Stair detection method based on iOS equipment carrying laser radar

Abstract

The invention discloses a real-time stair detection method based on iOS equipment carrying laser radar. Firstly, coarse recognition is performed on stairs through YOLOv8 (S1), then depth data is acquired by using a laser radar and converted into point cloud data, and voxel downsampling, radius denoising and height filtering preprocessing are performed (S2). In the fine detection stage, judgment is carried out according to the height relation between the point cloud and the ground: when the point cloud is lower than the ground, a plane is detected by adopting an RANSAC algorithm, and downstairs are identified by combining the area and a normal vector threshold value; and when the height is higher than the ground, analyzing the height difference, the normal vector and the number of point clouds of adjacent planes by using a region growing method, and judging stair climbing (S3). According to the method, computer vision and a point cloud processing technology are fused, efficient and accurate stair detection is realized, reliable stair sensing ability is provided for visually impaired people, and travel safety is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lidar perception and visually impaired assistance, and particularly relates to a staircase detection method based on an iOS device designed specifically for visually impaired persons. Background Art

[0002] As a vulnerable group, visually impaired persons often encounter problems with inconvenient travel in life. Staircases are common passable areas in daily travel, and the visually impaired group has a need for real-time perception of staircases. Currently, the main auxiliary devices for visually impaired persons to travel are blind canes, guide dogs, and electronic auxiliary devices. The blind cane is mainly used for obstacle avoidance, and it has a lag in perceiving staircases as it needs to be continuously tapped. The upfront investment cost of guide dogs is high, and they cannot be popularized for use. Although electronic assistance can use various sensor application algorithms to detect the environment for staircases, it currently has the disadvantages of high price and inconvenience in carrying.

[0003] With the continuous development of technology, the performance of mobile devices has been gradually improved, and at the same time, more and more various sensors are carried, and electronic assistance is gradually carried on mobile terminals. Apple Inc. added a lidar sensor to its Pro series devices in 2020. The lidar sensor is a sensor technology widely used in ranging, three-dimensional map construction, and environmental perception. It obtains the distance and position information of target objects by emitting laser beams and measuring the reflection time in space, thereby realizing high-precision perception and positioning of the surrounding environment. Lidar has the advantages of high precision, long distance, and all-weather operation, and is widely used in fields such as augmented reality, autonomous driving, robot navigation, and map making. Detecting whether there are staircases ahead and the type of staircases using lidar technology can provide decision-making support for the daily travel of visually impaired persons.

[0004] However, the types of staircases are complex and the number of floors varies, and the detection algorithm needs to consider different types. And for visually impaired persons, the algorithm for detecting staircases needs to consider real-time performance to make effective decisions. At the same time, the point cloud data of the staircases scanned by the Apple lidar presents different data qualities. When going up the stairs ahead, the side of the point cloud presents an obvious staircase step shape, but when going down the stairs ahead, the side point cloud of the staircase presents a slope shape with poor data quality. Therefore, targeted algorithm detection is required for different types of up and down staircases. Summary of the Invention

[0005] The purpose of the present invention is to provide a staircase detection method based on an iOS device, which is mainly applicable to the point cloud data of staircases scanned by an iOS device equipped with a lidar. It solves the problem of inapplicability of traditional methods under this data source, and can detect whether there are staircases ahead and identify the type of staircases in real time, efficiently, and accurately, helping visually impaired persons perceive the staircases ahead and make decisions.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The stair detection method based on iOS devices according to the present invention includes the following steps:

[0008] S1. Use the collected stair dataset to train the YOLO V8 object detection network framework to generate a stair detection model, and perform rough detection on the image data collected by the device;

[0009] S2. Obtain the depth map data collected by the device in real time, convert it into point cloud data, and perform preprocessing operations such as voxel downsampling, radius denoising, and straight-through filtering clipping on the point cloud to remove point cloud noise and redundancy, and improve the data processing efficiency;

[0010] S3. Judge the height relationship between the front point cloud and the ground according to the point cloud data. When it is lower than the ground, use RANSAC to detect the plane. When the plane area and normal vector meet the threshold, it is determined as going downstairs. When the point cloud is higher than the ground, use the region growing method to detect the plane, determine the normal vector of each plane, screen out the horizontal plane for analysis, and confirm whether the front is going upstairs according to the number of planes, the height of adjacent planes, and the relationship between the number of point clouds of the planes.

