Rapid edge smoothing method and system based on three-dimensional point cloud information

Abstract

The invention relates to a rapid edge smoothing method based on three-dimensional point cloud information, and the method comprises the steps: obtaining the three-dimensional point cloud information of a to-be-detected object, carrying out the edge smoothing of a foreground of the extracted three-dimensional point cloud information of the to-be-detected object through bilateral filtering, carrying out the rapid bilateral filtering of the smoothed foreground of the to-be-detected object, and carrying out the edge smoothing of the to-be-detected object. According to the technical scheme, the foreground is accurately extracted based on the three-dimensional point cloud, the rapid bilateral filtering is adopted, compared with a traditional mode, the time efficiency is greatly improved, the processing speed is increased, the effect is equivalent, bilateral filtering can be replaced, the noise generated by jump of depth information is removed through a statistical method, and the accuracy of the detection result is improved. The positioning precision of the specific point of the to-be-detected object is improved, and the product fluency, the system stability and the robustness are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of three-dimensional structure signal processing, and in particular to a method and system for fast edge smoothing based on three-dimensional point cloud information. BACKGROUND

[0002] With the development of computer vision technology and artificial intelligence technology, using computer vision technology to detect and locate human meridians to realize the digital treatment of traditional Chinese medicine has become a very promising exploration direction. Since the accurate detection of human meridians is of great significance for analyzing the physiology, pathology and treatment of diseases of the human body, the SY-4 digital Chinese medicine meridian treatment system independently developed by Zhongke Shangyi Health Technology (Beijing) Co., Ltd. needs to use a single camera and quickly and accurately detect the meridians of the user during the conditioning process to ensure the smoothness and integrity of the conditioning process and bring a good experience to the user.

[0003] With the development of three-dimensional point cloud information, a problem encountered in the development of the digital Chinese medicine meridian treatment system SY-4 of Zhongke Shangyi Health Technology (Beijing) Co., Ltd. is that the depth information of the actual camera is jump-like, and the obtained three-dimensional point cloud information inevitably has noise at the edge of the person, especially in the head, neck and waist, which will directly affect the accuracy of locating the meridians. In order to overcome this problem, a method and system for fast edge smoothing based on three-dimensional point cloud information are proposed, which can quickly smooth the edge and improve the accuracy of locating the specific points on the object to be measured. SUMMARY

[0004] Therefore, the present application proposes a method for fast edge smoothing of human three-dimensional point cloud information, which is suitable for a Chinese medicine meridian treatment robot, and has the characteristics that it includes the following steps:

[0005] obtaining initial three-dimensional point cloud information of the human body to be measured by scanning with a robot;

[0006] performing edge smoothing on the initial three-dimensional point cloud information of the human body using bilateral filtering to obtain first preprocessed three-dimensional point cloud information;

[0007] processing the first preprocessed three-dimensional point cloud information using fast bilateral filtering to obtain second preprocessed three-dimensional point cloud information;

[0008] removing the noise of the head of the human body in the second preprocessed three-dimensional point cloud information to obtain smoothed three-dimensional point cloud information of the object to be measured for subsequent processing by the Chinese medicine meridian treatment robot.

[0009] In one possible implementation, processing the first preprocessed three-dimensional point cloud information using the fast bilateral filtering method to obtain the second preprocessed three-dimensional point cloud information includes the following steps:

[0010] extracting an outer edge of the foreground of the first pre-processed three-dimensional point cloud information using statistical filtering to obtain an outer edge processed three-dimensional point cloud information;

[0011] performing bilateral filtering on the extracted outer edge;

[0012] fusing the result of the bilateral filtering with the remaining inner information extracted using statistical filtering to obtain the second pre-processed three-dimensional point cloud information.

[0013] In one possible implementation, the step of extracting the outer edge of the foreground of the first pre-processed three-dimensional point cloud information using statistical filtering comprises the steps of:

[0014] performing neighborhood search on each point in the first pre-processed three-dimensional point cloud;

[0015] determining whether the point cloud block searched by the neighborhood search conforms to a preset Gaussian distribution of a given mean and variance;

[0016] if the point cloud block does not conform to the preset Gaussian distribution, regarding the point cloud block as a noise point and removing the noise point to finally extract the outer edge processed three-dimensional point cloud information.

