Point cloud denoising method

Abstract

The embodiment of the disclosure discloses a point cloud denoising method, which comprises the following steps: obtaining to-be-denoised point cloud data, the to-be-denoised point cloud data being point cloud data collected in a specified area based on a preset collection track; performing filtering processing on point cloud data outside the specified area from the to-be-denoised point cloud data based on the collection track; detecting a mirror surface range in the specified area; and performing deletion processing on point cloud data collected from inside the mirror surface range in the point cloud data after the filtering processing, thereby obtaining denoised point cloud data.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to three-dimensional reconstruction technology, point cloud processing technology, and in particular to a point cloud denoising method. BACKGROUND

[0002] With the continuous development of laser radar detection technology, the depth information of the measured target can be quickly obtained at present, and then the point cloud data of the measured target is generated. Influenced by the measurement environment and the surface characteristics of the measured object, noise is inevitably present. For example, the situation of collecting laser point cloud in indoor living environment is very complex. The mirror and other reflective objects will interfere with laser collection, causing the original point cloud collected by the laser equipment to be very messy, and it is also possible to collect invalid point cloud clusters outdoors.

[0003] In view of the above situation, in the related art, a point cloud denoising method is adopted by manually editing using third-party software, such as manually cropping noise points. The manual editing denoising method is relatively troublesome to operate and cannot guarantee that the processed point cloud data is clean and neat. The denoising method is inefficient and has poor processing effect. SUMMARY

[0004] In view of the above problems of the prior art, the embodiments of the present disclosure provide a point cloud denoising method.

[0005] The first aspect of the embodiments of the present disclosure provides a point cloud denoising method, comprising:

[0006] Obtaining to-be-denoised point cloud data, the to-be-denoised point cloud data being point cloud data collected in a specified region based on a preset collection track;

[0007] Based on the collection track, performing filtering processing on point cloud data outside the specified region from the to-be-denoised point cloud data;

[0008] Detecting a mirror range in the specified region;

[0009] In the filtered point cloud data, deleting point cloud data collected from inside the mirror range to obtain denoised point cloud data.

[0010] As a possible implementation manner of the first aspect, based on the collection track, performing filtering processing on point cloud data outside the specified region from the to-be-denoised point cloud data, comprising:

[0011] Performing clustering processing on the point cloud data based on the distance between the nearest neighbors to obtain at least one point cloud cluster;

[0012] Obtaining a first distance between the at least one point cloud cluster and the collection track;

[0013] Point cloud clusters with a first distance greater than or equal to a preset first distance threshold are determined as point cloud data outside the specified region;

[0014] From the at least one point cloud cluster, the point cloud data outside the specified region is filtered.

[0015] As a possible implementation manner of the first aspect, based on the collection track, the point cloud data outside the specified region is filtered from the to-be-de-noised point cloud data, including:

[0016] The to-be-de-noised point cloud data is projected onto a horizontal plane to obtain an occupancy grid map corresponding to the to-be-de-noised point cloud data;

[0017] At least one connected domain in the occupancy grid map is obtained;

[0018] The connected domain that is not passed through by the collection track is determined as the point cloud data outside the specified region;

[0019] From the at least one connected domain, the point cloud data outside the specified region is filtered.

[0020] As a possible implementation manner of the first aspect, the mirror surface range in the specified region is detected, including:

[0021] From the to-be-de-noised point cloud data, a light spot with an intensity value greater than a preset first intensity threshold is detected;

[0022] Based on the intensity value of the point cloud in the surrounding area of the light spot, the mirror surface range in the specified region is determined.

[0023] As a possible implementation manner of the first aspect, in the point cloud data after the filtering processing, the point cloud data collected from inside the mirror surface range is deleted to obtain the de-noised point cloud data, and the method further includes:

[0024] After the point cloud data collected from inside the mirror surface range is deleted, the point cloud noise points inside the wall in the specified region are filtered to obtain the de-noised point cloud data.

[0025] As a possible implementation manner of the first aspect, the point cloud noise points inside the wall in the specified region are filtered, including:

[0026] The point cloud data after the deletion processing is subjected to plane segmentation processing, and a plurality of wall planes in the specified region are obtained based on the result of the plane segmentation processing;

[0027] In the plurality of wall planes, two wall planes that are parallel to each other and have a distance satisfying a preset condition are determined as point cloud planes on two sides of a same wall;

[0028] The point cloud noise points in the middle of the point cloud planes on two sides of the same wall are removed.

[0029] As a possible implementation manner of the first aspect, the point cloud noise in the middle of the point cloud planes on both sides of the same wall is removed, including:

[0030] Projecting the point cloud data in the middle of the point cloud planes on both sides of the same wall onto a horizontal plane to obtain a projection image;

[0031] Removing the point cloud data corresponding to the region in the projection image where the ground does not exist.

[0032] As a possible implementation manner of the first aspect, before the point cloud data outside the specified region is filtered from the to-be-de-noised point cloud data based on the collection trajectory, the method further includes:

[0033] Obtaining the shooting distance of the point from the to-be-de-noised point cloud data;

[0034] Filtering out the point in the to-be-de-noised point cloud data whose shooting distance is greater than or equal to a preset second distance threshold.

[0035] As a possible implementation manner of the first aspect, before the point cloud data outside the specified region is filtered from the to-be-de-noised point cloud data based on the collection trajectory, the method further includes:

[0036] Obtaining the intensity value of the point from the to-be-de-noised point cloud data;

[0037] Filtering out the point in the to-be-de-noised point cloud data whose intensity value is less than or equal to a preset second intensity threshold.

[0038] As a possible implementation manner of the first aspect, before the point cloud data outside the specified region is filtered from the to-be-de-noised point cloud data based on the collection trajectory, the method further includes:

[0039] Statistically obtaining the k-nearest neighbor average distance of the point in the to-be-de-noised point cloud data;

[0040] Filtering out the point in the to-be-de-noised point cloud data whose k-nearest neighbor average distance is greater than or equal to a preset third distance threshold; wherein the preset third distance threshold is determined based on the standard deviation of the k-nearest neighbor average distances of all points in the to-be-de-noised point cloud data.

[0041] The second aspect of the embodiments of the present disclosure provides a point cloud de-noising device, including:

[0042] An acquisition unit is configured to acquire to-be-de-noised point cloud data, the to-be-de-noised point cloud data being point cloud data collected in a specified region based on a preset collection trajectory;

[0043] A first processing unit is configured to filter point cloud data outside the specified region from the to-be-de-noised point cloud data based on the collection trajectory;

[0044] The detection unit is configured to detect a mirror surface range in the specified area.