[0011] Further, step S1 specifically includes:

[0012] S11. Collect images of stairs to make training set data;

[0013] S12. Use the dataset to train the YOLO V8 object detection network framework to generate a model;

[0014] S13. Read the video obtained by the device frame by frame as a target picture, and use the model to detect the image to roughly identify the stairs.

[0015] Further, step S2 specifically includes:

[0016] S21. Obtain the depth map data collected by the device in real time, and convert the depth map into point cloud data;

[0017] S22. Use straight-through filtering to clip the point cloud;

[0018] S23. Use the voxel downsampling algorithm to reduce the initial number of point clouds, reduce redundancy, and uniform the point cloud density

[0019] S24. Use the radius filtering denoising algorithm to remove the noise points suspended in the air.

[0020] Further, in step S21, the depth map data collected by the device in real time is obtained, and then the internal parameters of the camera, including the focal length and the principal point, are obtained from the parameter configuration file. This parameter describes the geometric imaging of the camera. Each pixel in the depth map is traversed. For the position (X, Y) of each pixel, its coordinates in the normalized camera coordinate system are calculated through the following formula. Then, using the depth value in the depth image and the normalized camera coordinates, the real-space coordinates corresponding to each pixel in the camera coordinate system are calculated. Next, the external parameter information of the camera is obtained, and the camera coordinate system is converted into the world coordinate system through the external parameter information. Finally, the point cloud data is output.

[0021] Further, the pass-through filter in step S22 means filtering the points whose specified dimension direction is not within the given threshold. The specific dimension and threshold are specified in advance, and then by traversing the point cloud, it is judged whether the point is within the value range, and only the points within the threshold range are retained. By this method, the point cloud with too high Z-axis is removed to prevent the influence of the ceiling and other objects, and at the same time, the point cloud in the too far direction of the Y-axis is removed.

[0022] Further, the voxel downsampling filter algorithm in step S23 creates a number of three-dimensional voxel grids for the input point cloud data with a certain length, width, and height. Then, within each voxel, the center of gravity of all the points in the voxel is used to approximately represent other points in the voxel, so that all the points in the voxel can be represented by one center of gravity point. After processing all the voxels, the downsampled point cloud can be obtained.

[0023] Further, the radius filter denoising algorithm in step S24 traverses all the point clouds and calculates whether each point cloud contains N adjacent points within the specified search radius R. If there is no such point, it is regarded as a noise point and deleted, and if there is, it is retained.

[0024] Further, step S3 specifically includes:

[0025] S31. Determine the up and down types of the stairs by judging the height relationship between the front point cloud and the ground. First, obtain the lowest point Z1 of the point cloud within N meters from the device position, and then obtain the lowest point Z2 of the overall point cloud. Compare the two lowest points. When Z2 is lower than Z1 by a certain height threshold, it is considered that the front point cloud is of the potential going-downstairs type, otherwise it is of the potential going-upstairs type;

[0026] S32. When the point cloud is lower than the ground, the stair point cloud presents an obvious inclined plane feature. Therefore, the RANSAC (Random Sampling Consensus) method can be used to detect the plane, and it is judged whether the number of points in the plane meets the requirements and whether the normal vector of the plane conforms to the slope threshold. When the above conditions are met, it is considered that the front is going downstairs;

[0027] S33. When the point cloud is above the ground, use the region growing method to detect the plane. After finding the plane, calculate the orientation of the normal vector of the plane. If the normal vector meets the horizontal plane threshold range, it is retained, and the rest of the planes are deleted. Finally, multiple planes parallel to the horizontal plane are obtained.

[0028] S34. Sort all the planes according to their heights.

[0029] S35. Calculate the height difference between each plane and its adjacent plane in sequence. When the heights of two planes are close within a certain threshold, they are considered to be planes of the same layer, and one of the planes is deleted to prevent affecting subsequent staircase detection.

[0030] S36. Due to the characteristics of the staircase point cloud, the horizontal plane area of the second - layer staircase is mostly less than or equal to that of the first - layer staircase. Therefore, calculate the number of points in the point cloud of the lowest - layer and the second - lowest - layer planes. When the number of points in the second layer is significantly more than that in the first layer, there is no staircase ahead.

[0031] S37. The common height range of each layer of the staircase is within 0.15m - 0.25m. When the height between the first - layer and the second - layer planes and the height between the second - layer and the third - layer planes meet the above requirements, it is confirmed that the point cloud ahead contains a staircase.