[0017] In one possible implementation, the given mean is set by a user.

[0018] In one possible implementation, the step of removing the noise of the head of the human body to be detected after the fast bilateral filtering comprises:

[0019] projecting the x-axis of the second pre-processed three-dimensional point cloud information on the y-axis of the three-dimensional point cloud information;

[0020] counting the number of three-dimensional points on the y-axis of the second pre-processed three-dimensional point cloud information;

[0021] dividing the projection into a plurality of regions, counting the number of three-dimensional points in different regions; and

[0022] determining whether a specified region is a noise according to the number of three-dimensional points in the specified region.

[0023] In one possible implementation, the number of the plurality of regions is at least 100.

[0024] In one possible implementation, the range of the specified region where the noise exists is set by a user.

[0025] In one possible implementation, the specified region is determined as a noise when the number of three-dimensional points in the specified region is less than a point number threshold.

[0026] In one possible implementation, the point number threshold is set by a user.

[0027] Furthermore, the present application also relates to a computer program product comprising instructions for implementing the above method.

[0028] Advantages of the present application:

[0029] The technical solution of the present application extracts the foreground based on three-dimensional point cloud information accurately and adopts fast bilateral filtering, which greatly improves the time efficiency compared with the traditional bilateral filtering, greatly improves the processing speed, and the effect is basically the same, which can effectively replace the bilateral filtering; secondly, the statistical method is used to remove part of the noise generated by the depth information jump at the specific point on the object to be measured, which significantly improves the positioning accuracy of the specific point on the object to be measured, avoids noise interference, enhances the smoothness of product conditioning, improves the system stability and robustness, and optimizes the overall application effect.

[0030] Other features and aspects of the present application will become apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the present application and serve to explain the principles of the present application.

[0032] Figure 1 A method flowchart of fast edge smoothing based on three-dimensional point cloud information of the embodiment of the present application is shown;

[0033] Figure 2 A flowchart of the fast bilateral filtering method of the embodiment of the present application is shown;

[0034] Figure 3 A method flowchart of statistical filtering of the embodiment of the present application is shown;

[0035] Figure 4 A flowchart of removing noise of the head of the human body of the embodiment of the present application is shown. DETAILED DESCRIPTION

[0036] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference signs in the drawings represent functionally identical or similar elements. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

[0037] It should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like are used herein to describe the orientation or position of the device or element as shown in the drawings, and are used only for convenience in describing the present application or simplifying the description, and do not indicate or imply that the device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore should not be construed as limiting the present application.

[0038] In addition, the terms "first", "second", "third", etc. are used only for descriptive purposes and should not be construed as indicating or implying relative importance or a required or preferred order of steps or elements. Thus, features defined with "first", "second" or "third" can include one or more of the features, explicitly or implicitly.

[0039] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations.

[0040] In addition, in order to better illustrate the present application, a large number of specific details are given in the specific embodiments below. Those skilled in the art should understand that the present application can also be implemented without some specific details. In some examples, methods, means, elements and circuits that are well known to those skilled in the art are not described in detail, in order to highlight the main ideas of the present application.

[0041] With specific reference to Figure 1 , Figure 1 A flow chart of a method for fast edge smoothing based on three-dimensional point cloud information according to an embodiment of the present application is shown.

[0042] The application of the present application is a method for fast edge smoothing based on three-dimensional point cloud information, which is applied in the field / equipment of stereo structure signal processing, and plays a role in accelerating bilateral filtering and efficiently and accurately removing noise. A method for fast edge smoothing of human three-dimensional point cloud information is suitable for a traditional Chinese medicine meridian treatment robot, and characterized in that it comprises the following steps: 101, obtaining initial three-dimensional point cloud information of a human body to be measured by scanning through a robot; 102, performing edge smoothing on the initial three-dimensional point cloud information of the human body by using bilateral filtering to obtain first preprocessed three-dimensional point cloud information; 103, processing the first preprocessed three-dimensional point cloud information by using fast bilateral filtering to obtain second preprocessed three-dimensional point cloud information; and 104, removing noise of the head of the human body in the second preprocessed three-dimensional point cloud information to obtain smoothed three-dimensional point cloud information of the object to be measured for subsequent processing of the traditional Chinese medicine meridian treatment robot. The method realizes fast edge smoothing and accurate denoising of human three-dimensional point cloud, provides high-quality point cloud data for the traditional Chinese medicine meridian treatment robot, guarantees the accuracy of subsequent meridian positioning and treatment operation, and improves the running fluency and stability of the robot system. The fast bilateral filtering processing only performs time-consuming bilateral filtering on the edge region, greatly reduces the calculation amount, greatly improves the processing speed, and the filtering effect is basically the same as that of the traditional bilateral filtering. The fast bilateral filtering processing can efficiently replace the traditional method and greatly improve the time efficiency of the system.