[0045] The second processing unit is configured to perform deletion processing on the point cloud data collected from inside the mirror surface range in the filtered point cloud data, to obtain denoised point cloud data.

[0046] As a possible implementation manner of the second aspect, the first processing unit is configured to:

[0047] perform clustering processing on the point cloud data based on the nearest neighbor point distance, to obtain at least one point cloud cluster;

[0048] obtain a first distance between the at least one point cloud cluster and the collection track;

[0049] determine, as the point cloud data outside the specified area, the point cloud cluster whose first distance is greater than or equal to a preset first distance threshold;

[0050] perform filtering processing on the point cloud data outside the specified area from the at least one point cloud cluster.

[0051] As a possible implementation manner of the second aspect, the first processing unit is configured to:

[0052] project the point cloud data to be denoised onto a horizontal plane, to obtain an occupancy grid map corresponding to the point cloud data to be denoised;

[0053] obtain at least one connected domain in the occupancy grid map;

[0054] determine, as the point cloud data outside the specified area, the connected domain that is not passed through by the collection track;

[0055] perform filtering processing on the point cloud data outside the specified area from the at least one connected domain.

[0056] As a possible implementation manner of the second aspect, the detection unit is configured to:

[0057] detect a light spot with an intensity value greater than a preset first intensity threshold from the point cloud data to be denoised;

[0058] determine the mirror surface range in the specified area based on intensity values of point clouds in a surrounding area of the light spot.

[0059] As a possible implementation manner of the second aspect, the apparatus further includes a third processing unit configured to:

[0060] perform filtering on point cloud noise inside a wall in the specified area after performing deletion processing on the point cloud data collected from inside the mirror surface range, to obtain the denoised point cloud data.

[0061] As a possible implementation manner of the second aspect, the third processing unit includes:

[0062] the segmentation subunit is configured to perform a plane segmentation process on the point cloud data after the deletion processing, and obtain a plurality of wall planes in the specified area based on a result of the plane segmentation process;

[0063] the matching subunit is configured to determine, among the plurality of wall planes, two wall planes that are parallel to each other and have a distance satisfying a preset condition as point cloud planes on two sides of a same wall;

[0064] the elimination subunit is configured to eliminate point cloud noise points between the point cloud planes on two sides of the same wall.

[0065] As a possible implementation manner of the second aspect, the elimination subunit is configured to:

[0066] project the point cloud data between the point cloud planes on two sides of the same wall onto a horizontal plane to obtain a projection image;

[0067] eliminate the point cloud data corresponding to a region in the projection image where no ground exists.

[0068] As a possible implementation manner of the second aspect, the device further includes a fourth processing unit configured to:

[0069] obtain a shooting distance of a point from the point cloud data to be denoised;

[0070] filter out points in the point cloud data to be denoised whose shooting distance is greater than or equal to a preset second distance threshold.

[0071] As a possible implementation manner of the second aspect, the device further includes a fourth processing unit configured to:

[0072] obtain an intensity value of a point from the point cloud data to be denoised;

[0073] filter out points in the point cloud data to be denoised whose intensity value is less than or equal to a preset second intensity threshold.

[0074] As a possible implementation manner of the second aspect, the device further includes a fourth processing unit configured to:

[0075] statistically obtain a k-nearest neighbor average distance of points in the point cloud data to be denoised;

[0076] filter out points in the point cloud data to be denoised whose k-nearest neighbor average distance is greater than or equal to a preset third distance threshold; wherein the preset third distance threshold is determined based on a standard deviation of the k-nearest neighbor average distances of all points in the point cloud data to be denoised.

[0077] A third aspect of the embodiments of the present disclosure provides an electronic device, comprising:

[0078] a memory configured to store a computer program product;

[0079] a processor configured to execute a computer program product stored in the memory, and the computer program product, when executed, implements the method of any one of the first aspect.

[0080] A fourth aspect of the embodiments of the present disclosure provides a computer readable storage medium having computer program instructions stored thereon, and the computer program instructions, when executed by a processor, implement the method of any one of the first aspect.

[0081] Based on the embodiments of the present disclosure, invalid point cloud data outside the specified area is filtered based on the collected trajectory, and the noise caused by the mirror and other reflective objects is removed by detecting the mirror, so that the filtering effect of the point cloud data collected in the specified area can be improved.

[0082] The technical solutions of the present disclosure will be described in further detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0083] The accompanying drawings, which form a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0084] The present disclosure can be more clearly understood with reference to the following detailed description, taken in conjunction with the accompanying drawings, in which:

[0085] Figure 1 Flow chart of an embodiment of the point cloud denoising method of the present disclosure;

[0086] Figure 2 Flow chart of an embodiment of the point cloud denoising method of the present disclosure;

[0087] Figure 3 Flow chart of an embodiment of the point cloud denoising method of the present disclosure;

[0088] Figure 4 Point cloud data projection schematic diagram of an embodiment of the point cloud denoising method of the present disclosure;

[0089] Figure 5 Point cloud data projection schematic diagram of an embodiment of the point cloud denoising method of the present disclosure;

[0090] Figure 6 Mirror reflection schematic diagram of an embodiment of the point cloud denoising method of the present disclosure;

[0091] Figure 7 Flow chart of an embodiment of the point cloud denoising method of the present disclosure;

[0092] Figure 8 Effect diagram of an embodiment of the point cloud denoising method of the present disclosure;

[0093] Figure 9An effect diagram of an embodiment of the point cloud denoising method of the present disclosure;

[0094] Figure 10 A structural schematic diagram of an embodiment of the point cloud denoising device of the present disclosure;

[0095] Figure 11 A structural schematic diagram of an embodiment of the point cloud denoising device of the present disclosure;

[0096] Figure 12 A structural schematic diagram of an embodiment of the point cloud denoising device of the present disclosure;

[0097] Figure 13 A structural schematic diagram of an embodiment of the point cloud denoising device of the present disclosure;

[0098] Figure 14 A block diagram of an electronic device of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0099] First, the existing method is introduced, and then the technical solution of the present application is introduced in detail.

[0100] In the related art, the point cloud denoising is performed in a manual editing manner by using a third-party software. The manual editing denoising manner is troublesome to operate and cannot guarantee that the processed point cloud data is clean and neat. In addition, for the interference of a mirror surface and the like, a processing manner of shielding the reflective object is also used to reduce the noise points caused by the mirror surface. However, the preparation work before collection is relatively cumbersome. In summary, the existing point cloud denoising manner is low in efficiency and poor in processing effect.

