086] in, is the image information loss degree of the i-th region of the j-th frame image data, is the depth data variance of the i-th region of the j-th frame image data, is the depth data variance of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data; To take the absolute value function, is the linear normalization function, This is a preset hyperparameter used to avoid the denominator being 0. The hyperparameter set in this embodiment is described using 1 as an example.

[0087] This embodiment uses the variance of the depth data to describe the changing relationship between the current position of the guide device and the position of the object corresponding to the ith region. Since different positions are at different distances from the guide device relative to the real object, and in different video frames, the movement of the object and the change in the position of the guide device will cause the collected depth data to shift, and the variance describes the distribution range of the depth data. When the position of the guide device and the object shifts, the object moves as a whole. The smaller the difference value, the more likely the object is to be a linear displacement compared to the guide device, that is, the change in the image data displayed to the guide device is more useful for analyzing the degree of information loss. Therefore, when The smaller the value, the more relevant the image information loss degree is to the guide information loss coefficient.

[0088] Step S004: Process each pixel in the area of ​​the image data according to the blind guide information loss coefficient to obtain a preprocessed image of the image data, and the preprocessed image is used for edge computing of the blind guide device.

[0089] It should be noted that when significant information loss is found in a single area of ​​the image data, it means that the pixel values ​​in this area have suffered relatively serious interference, which in turn has led to the loss of a large number of details. A differentiated strategy needs to be adopted, that is, using convolution kernels of different specifications to perform adaptive dynamic blur adjustment on the affected areas. That is, for areas with a larger guide information loss coefficient, it is necessary to increase the size of the convolution kernel or expand the receptive field so as to capture a wider range of contextual information, thereby further reducing the information inside it, thereby performing more effective compensation and reducing the difficulty of data processing in edge computing; for areas with a smaller guide information loss coefficient, that is, clearer areas, reduce the size of the convolution kernel or reduce the receptive field to ensure that the local detail features can be more finely retained and enhanced.

[0090] Preferably, in order to better restore the edge details and internal texture of the original image, a blind area convolution technique is used to perform region-based detail restoration. The blind area convolution technique is an existing well-known technique and will not be described in detail in this embodiment.

[0091] Preset global convolution kernel The size of the global convolution kernel is preset to 5*5. It should be noted that this embodiment does not specifically limit the size of the global convolution kernel, and the preset size is only used as a reference for this embodiment.

[0092] For the i-th region of the j-th frame image data, the input channel of the current RGB image is 3, and the output channel of the blind area convolution technology is 1.

[0093] The convolution kernel size of the nth pixel in the ith region of the jth frame image data The calculation method is:

[0094]

[0095] in, It is expressed as the guide information loss coefficient of the i-th region of the j-th frame image data.

[0096] Furthermore, based on the convolution kernel size of the n-th pixel in the ith area of ​​the j-th frame image data, a blind area convolution technique is used to perform a convolution calculation on the n-th pixel in the ith area of ​​the j-th frame image data to obtain an output grayscale value of the n-th pixel in the ith area of ​​the j-th frame image data. The blind area convolution technique and the convolution calculation are existing public technologies and will not be described in detail in this embodiment.