[0043] With specific reference Figure 2 , Figure 2 A flowchart of the fast bilateral filtering method of the embodiment of the present application is shown.

[0044] In a possible implementation, processing the first preprocessed three-dimensional point cloud information by using the fast bilateral filtering method to obtain the second preprocessed three-dimensional point cloud information comprises the following steps: 201, extracting the outer edge of the foreground of the first preprocessed three-dimensional point cloud information by using statistical filtering to obtain outer edge processed three-dimensional point cloud information; 202, performing bilateral filtering on the extracted outer edge; and 203, fusing the result after the bilateral filtering with the internal information remaining after the extraction by using statistical filtering to obtain the second preprocessed three-dimensional point cloud information. The information contained in the edge remaining after the extraction by using statistical filtering is fused with the result after the bilateral filtering to obtain the overall result after the bilateral filtering. Only the time-consuming bilateral filtering is performed on the edge region, the calculation amount is greatly reduced, and the processing speed is significantly improved.

[0045] Preferably, the formula for calculating the weight of the bilateral filtering is:

[0046]

[0047] In formula (1), where represents the number of nearby points for which the weight is calculated, also known as the window size, represents the number of three-dimensional points, and represents the standard deviation in the Gaussian distribution. By adaptively adjusting the window size and standard deviation, the edge features are preserved while effectively smoothing the non-edge region, improving the accuracy of subsequent processing.

[0048] w(x,y,z,i) is a weight function used to calculate the weight between point (x,y,z) and the ith point in the neighborhood in three-dimensional space. In the bilateral filtering algorithm, this weight is used to measure the similarity or influence between two points.

[0049] x, y, and z represent coordinate values in three-dimensional space, corresponding to the three dimensions in space, respectively, for determining the position of a point in three-dimensional space. i represents the index of the point in the neighborhood, i∈[1,n] represents that there are n points in the neighborhood, i takes values from 1 to n in turn, for traversing all points in the neighborhood to calculate the weight corresponding to each point. i i i represents the three-dimensional coordinate values of the ith point in the neighborhood, which is used together with the current point (x,y,z) to calculate the weight.

[0050] σ is a standard deviation parameter, usually used to control the decay rate of the weight function. The larger the value of σ, the slower the decay of the weight function, and vice versa. It affects the weight distribution of points in the neighborhood during the filtering process. j ω n is a weight vector that contains a set of weight values related to the jth point. When processing multiple points, each point corresponds to a weight vector, which is used for subsequent calculation or operation.

[0051] m1, m2,..., m j are specific weight values in the weight vector ω j , which correspond to the weights of different points in the neighborhood.

[0052] j represents the index of the weight vector, j∈[1,m] represents that there are m weight vectors, j takes values from 1 to m in turn, for traversing all weight vectors. W is a weight matrix composed of multiple weight vectors ω old . In three-dimensional point cloud processing, the weight matrix is usually used to represent the weight distribution of different points or regions for filtering, smoothing, and other operations.

[0053] I(x,y,z) new ​​represents the new value of the three-dimensional data at point (x, y, z) after filtering or processing, which is the result data after processing.

[0054] With specific reference Figure 3 , Figure 3 A method flowchart of statistical filtering of embodiments of the present application is shown.

[0055] In one possible implementation, the statistical filtering is used to extract the outer edge of the foreground of the first pre-processed three-dimensional point cloud information, including the following steps: 301, performing a neighborhood search for each point in the first pre-processed three-dimensional point cloud; 302, judging whether the point cloud block searched by the neighborhood search conforms to a preset Gaussian distribution of given mean and variance; and 303, if the point cloud block does not conform to the preset Gaussian distribution, regarding the point cloud block as a noise point and removing it, and finally extracting the outer edge of the pre-processed three-dimensional point cloud information. The outlying noise points can be quickly located, the noise removal rate is significantly improved, the point cloud data is purified, more pure basic information is provided for subsequent edge extraction and filtering, and the accuracy of the outer edge extraction is improved.