[0101] Based on the technical problems existing in the prior art, the embodiments of the present application provide a point cloud denoising method, which filters out invalid point cloud data outside a specified region based on a collection track, and deletes noise points caused by reflective objects such as a mirror surface by detecting the mirror surface, so as to improve the filtering effect of the point cloud data collected in the specified region, realize automatic denoising, and thus solve the technical problems of low denoising efficiency, complicated operation, and poor effect mentioned in the prior art.

[0102] Figure 1 A flowchart of an embodiment of the point cloud denoising method of the present disclosure. As shown in Figure 1 may specifically include:

[0103] In step S110, the point cloud data to be denoised is acquired, which is point cloud data collected in a specified region based on a preset collection track.

[0104] In one example, the specified area can include the indoor area of a specified house. Specifically, the collection trajectory can be pre-set. For example, the collection trajectory in the indoor area of a house can be to move around the wall on a trajectory line at a distance of L meters from the wall in each room. Wherein, L can be set to a distance of 0.5 meters or more. According to the layout of the house, the point cloud collection device can be controlled to move in the indoor area according to the preset collection trajectory in sequence, and the point cloud data is collected in each room in turn. Wherein, the point cloud collection device can be a mobile scanning device that emits laser, such as a three-dimensional laser scanner, a laser radar, etc. The mobile scanning device continuously emits laser during movement, and the collection trajectory can be considered as a connecting line connecting the laser emission points, that is, the trajectory of the movement of the light source and the emission point.

[0105] The point cloud data collected in the actual application scene is easy to produce noise points. Noise points are irrelevant or unwanted interference signals or errors in point cloud data. Point cloud denoising is to remove noise points from the collected point cloud data to be denoised by appropriate processing methods to improve data quality and accuracy. In step S110, the point cloud denoising tool, as the main body of the point cloud denoising method of the present disclosure, obtains the point cloud data to be denoised from the point cloud collection device, and performs denoising processing on the point cloud data to be denoised in the subsequent steps.

[0106] In step S120, based on the collection trajectory, the point cloud data outside the specified area is filtered from the point cloud data to be denoised.

[0107] In the process of indoor shooting by the point cloud collection device, the laser can pass through the window to shoot the trees, buildings, etc. outside. Therefore, the point cloud data to be denoised collected will include point cloud clusters belonging to outdoor targets, which are outside the specified area and belong to invalid point cloud clusters. Under normal circumstances, these invalid point cloud clusters are not connected to the main body of the house. By using this characteristic, the invalid point cloud clusters can be filtered out. Since the points in the invalid point cloud clusters are relatively far away from the specified area, the point cloud data to be denoised can be denoised based on the collection trajectory. Specifically, the point cloud clusters close to the collection trajectory can be retained, and the point cloud clusters far away from the collection trajectory can be filtered out, so as to filter out the invalid point cloud data outside the specified area from the point cloud data to be denoised.

[0108] In step S130, the mirror surface range in the specified area is detected. Wherein, the mirror surface can include the surface area of objects such as mirrors and glass with excellent reflection characteristics. When laser is vertically incident on the mirror surface, a high-intensity reflection echo will be generated. By using the echo characteristics of the above mirror reflection, the high-intensity light spot position in the specified area is detected to find all the mirror surface ranges in the specified area. Based on the position of the high-intensity light spot, the mirror surface range in the house is determined.

[0109] Step S140, in the filtered point cloud data, the point cloud data collected from inside the mirror surface range is deleted to obtain the denoised point cloud data.

[0110] In the collected point cloud data to be denoised, all the points passing through the mirror surface range can be considered as unreal and should not be collected. That is, those points collected through the mirror surface range are not real, but mirror points. Therefore, after detecting the mirror surface range, all the mirror point clouds collected by the light passing through the inside of the range are deleted. In the filtered point cloud data obtained after step S120, all the light reflected from the mirror is deleted in step S140, and the noise caused by the reflective object such as the mirror is deleted to obtain the denoised point cloud data.

[0111] In the above embodiment, first, the invalid point cloud data outside the specified region is filtered based on the collection trajectory, and then the point cloud data collected from inside the mirror surface range is deleted to delete the interference noise caused by the reflective object such as the mirror, and finally the denoised point cloud data is obtained. In another embodiment, the interference noise caused by the reflective object such as the mirror can also be deleted first, that is, the point cloud data collected from inside the mirror surface range is deleted in the point cloud data to be denoised, and then the invalid point cloud data outside the specified region is filtered based on the collection trajectory, and finally the denoised point cloud data is obtained.

[0112] Based on the embodiments of the present disclosure, the invalid point cloud data outside the specified region is filtered based on the collection trajectory, and the noise caused by the reflective object such as the mirror is deleted by detecting the mirror, which can improve the filtering effect of the point cloud data collected in the specified region.

[0113] As shown in FIG. 1, Figure 2 in an embodiment, Figure 1 Step S120 in the embodiment, based on the collection trajectory, the point cloud data outside the specified region is filtered from the point cloud data to be denoised, including:

[0114] Step S210, the point cloud data is clustered based on the distance of the nearest neighbor point to obtain at least one point cloud cluster;

[0115] Step S220, the first distance between at least one point cloud cluster and the collection trajectory is obtained;

[0116] Step S230, the point cloud cluster with the first distance greater than or equal to the preset first distance threshold is determined as the point cloud data outside the specified region;

[0117] Step S240, the point cloud data outside the specified region is filtered from at least one point cloud cluster.

[0118] The nearest neighbor distance of the point cloud refers to finding the nearest neighbor point of a given point in the point cloud and calculating the Euclidean distance between them. The point cloud clustering algorithm is an algorithm for segmenting point cloud data into different categories. In step S210, the points in the point cloud data to be denoised are clustered based on the nearest neighbor distance, and a plurality of different point cloud clusters are obtained. In step S220, the first distance between each point cloud cluster in at least one point cloud cluster and the collection trajectory is calculated. For example, the shortest distance between the geometric center point of the point cloud cluster and all points on the collection trajectory can be calculated as the first distance. In step S230, only the point cloud clusters close to the collection trajectory are retained. For example, the point cloud collection device captures trees, buildings, etc. outside through the window, which are far away from the collection trajectory and belong to the point cloud data outside the specified area, and should be deleted. Specifically, a preset first distance threshold is set in advance. For the point cloud cluster whose first distance is greater than or equal to the threshold, the position is far away from the collection trajectory, and it is deleted as the point cloud data outside the specified area in step S240.