[0032] Furthermore, the RANSAC algorithm in step S32 determines a plane based on three points. Randomly select three points and calculate the model parameters according to the quadratic equation Ax + By + Cz + d = 0 to fit a plane. Then calculate the distance from the remaining points to the plane. If it is less than the set distance threshold D, it is considered an inlier of the same plane, otherwise it is an outlier. Finally, record the number of inliers, and then iterate and repeat the above operations. Compare the number of inliers each time. Eventually, after reaching the iteration times, end the iteration and output the plane with the most inliers.

[0033] Furthermore, the region growing method in step S33 is an algorithm for segmenting point cloud data. The basic idea is to start from seed points and gradually grow continuous point cloud regions. First, select one or more seed points as the starting points for growth, and then define the neighborhood around the seed points, usually a spherical or cubic neighborhood centered on the seed points. During the growth process, according to the pre - set growth criteria, such as the distance between points, the direction of the normal vector, the similarity of color or other features, gradually add the points in the neighborhood to the current region. The growth process continues until no new points can be added or the pre - set stop condition is reached. Finally, the formed continuous region is the segmented point cloud region.

[0034] After adopting the above technical solutions, compared with the background technology, the present invention has the following advantages:

[0035] The present invention realizes a real-time, efficient and accurate stair detection function based on an iOS device equipped with a laser radar, helping visually impaired people to predict the stairs ahead and the type of stairs in advance when traveling. First, the model is used to perform a rough detection of the stairs in the image to reduce the consumption of real-time calculation of point cloud data. Secondly, the straight-through filtering algorithm removes the influence of the ceiling and sets the detection distance within a fixed range to ensure accuracy. The voxel downsampling filter enhances the calculation efficiency; the radius denoising filter removes discrete noise points around the stairs. Finally, according to the type of unevenness of the front point cloud relative to the ground, different methods are selected to detect the up and down stairs. The regional growing method is used to completely segment each plane of the up stairs for rapid detection of the stairs. At the same time, RANSAC plane detection is used to solve the point cloud data quality problem of the down stairs. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0037] Figure 2 This is a schematic diagram of the pretreatment process of the present invention;

[0038] Figure 3 The figure is a schematic diagram of the stair detection process based on different ground types according to the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] Example

[0041] refer to Figure 1 As shown, the present invention discloses an indoor navigation and obstacle avoidance method based on iOS device designed for visually impaired people, comprising the following steps:

[0042] S1. Use the collected staircase dataset to train the YOLO V8 target detection network framework to generate a staircase detection model, and perform rough detection on the image data collected by the device;

[0043] S2. Obtain the depth map data collected by the device in real time, and then convert it into point cloud data. Perform voxel downsampling, radius denoising, and straight-through filtering and clipping preprocessing operations on the point cloud to remove point cloud noise and redundancy and improve data processing efficiency.

[0044] S3. Determine the height relationship between the front point cloud and the ground based on the point cloud data. When it is lower than the ground, use RANSAC to detect the plane. When the plane area and normal vector meet the threshold, it is determined as going downstairs. When the point cloud is higher than the ground, use the region growing method to detect the plane, determine the normal vector of each plane, screen out the horizontal planes for analysis, and confirm whether it is going upstairs in front according to the number of planes, the height of adjacent planes, and the relationship between the number of point clouds on the planes.

[0045] Further, step S1 specifically includes:

[0046] S11. Collect images of stairs to make training set data;

[0047] S12. Use the data set to train the YOLO V8 object detection network framework to generate a model;

[0048] S13. Read the video obtained by the device frame by frame as a target picture, and use the model to detect the image to roughly identify the stairs.

[0049] Reference Figure 2 As shown, step S2 specifically includes:

[0050] S21. Obtain the depth map data collected by the device in real time, and convert the depth map into point cloud data;

[0051] S22. Use the passthrough filter to crop the point cloud;

[0052] S23. Use the voxel downsampling algorithm to reduce the initial number of point clouds, reduce redundancy, and uniform the point cloud density

[0053] S24. Use the radius filter denoising algorithm to remove the noise points suspended in the air.

