Search

CN-122018491-A - Intelligent wheelchair self-adaptive positioning and obstacle avoidance method and system based on multi-sensor fusion

CN122018491ACN 122018491 ACN122018491 ACN 122018491ACN-122018491-A

Abstract

The invention relates to the technical field of image recognition, in particular to an intelligent wheelchair self-adaptive positioning and obstacle avoidance method and system based on multi-sensor fusion, comprising the following steps: and acquiring a forward-looking road navigation image according to the intelligent wheelchair vision sensor, intercepting a driving region of interest under the forward-looking road navigation image according to the intelligent wheelchair chassis width parameter, and extracting a pixel gray level distribution matrix in the driving region of interest. According to the invention, the discretization coordinate sampling sequence is operated to generate the obstacle contour curvature energy integral value, tangential angle change and local direction sensitive items are introduced in the process, the sharpness and morphological risk of the edge of the obstacle can be captured, the integral value is associated and matched to a transverse avoidance distance threshold, the avoidance space is automatically enlarged for sharp high-risk obstacles, the avoidance distance is properly reduced for smooth low-risk obstacles, and the narrow space is furthest utilized on the premise of ensuring safety.

Inventors

  • LIU HUAZHU

Assignees

  • 深圳市脑行者智行科技有限公司

Dates

Publication Date
20260512
Application Date
20260225

Claims (9)