[0119] In the above embodiment, the implementation of the nearest neighbor distance search algorithm is relatively simple, does not need to assume the distribution form of the data, and is suitable for various types of data. This makes this way have high flexibility when processing complex point cloud data. Based on the nearest neighbor distance, the point cloud data is clustered, which can effectively detect and identify the objects in the space environment, and then delete the point cloud data outside the specified area, to achieve good denoising effect.

[0120] As shown in FIG. 1, Figure 3 In one embodiment, Figure 1 In step S120, the point cloud data outside the specified area is filtered from the point cloud data to be denoised based on the collection trajectory, including:

[0121] In step S310, the point cloud data to be denoised is projected onto a horizontal plane to obtain an occupancy grid map corresponding to the point cloud data to be denoised;

[0122] In step S320, at least one connected domain in the occupancy grid map is obtained;

[0123] In step S330, the connected domain not passed through by the collection trajectory is determined as the point cloud data outside the specified area;

[0124] In step S340, the point cloud data outside the specified area is filtered from the at least one connected domain.

[0125] The point cloud collection device takes pictures of trees, buildings, etc. outside through the window. These objects outside the specified area belong to invalid point cloud data. If the point cloud data is projected onto the horizontal plane, the indoor layout of the house is connected, and the trees outside the window scanned through the window are some isolated point cloud clusters that are not connected to the indoor layout. Thus, the three-dimensional point cloud data can be projected onto the two-dimensional horizontal plane, converted into a two-dimensional connected domain, and then the invalid point cloud data outside the specified area is removed according to the above features.

[0126] Specifically, in step S310, the point cloud data to be denoised is projected onto the horizontal plane, the spatial data is converted from the three-dimensional coordinate system to the two-dimensional coordinate system through the projection technology, and the occupancy grid map is generated in the process. The occupancy grid map obtained by projection refers to the occupancy grid map, which divides the space into a plurality of small grids (grids), each grid represents a specific space area, and whether these areas are occupied can be marked.

[0127] In step S320, the connected domain in the occupancy grid map is calculated. The connected domain generally refers to the image area composed of foreground pixel points with the same pixel value and adjacent positions in the image. In the occupancy grid map, the connected domain analysis is achieved by finding and marking adjacent pixel points with the same occupancy state. In one example of calculation, the basic algorithm of connected domain analysis includes scan line and seed filling algorithm. Among them, the scan line algorithm scans the image row by row, uses a stack to store the boundary points, and then expands to the entire connected region. The seed filling algorithm starts from a seed point and gradually fills the adjacent pixel points around it until there are no more adjacent pixels to fill.

[0128] Figure 4 A schematic diagram of projecting the point cloud data to be denoised of a set of houses onto the horizontal plane is shown. From the occupancy grid map obtained by projecting the point cloud data to be denoised of a set of houses onto the horizontal plane, it can be seen that the white pixel points are mostly connected. Figure 4 It can be seen that most of the white pixel points are connected. The area composed of these connected white pixel points belongs to the indoor layout of the house, i.e. belongs to the specified area. There are also some scattered and disconnected white pixel points outside the specified area. These white pixel points outside the specified area are connected domains that the collection track does not pass through, and belong to invalid point cloud data. In step S330, the connected domains that the collection track does not pass through are determined as point cloud data outside the specified area. In step S340, the point cloud data outside the specified area is filtered out from at least one connected domain. Figure 5 An effect diagram after filtering out the point cloud data outside the specified area is shown. Referring to Figure 4 and Figure 5 By comparing the connected domains before and after filtering, it can be seen that the point cloud data becomes clean and neat after filtering out the invalid point cloud clusters.

[0129] It is very important to analyze the connected domain in the occupancy grid map, and the analysis result can be used to better understand and process the environmental information. By identifying the connected domain, the position and shape of each component in the specified area can be more accurately determined, and then a good filtering effect can be achieved on the to-be-de-noised point cloud data.

[0130] In an embodiment, Figure 1 In step S130, the mirror surface range in the specified area is detected, including:

[0131] Detecting a light spot with an intensity value greater than a preset first intensity threshold from the to-be-de-noised point cloud data;

[0132] Determining the mirror surface range in the specified area based on the intensity values of the point clouds in the surrounding area of the light spot.

[0133] The mirror surface range can include the surface area of a mirror, glass, or other reflective object. For the mirror surface range, if the laser is vertically incident to the mirror surface range, the reflected light in this case is a high-intensity echo. However, if the laser is not vertically incident to the mirror surface range, that is, in the case of oblique incidence, the intensity of the reflected light is very low. In an example, only in the case of strict vertical incidence, such as when the angle between the incident light and the plane of the mirror surface is 89 degrees, 90 degrees, or 91 degrees, a high-intensity return wave can be obtained. If a light spot with a very high intensity is found in a region, and the light intensity in the surrounding area of the light spot is very weak, it can be determined that the light spot and the surrounding area belong to the mirror surface range in this case.

[0134] In an embodiment, the mirror surface range can be determined using the mirror frame point cloud, and then the point cloud data collected from the inside of the mirror surface range can be deleted. In another embodiment, if it is a frameless mirror attached to a wall or a cabinet, the intensity characteristics of the mirror reflection are used to find the mirror frame point cloud and determine the mirror surface range. Specifically, a preset first intensity threshold is set in advance, and a light spot with an intensity value greater than the preset first intensity threshold is detected from the to-be-de-noised point cloud data. The light spot is a high-intensity echo produced by the vertical incidence of laser to the mirror surface. Then, the mirror frame position is calculated based on the intensity values of the point clouds in the surrounding area of the light spot, and the mirror surface range in the specified area is determined.

[0135] Specifically, in the mirror range, the intensity of the reflected light is very low except for the light spot. In the wall area outside the mirror range, the intensity of the reflected light is normal. For example, if a mirror is pasted on the wall, the intensity of the reflected echo on the wall can be 0.8, the intensity of the reflected echo of the light spot generated by the vertical reflection of the mirror can be 1, and the intensity of the reflected echo of the point reflection in the mirror area around the light spot can be 0.1 or 0.01, or even zero. However, in the wall area outside the mirror range, the intensity of the reflected light is normal. That is, there is a light spot with particularly high intensity in the middle of the mirror range, and the intensity of the light is particularly low in other areas of the mirror range. According to the above mirror reflection characteristics, the frame position of the mirror area can be calculated.