[0054] Further, in step S21, obtain the depth map data collected by the device in real time, and then obtain the internal parameters of the camera from the parameter configuration file, including the focal length and the principal point. This parameter describes the geometric imaging of the camera. Traverse each pixel in the depth map. For the position (X, Y) of each pixel, calculate its coordinates in the normalized camera coordinate system through the following formula. Then use the depth value in the depth image and the normalized camera coordinates to calculate the real space coordinates of each pixel in the camera coordinate system. Then obtain the external parameter information of the camera, and convert the camera coordinate system into the world coordinate system through the external parameter information, and finally output the point cloud data.

[0055] Further, the passthrough filter in step S22 means filtering out the points whose specified dimension direction is not within the given threshold. Specify the specific dimension and threshold in advance, and then judge whether the point is within the value range by traversing the point cloud, and only retain the points within the threshold range. Set to remove the point clouds that are too high more than 2 meters outside the Z axis to prevent the influence of the ceiling and other objects, and at the same time remove the point clouds more than 4 meters away from the user's position.

[0056] Further, the voxel downsampling filtering algorithm in step S23 creates a number of three-dimensional voxel grids for the input point cloud data with a certain length, width, and height. Then, within each voxel, the center of gravity of all the points in the voxel is used to approximate the other points in the voxel, so that all the points in the voxel can be represented by one center of gravity point. After processing all the voxels, the downsampled point cloud can be obtained. An overly large voxel grid will lose the accuracy of the staircase point cloud and cannot identify the plane, while an overly small voxel grid results in an insignificant downsampling effect. Therefore, the length, width, and height of the voxel grid are all set to 0.1 m.

[0057] Further, the radius filtering denoising algorithm in step S24 traverses all the point clouds and calculates whether each point cloud contains N adjacent points within a specified search radius R. If there is no such point, it is regarded as a noise point and deleted; if there is, it is retained. The radius denoising algorithm involves two parameters, namely the size of the search radius and the number of neighborhood points. Noise is usually discretely distributed. The size of the search radius R is set to 0.5 m, and the minimum number of points N is set to 10.

[0058] Reference Figure 3 As shown, step S3 specifically includes:

[0059] S31. Determine the up and down types of the staircase by judging the height relationship between the front point cloud and the ground. First, obtain the lowest point Z1 of the point cloud within 1 meter of the device position, and then obtain the lowest point Z2 of the overall point cloud. Compare the two lowest points. When Z2 is 0.25 meters lower than the lowest value of Z1, it is considered that the front point cloud is a potential down-staircase type; otherwise, it is a potential up-staircase type.

[0060] S32. When the point cloud is lower than the ground, the staircase point cloud presents an obvious inclined plane feature. Therefore, the RANSAC (Random Sampling Consensus) method can be used to detect the plane, and it is determined whether the number of point clouds of the plane reaches more than 200 and whether the normal vector of the plane is within 0.7 - 0.9. When the above conditions are met, it is considered that the front is a down-staircase.

[0061] S33. When the point cloud is higher than the ground, the region growing method is used to detect the plane. After finding the plane, calculate the orientation of the normal vector of the plane. The planes whose normal vectors satisfy the horizontal plane threshold of more than 0.95 are retained, and the rest of the planes are deleted. Finally, multiple planes parallel to the horizontal plane are obtained.

[0062] S34. Sort all the planes according to height.

[0063] S35. Calculate the height difference between each plane and the adjacent plane in sequence. When the height difference between two planes is close to within 0.1, it is considered that they are planes on the same layer, and one of the planes is deleted to prevent affecting subsequent staircase detection.

[0064] S36. Due to the characteristics of the stair point cloud, the horizontal area of the second - floor stair is mostly less than or equal to that of the first - floor stair. Therefore, calculate the number of point clouds on the lowest - level and the second - lowest - level planes. When the number of point clouds on the second floor is significantly more than twice that of the first floor, then what lies ahead is not a stair.

[0065] S37. The common height range of each floor of the stair is within 0.15m - 0.25m. When the height between the first - floor and the second - floor planes and the height between the second - floor and the third - floor planes both meet the above requirements, then it is confirmed that the point cloud ahead contains a stair.

[0066] Furthermore, the RANSAC algorithm in step S32 determines a plane based on three points. Randomly select three points and calculate the model parameters according to the quadratic equation Ax + By + Cz + d = 0 to fit a plane. Then calculate the distance from the remaining points to the plane. If it is less than the set distance threshold of 0.02 meters, then it is considered an inlier of the same plane, otherwise it is an outlier. Finally, record the number of inliers, and then iterate and repeat the above operations. Compare the number of inliers each time. Eventually, after reaching the iteration count of 1000, end the iteration and output the plane with the most inliers. The minimum number of points in a plane is 50.