[0097] Convolution calculation is performed according to the convolution kernel size of all pixels in all areas of the j-th frame image data to obtain the output grayscale values ​​of all pixels. The output grayscale values ​​of all pixels constitute the preprocessed image of the j-th frame image data. The preprocessed image is input into the edge computing module of the guide device for real-time processing, and feedback is given to the blind based on the processed results.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for collecting image data of a blind guide system based on edge computing, characterized in that: The method comprises the following steps: Obtain image data and depth data of the same frame collected by the blind guide system; Dividing the image data to obtain a plurality of image data regions; Based on the information intensity of the pixels in the region of the image data and the grayscale difference with the adjacent region, the information richness of the region of the image data is obtained; Performing region matching on adjacent frames to obtain the corresponding region of the image data region in the adjacent frames, as well as the optical flow points of the image data region and the corresponding region; The degree of image information loss of the region of the image data is obtained according to the difference in information richness between the region of the image data and the corresponding region of the adjacent frame, and the density of optical flow points in the region matching process; Based on the image information loss degree of the image data region and the light invariance of the depth data corresponding to the image data region, a guide blind information loss coefficient of the image data region is obtained; Process each pixel point in the area of ​​the image data according to the blind guide information loss coefficient to obtain a preprocessed image of the image data, and the preprocessed image is used for edge computing of the blind guide device; The method for obtaining the information intensity is as follows: obtaining the gradient of each pixel point in the ith region of the jth frame image data based on the Laplace operator, recorded as the information intensity of each pixel point in the ith region of the jth frame image data; The specific method of obtaining the texture clarity is as follows: the variance of the information intensity of all pixels in the i-th region of the j-th frame image data is recorded as the texture clarity of the i-th region in the j-th frame image data; Among them, the specific way to obtain information richness is: The texture clarity of the i-th region in the j-th frame image data is recorded as , the information richness of the i-th region of the j-th frame image data The calculation method is: in, is the adjacency information difference coefficient of the i-th region of the j-th frame image data, is a linear normalization function; The specific method of obtaining the adjacency information difference coefficient is as follows: The region adjacent to the i-th region in the j-th frame image data is recorded as the adjacent region of the i-th region in the j-th frame image data, and the adjacency information difference coefficient of the i-th region in the j-th frame image data is The calculation method is: in, is the number of adjacent regions of the i-th region of the j-th frame image data, is the first region in the jth frame image data The mean gray value of adjacent regions, is the mean gray value of the i-th region in the j-th frame image data; is the absolute value function; The specific method for obtaining the degree of image information loss is as follows: The information richness of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data is recorded as ; The number of optical flow points in the i-th region of the j-th frame image data is recorded as , the number of optical flow points of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data is recorded as ; The degree of image information loss in the i-th region of the j-th frame image data The calculation method is: in, is the number of pixels in the corresponding area of ​​the i-th area of ​​the j-th frame image data, is the number of pixels in the i-th region of the j-th frame image data, To take the absolute value function, is the information richness of the i-th region of the j-th frame image data; The specific method for obtaining the blind guide information loss coefficient is as follows: The blind guide information loss coefficient of the i-th region of the j-th frame image data The calculation method is: in, is the image information loss degree of the i-th region of the j-th frame image data, is the depth data variance of the i-th region of the j-th frame image data, is the depth data variance of the corresponding area of ​​the i-th area of ​​the j-th frame image data in the j-1-th frame image data; To take the absolute value function, is the linear normalization function, To preset hyperparameters; Among them, the specific method of obtaining the preprocessed image of the image data is: adjusting the convolution kernel size corresponding to each pixel point by the guide information loss coefficient, performing convolution calculation on all pixel points in all areas of the image data through the adjusted convolution kernel size, and obtaining the output grayscale values ​​of all pixel points. The output grayscale values ​​of all pixel points constitute the preprocessed image of the image data.

2. According to the method for collecting image data of a blind guide system based on edge computing in claim 1, it is characterized in that: The specific method of obtaining the area of ​​the image data is: A superpixel segmentation algorithm is used to divide each frame of image data to obtain several regions in each frame of image data.

3. According to the method for collecting image data of a blind guide system based on edge computing in claim 1, it is characterized in that: The specific method of obtaining the corresponding area is: An optical flow matching model is constructed for all the collected frame image data, and the pixel points in all the frame image data are matched to obtain the corresponding area of ​​each area in each frame image data in the previous frame image data, wherein the optical flow matching model is a model obtained based on the optical flow method.

4. According to the method for collecting image data of a blind guide system based on edge computing in claim 1, it is characterized in that: The specific method of obtaining the convolution kernel size is: Preset global convolution kernel ; The convolution kernel size of the nth pixel in the ith region of the jth frame image data The calculation method is: in, It is expressed as the guide information loss coefficient of the i-th region of the j-th frame image data.

Images