[0136] As shown in Figure 6 , the frame calculation method is to emit a line from the position of the high-intensity light spot along the plane by 360 degrees, pass through the area with very low light intensity to find the frame point cloud, that is, to find the position with normal intensity of the wall, to obtain a polygon ring. The boundary of the polygon ring is the frame of the mirror. The intensity of the reflected light outside the polygon ring becomes normal, that is, the normal wall. Referring to Figure 6 , the middle light spot is the reflected light with the highest intensity, the intensity of the reflected light in the range of the radiation line is very low, and the boundary of the polygon ring is the boundary of the mirror and the wall. 360 rays can be emitted around the light spot along the plane by 360 degrees to find the frame, that is, to find the position with normal intensity of the wall, that is, the boundary between the mirror and the wall. Connecting the 360 vertices found to form a ring forms an enclosed internal area, which is the mirror range.

[0137] In an example, after deleting the point cloud data collected from the inside of the mirror range, a hole is obtained on the wall. The hole can be filled back by remedial treatment, for example, some points can be obtained by using uniformly sampled point cloud data to fill the mirror range, so as to obtain the wall again.

[0138] In addition, the detection accuracy of the mirror can also be enhanced by pasting auxiliary reflective stickers. The characteristics of the reflective stickers are that high-intensity echoes can be obtained at any angle of shooting, which facilitates the detection of high-intensity light spots. For a mirror, a high-intensity light spot can only be generated under very strict perpendicular incidence. Therefore, when collecting point cloud data, the mirror may not be vertically incident when it is just scanned, resulting in missed points and failure to collect the mirror. In view of this situation, auxiliary reflective stickers can be pasted so that a high-intensity echo can be obtained at any angle of shooting, which facilitates the detection of high-intensity light spots. The high-intensity light spot can be used to locate the position of the mirror. In this way, it can be ensured that the mirror and other reflective objects are detected and will not be missed, further improving the accuracy of the detection.

[0139] In the embodiments of the present disclosure, the mirror, glass and other objects with excellent reflection characteristics can be accurately detected through the mirror surface detection, so that the mirror interference noise points caused by the above-mentioned reflective objects can be more accurately filtered out, and the filtered point cloud data is clean enough to achieve good filtering effect.

[0140] In an embodiment, Figure 1 In step S140 in the embodiment, the point cloud data collected from inside the mirror surface range is deleted to obtain the denoised point cloud data, and the step further includes:

[0141] After the point cloud data collected from inside the mirror surface range is deleted, the point cloud noise points inside the wall in the specified region are filtered to obtain the denoised point cloud data.

[0142] The point cloud is a set of point data of the appearance surface of an object in the collected environment obtained by a point cloud collection device. In theory, the point cloud inside the wall cannot be scanned, but there may be various reasons causing the noise point cloud in the middle of the wall. Therefore, after the invalid point cloud data outside the specified region is filtered based on the collection trajectory in step S120 and the point cloud data collected from inside the mirror surface range is deleted in step S140, the point cloud noise points inside the wall in the specified region can also be filtered. In the denoising process, the invalid point cloud data is removed in step S120, the mirror interference noise points are deleted in step S140, and the noise points inside the wall are filtered. The execution order of the above three steps can be interchanged. After the noise points inside the wall are filtered, the obtained point cloud data is cleaner, and the point cloud denoising effect is further improved.

[0143] As shown in Figure 7 In an embodiment, filtering the point cloud noise points inside the wall in the specified region includes:

[0144] In step S410, the point cloud data after the deletion processing is subjected to plane segmentation processing, and a plurality of wall planes in the specified region are obtained based on the result of the plane segmentation processing.

[0145] Firstly, the point cloud data after the deletion processing is subjected to plane segmentation processing by using a plane segmentation algorithm of point cloud, and the points in the point cloud data are distributed to clusters belonging to the same plane. Then, the normal, length and height information of the plane obtained by the plane segmentation processing are used to obtain a plurality of wall planes in the specified region. For example, the plane with the normal along the horizontal plane, the length and the height exceeding a certain threshold value can be determined as a wall plane. All the wall planes in the specified region are filtered based on the above method.

[0146] An exemplary plane segmentation algorithm of point cloud includes the following steps:

[0147] 1) Preprocessing of point cloud data: including data filtering, downsampling, etc.

[0148] 2) Plane detection: using appropriate algorithms, such as Random Sample Consensus (RANSAC) algorithm or Region Growing algorithm, etc., to detect planes in point cloud data. These algorithms find specific plane models by fitting or clustering samples or point sets in data.

[0149] 3) Plane parameter estimation: after finding the plane, estimate its parameters. Usually use least squares method or method of minimizing residual sum of squares to fit the representation and description of the plane, and further utilize and analyze.

[0150] 4) Plane segmentation: according to the plane model and corresponding parameters, segment the point cloud data according to the plane, and assign points to clusters belonging to the same plane. This can be achieved by calculating the distance or angle between points and planes.

[0151] Step S420, among the plurality of wall planes, two wall planes that are parallel to each other and have a distance satisfying a preset condition are determined as point cloud planes on two sides of the same wall.

[0152] Specifically, the plurality of wall planes are matched with each other to find point cloud planes on two sides of the same wall, forming a pair. For example, two planes that are parallel to each other and have a distance approximately equal to the thickness of the wall are determined as point cloud planes on two sides of the same wall.

[0153] Step S430, the point cloud noise in the middle of the point cloud planes on two sides of the same wall is removed.

[0154] In this embodiment, the wall in the specified area can be accurately detected through plane segmentation and wall pairing, and the non-real point cloud noise in the wall can be removed, so that the point cloud data is cleaner.

[0155] In one embodiment, the point cloud noise in the middle of the point cloud planes on two sides of the same wall is removed, comprising:

[0156] Projecting the point cloud data in the middle of the point cloud planes on two sides of the same wall onto a horizontal plane to obtain a projection image;

[0157] Removing the point cloud data corresponding to the area in the projection image where the ground does not exist.

[0158] The existence of the door hole needs to be considered when removing the point cloud noise in the middle of the wall. The point cloud on the upper edge, the edge and the ground below the door hole is expected to be retained. The point cloud data in the middle of the point cloud planes on both sides of the same wall can be projected onto the horizontal plane to obtain a projection image. Referring to Figure 5 The effect diagram of the projection of the point cloud data on the horizontal plane is shown. The place with the door hole must have the ground after projection, and the projection of the wall has no ground. For a wall without a door, the wall is projected onto the horizontal plane, and there is no ground area. After the two wall planes are paired, if some points are found in the middle of the two walls, these points can be door holes, grounds, door frames or door beams, and these points are not removed and need to be retained.