  1. 1. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion is characterized by comprising the following steps of: acquiring a forward-looking road navigation image according to an intelligent wheelchair vision sensor, intercepting a driving region of interest under the forward-looking road navigation image according to intelligent wheelchair chassis width parameters, extracting a pixel gray level distribution matrix in the driving region of interest, calculating gray level symbiotic contrast values between adjacent pixel points in the pixel gray level distribution matrix, and generating road texture space distribution characteristics; Converting the data of the pavement texture space distribution characteristics from a space domain to a frequency domain spectrum coordinate system, extracting spectrum centroid frequency and spectrum width in the frequency domain spectrum coordinate system, calculating the spectrum centroid frequency and the spectrum width, and quantifying pavement visual roughness index; Identifying a close-range obstacle contour line positioned outside the boundary of the area corresponding to the road surface visual roughness index in the forward-looking road surface navigation image, performing discretization coordinate sampling on the close-range obstacle contour line to obtain a discretization coordinate sampling sequence, and performing operation on the discretization coordinate sampling sequence to generate an obstacle contour curvature energy integral value; And mapping the pavement visual roughness index comparison to a wheelchair motor torque control table, and correlating and matching the obstacle profile curvature energy integral value to a transverse avoidance distance threshold value to generate an intelligent wheelchair self-adaptive positioning and obstacle avoidance strategy instruction.
  2. 2. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to claim 1, wherein the step of obtaining the pixel gray distribution matrix is as follows: Reading a forward-looking road navigation image output by an intelligent wheelchair vision sensor, calling intelligent wheelchair chassis width parameters to convert the parameters into pixel widths, downwards positioning a rectangular belt below the bottom edge of the forward-looking road navigation image by taking the central column of an imaging coordinate system as a reference, setting left and right boundaries according to the pixel widths, defining longitudinal heights along the lower edge, cutting out the rectangular belt, correcting out-of-range pixels, and obtaining a driving region of interest; According to the running interested region, the gray value of each pixel is mapped into discrete gray level according to a fixed gray level quantization rule, the gray level is read point by point according to the sequence of a row index and a column index and written into a corresponding matrix position, and the edge missing pixels are filled in a neighboring pixel copying mode to keep the rows and columns intact, so that a pixel gray level distribution matrix is obtained.
  3. 3. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to claim 1, wherein the step of obtaining the road surface texture space distribution characteristics is as follows: According to the pixel gray level distribution matrix, respectively constructing an index sequence of adjacent pixel pairs in the vertical direction and the diagonal direction in the horizontal direction, calculating gray level symbiotic contrast values of the adjacent pixel pairs pair by pair, recording row-column coordinates, writing the gray level symbiotic contrast values into corresponding pixel grids of the spatial domain texture feature layer according to the recorded coordinates, and carrying out position combination on multi-direction results to obtain the road surface texture spatial distribution features.
  4. 4. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to claim 1, wherein the spectrum centroid frequency and spectrum width acquisition steps are as follows: According to the road surface texture space distribution characteristics, the texture intensity value corresponding to each pixel position is read according to a row-column sequence, the texture intensity value sequence is subjected to transformation operation on a frequency axis, frequency components and corresponding amplitude spectrums are obtained, all the amplitude spectrums are arranged from low frequency to high frequency, and the upper frequency limit is recorded, so that a frequency domain spectrum coordinate system mapping result is formed; According to the mapping result of the frequency domain spectrum coordinate system, calculating the weighted center frequency of spectrum energy, adopting the energy of each frequency point as a weight to obtain the spectrum centroid frequency, and calculating the standard deviation of spectrum energy distribution by taking the spectrum centroid frequency as the center to obtain the spectrum width.
  5. 5. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion of claim 1, wherein the step of obtaining the road surface visual roughness index is: according to the spectrum centroid frequency and the spectrum width, calculating a pavement visual roughness index, wherein a calculation formula is as follows: ; Wherein, the As an index of the visual roughness of the road surface, Is obtained by weighted average of frequency and energy, and the calculation formula is Wherein Is the first The frequency points of the frequency spectrum are selected, As the energy of the frequency point, Reflecting the degree of dispersion of the frequency distribution for the frequency spectrum width, the calculation formula is , The upper frequency limit of the mapping result of the frequency domain spectrum coordinate system is used for normalization processing to eliminate sampling rate difference, Is extremely small positive constant for preventing Zero-approaching denominator approaches zero.
  6. 6. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to claim 1, wherein the step of obtaining the discretized coordinate sampling sequence is: Reading boundary coordinates of a region corresponding to the road surface visual roughness index according to the forward-looking road surface navigation image, positioning a continuous edge chain outside the boundary, and carrying out region growth tracking according to a pixel eight-connection rule to obtain a close-range obstacle contour line; And according to the outline of the close-range obstacle, sampling along the outline arc length at fixed pixel intervals, and recording the transverse pixel coordinates and the longitudinal pixel coordinates of each sampling point to form a discretization coordinate sampling sequence.
  7. 7. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to claim 1, wherein the step of obtaining the obstacle profile curvature energy integral value is: according to the discretization coordinate sampling sequence, calculating an obstacle contour curvature energy integral value, wherein the calculation formula is as follows: ; Wherein, the To be an integrated value of the energy of the curvature of the obstacle profile, For the number of sample points of the discretized coordinate sample sequence, Is the first The included angle between the section barrier contour line and the horizontal direction is calculated by coordinates of two adjacent sampling points, and the calculation formula is that Wherein Is the first The lateral pixel coordinates of the individual sample points, Is the first The vertical pixel coordinates of the individual sample points, Represents the variation of the included angle of two adjacent sections of contour lines, is used for reflecting the variation amplitude of local curvature, For local direction sensitive terms, the weight increases as the contour direction is near vertical.
  8. 8. The intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to claim 1, wherein the intelligent wheelchair self-adaptive positioning and obstacle avoidance strategy instruction obtaining step is as follows: Reading an index column of a torque control table of the wheelchair motor according to the visual roughness index of the road surface, positioning the segment according to the boundary of the preset segment, and if the segment falls between the adjacent segments, performing interpolation according to the boundary proportion, outputting a target torque limit, a torque rising slope and a torque response delay to obtain output parameters of the torque control table of the wheelchair motor; reading the grading of the obstacle contour curvature energy integral value, matching a transverse distance interval according to the grade of the obstacle contour curvature energy integral value, and generating a transverse avoidance distance threshold by combining imaging proportion conversion into an actual transverse distance threshold; And according to the transverse avoidance distance threshold, invoking target torque limitation, torque rising slope and torque response delay in output parameters of a torque control table of the wheelchair motor, selecting deceleration or braking or detouring according to whether the transverse avoidance distance threshold is broken through, synthesizing instruction sets of a longitudinal control channel and a transverse control channel, and generating an intelligent wheelchair self-adaptive positioning and obstacle avoidance strategy instruction.
  9. 9. The system of the intelligent wheelchair adaptive positioning and obstacle avoidance method based on multi-sensor fusion according to any one of claims 1-8, comprising: The texture feature extraction module is used for acquiring a forward-looking road navigation image according to the intelligent wheelchair vision sensor, intercepting a driving region of interest under the forward-looking road navigation image according to the intelligent wheelchair chassis width parameter, extracting a pixel gray level distribution matrix in the driving region of interest, calculating gray level symbiotic contrast values between adjacent pixel points in the pixel gray level distribution matrix, and generating road texture space distribution features; the frequency domain roughness quantization module is used for converting the data of the pavement texture space distribution characteristics from a space domain to a frequency domain spectrum coordinate system, extracting the spectrum centroid frequency and the spectrum width in the frequency domain spectrum coordinate system, calculating the spectrum centroid frequency and the spectrum width, and quantizing the pavement visual roughness index; The obstacle contour analysis module is used for identifying a close-range obstacle contour line positioned outside the boundary of the area corresponding to the road surface visual roughness index in the forward-looking road surface navigation image, performing discretization coordinate sampling on the close-range obstacle contour line to obtain a discretization coordinate sampling sequence, and performing operation on the discretization coordinate sampling sequence to generate an obstacle contour curvature energy integral value; The self-adaptive control strategy generation module is used for mapping the road surface visual roughness index comparison to a wheelchair motor torque control table, and correlatively matching the obstacle profile curvature energy integral value to a transverse avoidance distance threshold value to generate an intelligent wheelchair self-adaptive positioning and obstacle avoidance strategy instruction.