[0067] Furthermore, the region - growing method in step S33 is an algorithm for segmenting point - cloud data. The basic idea is to start from seed points and gradually grow continuous point - cloud regions. First, select one or more seed points as the starting points for growth, and then define the neighborhood around the seed points, usually a spherical or cubic neighborhood centered on the seed points. During the growth process, according to the pre - set growth criteria, such as the distance between points, the direction of the normal vector, color, curvature, or the similarity of other features, gradually add the points in the neighborhood to the current region. The growth process continues until no new points can be added or the pre - set stop condition is reached. Finally, the formed continuous region is the segmented point - cloud region. The minimum number of points in each plane is 100, the maximum number of points is 8000, and the curvature is set to 0.0125.

[0068] As described above, it is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A stair detection method based on an iOS device equipped with a laser radar, characterized in that: The following steps are involved: S1. Use the previously collected staircase image dataset to train the YOLO V8 target detection network framework to generate a staircase detection model, and perform rough detection on the image data collected by the device; S2. Obtain the depth map data collected by the device in real time, convert the depth map into point cloud data, and use voxel downsampling, radius denoising algorithm, and straight-through filtering and clipping preprocessing operations on the point cloud to remove point cloud noise and redundancy and improve data processing efficiency; S3. According to the elevation distribution of point cloud data at different distances, rough detection is performed to determine the type of up and down stairs ahead. When the elevation distribution is getting lower and lower, RANSAC is used to detect the plane. If the plane area and normal vector meet the threshold, it is determined to be downstairs. When the elevation distribution is getting higher and higher, the regional growing method is used to detect the plane, determine the normal vector of each plane, select the horizontal plane for analysis, and confirm whether the front is upstairs based on the number of planes, the height of adjacent planes, and the relationship between the number of point clouds of the planes.

2. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 1, characterized in that: Step S1 specifically includes: S11, collect images of stairs to make training set data; S12. Use the dataset to train the YOLO V8 target detection network framework generation model; S13, reading the video acquired by the device frame by frame as a target image, using the model to detect the image, and roughly identifying the stairs.

3. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 1, characterized in that: Step S2 specifically includes: S21, obtaining depth map data collected by the device in real time, and converting the depth map into point cloud data; S22, crop the point cloud using straight-through filtering; S23, use voxel downsampling algorithm to reduce the number of initial point clouds, reduce redundancy, and even out the point cloud density; S24. Use a radius filter denoising algorithm to remove noise points suspended in the air.

4. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 3, characterized in that: Step S21 specifically includes: obtaining the depth map data collected by the device in real time, obtaining the internal parameters of the camera from the parameter configuration file, including the focal length and the principal point, which describe the geometric imaging of the camera, traversing each pixel in the depth map, and for each pixel position (X, Y), calculating its coordinates in the normalized camera coordinate system by the following formula, and then using the depth value in the depth image and the normalized camera coordinates to calculate the real space coordinates corresponding to each pixel in the camera coordinate system, and then obtaining the external parameter information of the camera, converting the camera coordinate system into the world coordinate system through the external parameter information, and finally outputting the point cloud data.

5. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 3, characterized in that: Step S22 specifically includes: using straight-through filtering to crop the point cloud, filtering points that are not within a given threshold in a specified dimension direction, specifying a specific dimension and a value range, and determining whether the point is within the value range by traversing the point cloud, and only retaining points within the value range. This method removes point clouds that are too high on the Z axis to prevent images of the ceiling and other objects, and removes point clouds that are too far in the Y axis direction to reduce data consumption.

6. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 3, characterized in that: Step S23 uses voxel filtering to thin the point cloud, which reduces the amount of data while making the point cloud density uniform. The voxel thinning method divides the point cloud into countless cubic cells for projection by setting cubic cells with fixed side lengths. The centroid points of all point clouds in each cube are used as sampling points. The sampling points represent the representative points of all points in the cells. After all cells are processed, voxel filtering can be completed.

7. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 6, characterized in that: Step S23 uses voxel filtering to thin the point cloud. A voxel side length greater than 5 cm will result in reduced accuracy of stair detection, and a voxel side length less than 1 cm will result in an insignificant thinning effect and a slower stair detection speed.

8. The stair detection method based on an iOS device equipped with a laser radar as claimed in claim 3, characterized in that: Step S24: The radius filtering denoising algorithm traverses all point clouds and calculates whether each point cloud contains N adjacent points within the specified search radius R. If there is no such point, it is considered as a noise point and deleted. If there is such a point, it is retained. The algorithm contains two parameters, namely the search radius value and the adjacent point threshold N. Since noise points usually exist in a discrete state, the two values ​​do not need to be set too large.

9. The stair detection method based on an iOS device equipped with a laser radar as claimed in claim 1, characterized in that: Step S3 specifically includes: S31, determine the height relationship between the front point cloud and the ground to detect the up and down type of the stairs, first obtain the point cloud within 0.3 meters from the device position, obtain the lowest point Z1, then obtain the lowest point Z2 of the entire point cloud, compare the two lowest points, when Z2 is lower than a certain height threshold of Z1, it is considered that the front point cloud is a potential downstairs type, otherwise it is a potential upstairs type; S32. When the point cloud is lower than the ground, the stair point cloud shows obvious slope features. Therefore, the RANSAC (Random Sampling Consensus) method can be used to detect the plane to determine whether the number of point clouds on the plane meets the requirement and whether the normal vector of the plane meets the slope threshold. If the above conditions are met, it is considered that there are stairs ahead; S33, when the point cloud is higher than the ground, the plane is detected using the region growing method. After the plane is found, the normal vector orientation of the plane is calculated. The planes whose normal vectors meet the horizontal plane threshold range are retained, and the rest are deleted, and finally multiple planes parallel to the horizontal plane are obtained; S34, then analyzing the retained planes and sorting all the planes by height; S35, calculating the height difference between each plane and the adjacent plane in sequence, and when the heights of two planes are close to a certain threshold, they are considered to be planes on the same floor, and one of the planes is deleted to prevent affecting subsequent stair detection; S36. Due to the characteristics of the stair point cloud, the horizontal plane area of ​​the second staircase is mostly smaller than or equal to the horizontal plane area of ​​the first staircase. Therefore, the number of point clouds of the lowest layer and the second lowest layer is calculated. When the number of point clouds of the second layer is obviously more than that of the first layer, there is no staircase in front. S37. The height of each staircase is usually in the range of 0.15m to 0.25m. When the height between the first and second floors and the height between the second and third floors meet the above requirements, it is confirmed that the front point cloud contains stairs.

10. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 9, characterized in that: In step S32, a plane is determined based on three points. Three points are randomly selected according to the square equation Ax+By+Cz+d=0, and the model parameters are calculated to fit into a plane. Then the distance of the remaining points from the plane is calculated. If it is less than the set distance threshold D, it is considered to be an inner point of the same plane, otherwise it is an outer point. Finally, the number of inner points is recorded, and then the above operation is repeated iteratively, and the number of inner points each time is compared. When the number of iterations is finally reached, the iteration is ended and the plane with the most inner points is output.

11. The stair detection method based on an iOS device equipped with a laser radar as claimed in claim 10, characterized in that: In step S32, the point cloud quantity threshold should be greater than 50, the plane normal vector should be greater than 0.9, and the iteration number threshold should be conducive to extracting a complete plane point cloud.

12. The stair detection method based on an iOS device equipped with a laser radar as claimed in claim 9, characterized in that: The region growing method of the point cloud in step S33 is an algorithm for segmenting point cloud data. The basic idea is to start from a seed point and gradually grow a continuous point cloud region. First, one or more seed points are selected as the starting point of growth, and then a neighborhood around the seed point is defined, usually a spherical or cubic neighborhood centered on the seed point. During the growth process, points in the neighborhood are gradually added to the current region according to pre-set growth criteria, such as the distance between points, normal vector direction, color or other feature similarity, etc. The growth process continues until no new points can be added or the pre-set stop condition is reached. Finally, the continuous region formed is the point cloud region obtained by segmentation.

13. A stair detection method based on an iOS device equipped with a laser radar as claimed in claim 12, characterized in that: In step S33, the range of the domain is within the threshold range, and the criteria for region growing, such as the direction of the normal vector and the minimum number of planes, should be conducive to the extraction of the plane.

Images