[0159] Based on the above characteristics of the point cloud data, it can be determined whether there is ground in the area after the projection of the two paired wall planes. The place with the ground is the door hole and needs to be retained. The point cloud data in the place without ground and not the door hole is considered to be some noise in the middle of the wall. Specifically, a wall is projected to obtain a long rectangle. If there is no door hole in the projection area, the points in the entire rectangular area are removed as noise. If there is a door hole, the rectangular area will be connected to a piece of ground point. Only the area without ground is removed, and the door structure can be retained.

[0160] In the embodiments of the present application, through the analysis of the projection image of the point cloud data, not only the noise between the walls can be removed, but also the door structure on the wall can be retained, so that the point cloud data after denoising is more real and clean, and a more accurate denoising effect is achieved.

[0161] In one embodiment, based on the collection track, before the point cloud data outside the specified area is filtered from the point cloud data to be denoised, the method further includes:

[0162] Obtaining the shooting distance of the point from the point cloud data to be denoised;

[0163] Filtering the points in the point cloud data to be denoised whose shooting distance is greater than or equal to a preset second distance threshold.

[0164] The points in the point cloud data can all carry emission point information, and the shooting distance of the points can be obtained according to the emission point information. When one or more laser beams irradiate the surface of an object, the reflected laser beams carry information such as direction and distance. The shooting distance of the points in the point cloud data can be determined according to the direction and distance of the source of the reflected light. The shooting distance can be used to perform basic filtering on the point cloud data. A second distance threshold is set in advance for the shooting scene, and points with a shooting distance greater than or equal to the preset second distance threshold in the to-be-de-noised point cloud data are filtered out. On the basis of the above basic filtering, further de-noising processing is performed, so that the effect of the final de-noised result is better.

[0165] In an implementation, before the points outside the specified region are filtered from the to-be-de-noised point cloud data based on the collection trajectory, the method further includes:

[0166] An intensity value of the point is obtained from the to-be-de-noised point cloud data.

[0167] The points with an intensity value less than or equal to a preset second intensity threshold in the to-be-de-noised point cloud data are filtered out.

[0168] Each point in the point cloud data can carry an intensity value, and the intensity value can be used to perform basic filtering on the point cloud data. A second intensity threshold can be set in advance, that is, the effective range of the point cloud intensity value is determined. In general cases, except for strict perpendicular incidence, the point cloud reflected by glass or a mirror surface often has the characteristic of very low intensity. In an example, the second intensity threshold can be set to 0.5. The points with an intensity value less than or equal to the preset second intensity threshold in the to-be-de-noised point cloud data are filtered out. On the basis of the above basic filtering, further de-noising processing is performed, so that the effect of the final de-noised result is better.

[0169] In an implementation, before the points outside the specified region are filtered from the to-be-de-noised point cloud data based on the collection trajectory, the method further includes:

[0170] A k-nearest neighbor average distance of the points in the to-be-de-noised point cloud data is counted.

[0171] The points with a k-nearest neighbor average distance greater than or equal to a preset third distance threshold in the to-be-de-noised point cloud data are filtered out. The preset third distance threshold is determined based on the standard deviation of the k-nearest neighbor average distances of all points in the to-be-de-noised point cloud data.

[0172] The k-nearest neighbor average distance of each point in the point cloud refers to the average Euclidean distance between each point in the point cloud and its nearest k neighboring points. Based on the k-nearest neighbor average distance, the point cloud data can be statistically processed, and the statistical information of the point cloud can be used to perform basic filtering on the point cloud data. Specifically, the k-nearest neighbor average distance of the points in the point cloud data to be denoised can be calculated, and the average value and the standard deviation of the k-nearest neighbor average distance of all points in the point cloud data to be denoised can be obtained. A third distance threshold can be set in advance, for example, the third distance threshold can be set to n times the standard deviation. The points in the point cloud data to be denoised whose k-nearest neighbor average distance is greater than or equal to the preset third distance threshold are determined as outliers. These outliers are far away from the neighboring points, and they are removed according to this characteristic. On the basis of the above basic filtering, further subsequent denoising processing is performed, which can make the effect of the final denoising result better.

[0173] In the denoising process, the execution order of the steps of removing invalid point cloud data in step S120, deleting mirror interference noise points in step S140, and filtering noise points inside the wall can be interchanged. In one example, the execution order of each of the above steps can also be set according to the characteristics of the point cloud noise in a specific scene. For example, if the proportion of invalid point cloud data in the point cloud noise is relatively large, the step S120 of removing invalid point cloud data can be executed first, and then the other denoising steps can be executed.

[0174] In yet another example, the basic filtering can be performed first, and then the invalid point cloud data can be removed, the mirror interference noise points can be deleted, and the noise points inside the wall can be filtered. The specific steps are as follows:

[0175] 1) Direct filtering using basic information of the point cloud. The basic information of the point cloud can include emission point information and intensity information of the points in the point cloud.

[0176] 2) Filtering using statistical information of the point cloud. The statistical information of the point cloud can include the average value and the standard deviation of the k-nearest neighbor average distance of all points in the point cloud data.

[0177] 3) Removing invalid point cloud clusters: based on the collection trajectory, invalid point cloud data outside the specified area is filtered.

[0178] 4) Deleting mirror interference noise points: point cloud data collected from inside the mirror range is deleted.

[0179] 5) Filtering point cloud using wall and other structure information: filtering point cloud noise inside the wall in the specified area.

[0180] In summary, the embodiments of the present disclosure propose a complete set of filtering algorithms for the laser point cloud data collected in an indoor scene. On the basis of filtering based on point cloud intensity information and point cloud statistical information, the point cloud data in the non-collection area is further filtered using the collection trajectory, and a method for detecting the mirror surface range and filtering the mirror point cloud is proposed. Finally, the noise points inside the wall are filtered, so that the filtered point cloud data is clean and neat. The above multi-layer processing through a series of filtering methods can achieve good filtering effect of the point cloud data collected in the indoor scene.

[0181] Figure 8 and Figure 9 respectively show the point cloud images before and after the denoising processing based on the point cloud denoising method according to the embodiments of the present disclosure. Referring to Figure 8 and Figure 9 By comparing the effect images before and after the denoising processing, it can be seen that after the denoising processing, the unreal and unnecessary points are obviously removed, so that the point cloud data becomes clean and neat.