Description

Intelligent wheelchair self-adaptive positioning and obstacle avoidance method and system based on multi-sensor fusion Technical Field The invention relates to the technical field of image recognition, in particular to an intelligent wheelchair self-adaptive positioning and obstacle avoidance method and system based on multi-sensor fusion. Background Image recognition technology is the processing, analysis, and understanding of images or video sequences with computer systems to identify targets and objects in a variety of different modes. The existing image recognition technology for intelligent wheelchairs is usually focused on semantic classification and position frame selection of objects in a scene, and tends to consider a road surface as a homogeneous geometric plane or only carry out binarization of drivable area division, and analysis of microscopic textures and frequency domain features of the road surface is ignored, so that actual roughness and friction coefficient differences of the road surface are difficult to perceive. When facing complex and changeable unstructured roads, the single perception mode makes the control system difficult to adjust the power output in real time according to road conditions. Therefore, improvements are needed. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides an intelligent wheelchair self-adaptive positioning and obstacle avoidance method and system based on multi-sensor fusion. In order to achieve the purpose, the intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion adopts the following technical scheme that the intelligent wheelchair self-adaptive positioning and obstacle avoidance method based on multi-sensor fusion comprises the following steps: acquiring a forward-looking road navigation image according to an intelligent wheelchair vision sensor, intercepting a driving region of interest under the forward-looking road navigation image according to intelligent wheelchair chassis width parameters, extracting a pixel gray level distribution matrix in the driving region of interest, calculating gray level symbiotic contrast values between adjacent pixel points in the pixel gray level distribution matrix, and generating road texture space distribution characteristics; Converting the data of the pavement texture space distribution characteristics from a space domain to a frequency domain spectrum coordinate system, extracting spectrum centroid frequency and spectrum width in the frequency domain spectrum coordinate system, calculating the spectrum centroid frequency and the spectrum width, and quantifying pavement visual roughness index; Identifying a close-range obstacle contour line positioned outside the boundary of the area corresponding to the road surface visual roughness index in the forward-looking road surface navigation image, performing discretization coordinate sampling on the close-range obstacle contour line to obtain a discretization coordinate sampling sequence, and performing operation on the discretization coordinate sampling sequence to generate an obstacle contour curvature energy integral value; And mapping the pavement visual roughness index comparison to a wheelchair motor torque control table, and correlating and matching the obstacle profile curvature energy integral value to a transverse avoidance distance threshold value to generate an intelligent wheelchair self-adaptive positioning and obstacle avoidance strategy instruction. Preferably, the step of obtaining the pixel gray distribution matrix includes: Reading a forward-looking road navigation image output by an intelligent wheelchair vision sensor, calling intelligent wheelchair chassis width parameters to convert the parameters into pixel widths, downwards positioning a rectangular belt below the bottom edge of the forward-looking road navigation image by taking the central column of an imaging coordinate system as a reference, setting left and right boundaries according to the pixel widths, defining longitudinal heights along the lower edge, cutting out the rectangular belt, correcting out-of-range pixels, and obtaining a driving region of interest; According to the running interested region, the gray value of each pixel is mapped into discrete gray level according to a fixed gray level quantization rule, the gray level is read point by point according to the sequence of a row index and a column index and written into a corresponding matrix position, and the edge missing pixels are filled in a neighboring pixel copying mode to keep the rows and columns intact, so that a pixel gray level distribution matrix is obtained. Preferably, the step of obtaining the spatial distribution characteristics of the pavement texture comprises the following steps: According to the pixel gray level distribution matrix, respectively constructing an index sequence of adjacent pixel pairs in the vertical di