[0056] Preferably, the KD tree is used to accelerate the neighborhood search, and a neighborhood point number threshold is set. The outlying noise points can be quickly located, the noise removal rate is significantly improved, and the data purity is improved.

[0057] In one specific embodiment, the specific practice flow of the statistical filtering based on the KD tree acceleration to extract the outer edge is as follows: obtaining the first pre-processed three-dimensional point cloud data after the initial bilateral filtering processing, containing the three-dimensional coordinate (x, y, z) information of all points. The mean μ and variance σ of the preset Gaussian distribution are set. The neighborhood point number threshold K is set. The related parameters of the KD tree are constructed.

[0058] The KD tree is constructed by arranging all points in the first pre-processed three-dimensional point cloud into a data set according to the coordinate format.

[0059] The KD tree is generated by taking the x-axis as the first splitting dimension, calculating the median of the x coordinates of all points, dividing the point cloud into two subsets on the left and right, and generating a root node. For each subset, the y-axis is taken as the second splitting dimension, the median of the y coordinates is calculated for splitting, and a child node is generated. The z-axis is taken as the third splitting dimension, and the child node is recursively split until the number of points contained in each leaf node is less than the preset threshold, and the KD tree construction is completed. Finally, the structure information of the KD tree is stored, including the splitting dimension, splitting value, left and right subtree pointers of each node, and the point index contained in the leaf node.

[0060] Neighborhood search and Gaussian distribution judgment, by traversing each point P in the first pre-processed point cloud, neighborhood search is performed through KD tree: taking point P as the center, a search radius r is set, and the nearest neighbor search algorithm of KD tree is used to quickly find all neighborhood points within the radius r from P. Record the neighborhood search results: count the number of points in the neighborhood N. If N is less than the threshold K of the number of neighborhood points, mark the point as a potential noise point directly, and skip the subsequent Gaussian distribution judgment.

[0061] Wherein the Gaussian distribution compliance judgment is to calculate the mean and variance of the coordinates of the neighborhood point set N≥K, and to judge whether the neighborhood point set conforms to the preset Gaussian distribution: if the mean of the neighborhood point set is within ±2σ of the preset mean μ, and the variance is within the reasonable interval of the preset variance σ 2 , it is determined that the point set conforms to the preset Gaussian distribution; otherwise, it is determined not to conform.

[0062] For points whose neighborhood point set does not conform to the preset Gaussian distribution, mark them as noise points and remove them from the first pre-processed three-dimensional point cloud. For points whose neighborhood point set conforms to the preset Gaussian distribution, keep them as valid points.

[0063] Edge recognition is performed on the retained valid points: the normal vector difference between each valid point and the points in its neighborhood is calculated, and the points whose normal vector change rate exceeds the preset threshold are marked as edge candidate points. Connectivity analysis is performed on the edge candidate points to remove isolated edge candidate points, and the remaining points form the outer edge processing three-dimensional point cloud information. The extracted outer edge processing three-dimensional point cloud information is stored for subsequent fast bilateral filtering processing.

[0064] Through the above process, the efficient neighborhood search capability of KD tree can greatly improve the processing speed of statistical filtering, and at the same time, through strict Gaussian distribution judgment and parameter control, the accuracy of outer edge extraction is ensured.

[0065] In one possible implementation, the mean value is set by the user. The adaptability of statistical filtering to point cloud data in different scenarios is enhanced. Due to factors such as different body types, scanning distances, and environmental light, the neighborhood distribution characteristics of point clouds may differ. The user can adjust the mean value parameter according to the actual situation to ensure that the Gaussian distribution judgment is more consistent with the true distribution of the current point cloud data. For example, for the case of low density of child head point cloud, the user can appropriately lower the mean value threshold to avoid misjudging valid edge points as noise; while for the normal scanning scenario of adults, the standard mean value can be set to strictly filter noise. This setting greatly improves the accuracy of noise point identification of statistical filtering, further ensures the accuracy of outer edge extraction, provides more reliable basic data for subsequent fast bilateral filtering, and indirectly improves the stability of meridian positioning.