[0182] As shown in Figure 10 , the present disclosure also provides an embodiment of a corresponding point cloud denoising device. The beneficial effects or technical problems solved by the device can be referred to the description in the method corresponding to each device, or referred to the description in the summary, which will not be repeated here.

[0183] In the embodiment of the point cloud denoising device, the device comprises:

[0184] The acquisition unit 100 is configured to acquire the to-be-denoised point cloud data, the to-be-denoised point cloud data being point cloud data collected in a specified area based on a preset collection trajectory;

[0185] The first processing unit 200 is configured to perform filtering processing on the point cloud data outside the specified area from the to-be-denoised point cloud data based on the collection trajectory;

[0186] The detection unit 300 is configured to detect a mirror surface range in the specified area;

[0187] The second processing unit 400 is configured to perform deletion processing on the point cloud data collected from the inside of the mirror surface range in the point cloud data after the filtering processing, to obtain denoised point cloud data.

[0188] In an implementation manner, the first processing unit 200 is configured to:

[0189] perform clustering processing on the point cloud data based on the distance between the nearest neighbors, to obtain at least one point cloud cluster;

[0190] obtain a first distance between the at least one point cloud cluster and the collection trajectory;

[0191] The point cloud cluster with the first distance greater than or equal to a preset first distance threshold is determined as the point cloud data outside the specified area.

[0192] The point cloud data outside the specified area is filtered from the at least one point cloud cluster.

[0193] In an embodiment, the first processing unit 200 is configured to:

[0194] Project the to-be-de-noised point cloud data onto a horizontal plane to obtain an occupancy grid map corresponding to the to-be-de-noised point cloud data.

[0195] Obtain at least one connected domain in the occupancy grid map.

[0196] The connected domain not passed through by the collection track is determined as the point cloud data outside the specified area.

[0197] The point cloud data outside the specified area is filtered from the at least one connected domain.

[0198] In an embodiment, the detection unit 300 is configured to:

[0199] Detect a light spot with an intensity value greater than a preset first intensity threshold from the to-be-de-noised point cloud data.

[0200] Determine a mirror surface range within the specified area based on intensity values of point clouds in a surrounding area of the light spot.

[0201] As shown in FIG. Figure 11 In an embodiment, the apparatus further includes a third processing unit 500 configured to:

[0202] After deleting the point cloud data collected from inside the mirror surface range, filter the point cloud noise inside the wall within the specified area to obtain de-noised point cloud data.

[0203] As shown in FIG. Figure 12 In an embodiment, the third processing unit 500 includes:

[0204] A segmentation sub-unit 510 configured to perform plane segmentation processing on the point cloud data after the deletion processing, and obtain a plurality of wall planes within the specified area based on a result of the plane segmentation processing.

[0205] A matching sub-unit 520 configured to determine, among the plurality of wall planes, two wall planes that are parallel to each other and have a distance satisfying a preset condition as point cloud planes on two sides of a same wall.

[0206] A culling sub-unit 530 configured to cull point cloud noise in the middle of the point cloud planes on two sides of the same wall.

[0207] In an embodiment, the culling sub-unit 530 is configured to:

[0208] Projecting the point cloud data in the middle of the point cloud planes on both sides of the same wall onto a horizontal plane to obtain a projection image;

[0209] Discarding the point cloud data corresponding to the area in the projection image where the ground does not exist.

[0210] As Figure 13 shown in an embodiment, the apparatus further includes a fourth processing unit 600 configured to:

[0211] Obtaining the shooting distance of a point from the point cloud data to be denoised;

[0212] Filtering out the point in the point cloud data to be denoised whose shooting distance is greater than or equal to a preset second distance threshold.

[0213] In an embodiment, the apparatus further includes a fourth processing unit 600 configured to:

[0214] Obtaining the intensity value of a point from the point cloud data to be denoised;

[0215] Filtering out the point in the point cloud data to be denoised whose intensity value is less than or equal to a preset second intensity threshold.

[0216] In an embodiment, the apparatus further includes a fourth processing unit 600 configured to:

[0217] Statistically obtaining the k-nearest neighbor average distance of a point from the point cloud data to be denoised;

[0218] Filtering out the point in the point cloud data to be denoised whose k-nearest neighbor average distance is greater than or equal to a preset third distance threshold, wherein the preset third distance threshold is determined based on the standard deviation of the k-nearest neighbor average distances of all points in the point cloud data to be denoised.

[0219] The point cloud denoising apparatuses of the embodiments of the present disclosure and the point cloud denoising methods of the embodiments of the present disclosure correspond to each other in terms of specific implementation and beneficial technical effects, and related contents can be mutually referred to, which will not be described here again.

[0220] Below, an electronic device according to embodiments of the present disclosure is described with reference to Figure 14 The electronic device can be either or both of the first device and the second device, or a single device independent of them, which can communicate with the first device and the second device to receive the collected input signals therefrom.

[0221] Figure 14 Fig. 1 illustrates a block diagram of an electronic device according to embodiments of the present disclosure.

[0222] As Figure 14 shown, the electronic device includes one or more processors and a memory.

[0223] The processor can be a central processing unit (CPU) or other form of processing unit having data processing and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions.

[0224] The memory can store one or more computer program products, and can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM), cache memory, and / or the like. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, and / or the like. One or more computer program products can be stored on the computer-readable storage media, and the processor can execute the computer program products to implement the point cloud denoising method of various embodiments of the present disclosure described above and / or other desired functions.

[0225] In one example, the electronic device can further include an input device and an output device, which are interconnected through a bus system and / or other forms of connection mechanism (not shown).

[0226] In addition, the input device can further include, for example, a keyboard, a mouse, and / or the like.

[0227] The output device can output various information, including determined distance information, direction information, and / or the like, to the outside. The output device can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and / or the like.

[0228] Of course, in order to simplify, Figure 14 Only some of the components in the electronic device related to the present disclosure are shown in FIG. 1, and components such as buses, input / output interfaces, and / or the like are omitted. In addition, the electronic device can further include any other appropriate components according to specific application cases.

[0229] In addition to the above-described method and device, embodiments of the present disclosure can be a computer program product including computer program instructions, which, when executed by a processor, cause the processor to perform the steps of the point cloud denoising method according to various embodiments of the present disclosure described in the above part of the specification.

[0230] A computer program product can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server.

[0231] Furthermore, embodiments of the present disclosure can also be a computer readable storage medium, having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the steps described in the foregoing disclosure of the point cloud denoising method according to various embodiments of the present disclosure.

[0232] The computer readable storage medium can be any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the above. More specific examples (a non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0233] The above describes the basic principles of the present disclosure in combination with specific embodiments, but it should be noted that the advantages, advantages, effects and the like mentioned in the present disclosure are only examples and are not limiting, and these advantages, advantages, effects and the like cannot be considered as the various embodiments of the present disclosure must have. In addition, the above specific details are only for the purpose of example and for the purpose of understanding, and the above details do not limit the present disclosure to the above specific details.

[0234] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between each embodiment can be referred to each other. For system embodiments, since they basically correspond to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0235] The block diagrams of devices, apparatuses, equipment, systems referred to in this disclosure are merely illustrative examples and are not intended to require or imply that the connection, arrangement, configuration must be as shown in the block diagrams. These devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner as will be appreciated by those skilled in the art. Words such as "include," "contain," "have," and the like are open-ended words that are to be interpreted to mean "including but not limited to," and are not to be interpreted as limiting the described embodiment to features, elements, and / or steps disclosed herein. The words "or" and "and" as used herein are to be interpreted as the word "and / or," and are not to be interpreted as requiring both features, elements, and / or steps disclosed herein. The word "such as" as used herein is to be interpreted as the phrase "such as but not limited to," and is not to be interpreted as limiting the described embodiment to features, elements, and / or steps disclosed herein.

[0236] The methods and apparatuses of this disclosure can be implemented in a number of ways. For example, the methods and apparatuses of this disclosure can be implemented using software, hardware, firmware, or any combination of these. The above described order of steps for the methods is merely illustrative, and the steps of the methods of this disclosure are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, the disclosure can also be implemented as a program recorded in a recording medium, which includes machine readable instructions for implementing the methods according to the disclosure. Thus, the disclosure also covers a recording medium storing a program for executing the methods according to the disclosure.

[0237] It is also important to note that the devices, equipment, and methods of this disclosure can be embodied in a variety of ways. These variations are contemplated as being within the scope of the present disclosure.

[0238] The above description of the disclosed aspects is given for illustrative purposes and is not intended to limit the scope of the disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0239] The above description has been given for illustrative and descriptive purposes. In addition, this description is not intended to limit embodiments of the disclosure to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain modifications, permutations, additions, and sub-combinations thereof.

Claims

1. A method of point cloud denoising, the method comprising: The method comprises the following steps: acquiring to-be-de-noised point cloud data, the to-be-de-noised point cloud data being point cloud data collected in a specified region based on a preset collection track; based on the collection track, performing filtering processing on point cloud data outside the specified region from the to-be-de-noised point cloud data; detecting a mirror surface range in the specified region; in the point cloud data after the filtering processing, performing deletion processing on point cloud data collected from inside the mirror surface range, the point cloud data collected from inside the mirror surface range including mirror image point cloud collected by light passing through the inside of the mirror surface range, to obtain de-noised point cloud data; the deletion processing on the point cloud data collected from inside the mirror surface range includes: after the deletion processing on the point cloud data collected from inside the mirror surface range, projecting point cloud data in the middle of point cloud planes on both sides of a same wall onto a horizontal plane to obtain a projection image; performing rejection on point cloud data corresponding to a region without ground in the projection image and performing reservation on point cloud data corresponding to a region with ground in the projection image to obtain de-noised point cloud data; wherein the point cloud data corresponding to the region without ground is noise inside a wall without a door, and the point cloud data corresponding to the region with ground is a door structure in the wall.

2. The method of claim 1, wherein, the filtering processing on the point cloud data outside the specified region from the to-be-de-noised point cloud data based on the collection track includes: performing clustering processing on point cloud data based on nearest neighbor point distance to obtain at least one point cloud cluster; acquiring a first distance between the at least one point cloud cluster and the collection track; determining a point cloud cluster with a first distance greater than or equal to a preset first distance threshold as point cloud data outside the specified region; performing filtering processing on the point cloud data outside the specified region from the at least one point cloud cluster.

3. The method of claim 1, wherein, the filtering processing on the point cloud data outside the specified region from the to-be-de-noised point cloud data based on the collection track includes: projecting the to-be-de-noised point cloud data onto a horizontal plane to obtain an occupancy grid map corresponding to the to-be-de-noised point cloud data; acquiring at least one connected domain in the occupancy grid map; determining a connected domain not passed through by the collection track as point cloud data outside the specified region; performing filtering processing on the point cloud data outside the specified region from the at least one connected domain.

4. The method of claim 1, wherein, the detection of the mirror surface range in the specified region includes: detecting a light spot with an intensity value greater than a preset first intensity threshold from the to-be-de-noised point cloud data; based on the intensity value of point cloud in a surrounding region of the light spot, determining a mirror surface range in the specified region.

5. The method of claim 1, wherein, The method further comprises: performing plane segmentation processing on the point cloud data after the deletion processing, and obtaining a plurality of wall planes in the specified region based on a result of the plane segmentation processing; in the plurality of wall planes, determining two wall planes that are parallel to each other and have a distance satisfying a preset condition as the point cloud planes on both sides of the same wall.

6. The method according to any one of claims 1 to 4, characterized in that, Before the filtering of the point cloud data outside the specified region from the to-be-denoised point cloud data based on the collection trajectory, the method further includes: acquiring a shooting distance of a point from the to-be-denoised point cloud data; filtering out a point in the to-be-denoised point cloud data whose shooting distance is greater than or equal to a preset second distance threshold.

7. The method according to any one of claims 1 to 4, characterized in that, Before the filtering of the point cloud data outside the specified region from the to-be-denoised point cloud data based on the collection trajectory, the method further includes: acquiring an intensity value of a point from the to-be-denoised point cloud data; filtering out a point in the to-be-denoised point cloud data whose intensity value is less than or equal to a preset second intensity threshold.

8. The method according to any one of claims 1 to 4, characterized in that, Before the filtering of the point cloud data outside the specified region from the to-be-denoised point cloud data based on the collection trajectory, the method further includes: counting a k-nearest neighbor average distance of a point in the to-be-denoised point cloud data; filtering out a point in the to-be-denoised point cloud data whose k-nearest neighbor average distance is greater than or equal to a preset third distance threshold; wherein the preset third distance threshold is determined based on a standard deviation of the k-nearest neighbor average distances of all points in the to-be-denoised point cloud data.

Images