[0066] Specifically referring to Figure 4 , Figure 4A flowchart of removing noise of a head of a human body according to an embodiment of the present application is shown.

[0067] In a possible implementation, removing noise of the head of the human body after the fast bilateral filtering includes the following steps: 401, projecting the x-axis of the second preprocessed three-dimensional point cloud information on the y-axis of the three-dimensional point cloud information; 402, counting the number of three-dimensional points on the y-axis of the second preprocessed three-dimensional point cloud information; 403, dividing the projection into multiple regions and counting the number of three-dimensional points in different regions; and 404, judging whether a specified region is noise according to the counted number of three-dimensional points in the specified region. The noise generated by the depth information jump of the head of the human body is removed specifically, the noise interference to the meridian planning is avoided, the positioning accuracy of the specific points in the head region is significantly improved, and the accuracy of the meridian planning is improved.

[0068] Preferably, the x-axis of the three-dimensional point cloud information is projected on the y-axis; and the method for counting the number of three-dimensional points on the y-axis is as follows:

[0069] H i =∑p(x,y,z|y∈[y min ,y max )) i∈[1,100]

[0070] wherein the projection is divided into 100 regions, y min and y max represent the minimum and maximum values in the region respectively, H i represents the number of three-dimensional points counted in the ith interval. Whether it is noise is judged according to the counted H i , and the principle is as follows:

[0071]

[0072] wherein numbers represents the noise number threshold, the actual statistics is 150. The i<5 is set according to the characteristics of the region. In this embodiment, a dynamic weight factor is introduced, and the threshold calculation method is automatically adjusted according to the point cloud density. The noise point detection can be adapted to complex scenes, and the misjudgment rate is significantly reduced.

[0073] In a possible implementation, the number of the divided multiple regions is at least 100. Through fine-grained region division, the accuracy of point cloud distribution statistics is significantly improved, and noise misjudgment or omission caused by too coarse region division is avoided.

[0074] In a specific embodiment, in the complex curved surface of the head region, more than 100 regions can accurately capture the local point cloud density difference, provide more detailed basis for subsequent noise judgment, greatly improve the noise recognition accuracy, and further guarantee the accuracy of meridian positioning.

[0075] In a possible implementation, the range of the designated region where noise exists is set by the user. The adaptability of the enhancement system to different application scenarios is improved to meet individual needs.

[0076] In an embodiment, for different body types (e.g., children, adults) or special postures (e.g., low head, high head), the user can flexibly adjust the range of the head noise region, avoiding the limitations caused by fixed region settings, so that the noise removal success rate of the system in diversified scenarios remains above 95%.

[0077] In a possible implementation, when the number of three-dimensional points in the designated region where noise exists is less than a point threshold, the designated region is determined to be noise. By quantifying the point cloud density difference, objective judgment of noise is achieved, avoiding subjective experience errors.

[0078] In an embodiment, in the top region of the head, if the number of point clouds is significantly reduced due to depth jumps, the system can automatically identify and remove noise in this region, greatly improving the positioning accuracy of specific points of the head, and directly improving the accuracy of meridian planning of a traditional Chinese medicine meridian treatment robot.

[0079] In a possible implementation, the point threshold is set by the user. The adaptability of the system to different scanning conditions is improved.

[0080] In an embodiment, in a scenario where low light causes a decrease in point cloud density, the user can appropriately lower the threshold to avoid mistakenly deleting valid points; and in high-resolution scanning, the threshold can be increased to strictly filter noise. This flexibility enhances the robustness of the system in complex environments, improves the overall adaptability of noise processing by about 40%, and ensures the stability of subsequent processing.

[0081] The method of the present application accurately extracts the foreground based on three-dimensional point cloud information, then uses fast bilateral filtering for edge smoothing, then uses a statistical method to remove noise from the top of the head, and finally uses the proportion and topological relationship of the human body to infer the meridians, thereby improving the efficiency of meridian planning, enhancing the smoothness of the conditioning of the products of the company, and increasing the stability and robustness of the system. In practical applications, bilateral filtering is used to smooth the edges well. In order to accelerate bilateral filtering, we propose a method of fast bilateral filtering. The specific effects are shown in the following Table 1:

[0082] Table 1 Comparison of time efficiency of bilateral filtering and fast bilateral filtering

[0083]

[0084] From table 1, it can be seen that the time efficiency of the fast bilateral filter is improved by about 1.6 times compared with the bilateral filter. And the results output by the bilateral filter and the fast bilateral filter are basically consistent. It can be seen that the fast bilateral filter can replace the bilateral filter and improve the time efficiency.

[0085] After the fast edge smoothing, there may be noise that cannot be removed and affects the meridian planning, especially the top of the head, which will be affected by the depth information jump. At this time, we use the statistical information method to remove. Without denoising on the top of the head, if there is noise on the top of the head, it will directly affect the accuracy of the meridian planning. After denoising on the top of the head, the accuracy of the meridian planning has been significantly improved.

[0086] In addition, the present application also relates to a computer program product comprising instructions for implementing the above method. The present computer program product greatly shortens the point cloud processing time, meets the real-time demand of the system, reduces the performance dependence on the hardware device, and improves the adaptability and running efficiency of the program in different hardware environments.

[0087] Preferably, the present computer program product uses a parallel computing framework to accelerate the filtering operation and supports multi-thread processing. The processing time can be greatly shortened to meet the real-time demand and reduce the dependence on hardware.

[0088] The above has described the embodiments of the present application, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles, practical application or improvement of the technology in the market of the embodiments, or to enable other ordinary skilled persons in the art to understand the embodiments disclosed herein.

Claims

1. A method for fast edge smoothing of human three-dimensional point cloud information, suitable for a traditional Chinese medicine meridian treatment robot, characterized in that, The method comprises the steps of: acquiring initial three-dimensional point cloud information of a human body to be measured by a robot; performing edge smoothing on the initial three-dimensional point cloud information of the human body using bilateral filtering to obtain first pre-processed three-dimensional point cloud information; processing the first pre-processed three-dimensional point cloud information using fast bilateral filtering to obtain second pre-processed three-dimensional point cloud information; removing noise of a head of the human body in the second pre-processed three-dimensional point cloud information to obtain smoothed three-dimensional point cloud information of the object to be measured for subsequent processing by a traditional Chinese medicine channel treatment robot.

2. The method of claim 1, wherein, The processing of the first pre-processed three-dimensional point cloud information using fast bilateral filtering to obtain second pre-processed three-dimensional point cloud information comprises the steps of: extracting an outer edge of a foreground of the first pre-processed three-dimensional point cloud information using statistical filtering to obtain outer edge processed three-dimensional point cloud information; performing bilateral filtering on the extracted outer edge; fusing the result of the bilateral filtering with internal information remaining after the extraction using statistical filtering to obtain the second pre-processed three-dimensional point cloud information.

3. The method of claim 2, wherein, The extraction of the outer edge of the foreground of the first pre-processed three-dimensional point cloud information using statistical filtering comprises the steps of: performing neighborhood search on each point in the first pre-processed three-dimensional point cloud information; judging whether a point cloud block searched by the neighborhood search conforms to a preset Gaussian distribution of a given mean value and variance; if the point cloud block does not conform to the preset Gaussian distribution, regarding the point cloud block as a noise point and removing the noise point to finally extract the outer edge processed three-dimensional point cloud information.

4. The method of claim 3, wherein, The given mean value is set by a user.

5. The method of claim 1, wherein, The removal of the noise of the head of the human body after the fast bilateral filtering comprises: projecting the x-axis of the second pre-processed three-dimensional point cloud information on the y-axis of the three-dimensional point cloud information; counting the number of three-dimensional points on the y-axis of the second pre-processed three-dimensional point cloud information; dividing the projection into a plurality of regions, counting the number of three-dimensional points in different regions; and judging whether a specified region is noise according to the counted number of three-dimensional points in the specified region.

6. The method of claim 3, wherein, The number of the divided plurality of regions is at least 100.

7. The method of claim 4, wherein, The range of the specified region where the noise exists is set by a user.

8. The method of claim 6, wherein, When the number of three-dimensional points in the specified region where the noise exists is less than a point number threshold value, the specified region is judged as noise.

9. The method of claim 8, wherein, The point number threshold value is set by a user.

10. A computer program product comprising instructions for implementing the method of any one of claims 1-9.

Images