An indoor scene point cloud semantic segmentation method and system for a wearable device

Abstract

The present application relates to the technical field of point cloud data processing, in particular to an indoor scene point cloud semantic segmentation method and system for wearable devices, the present application can better capture the tiny details in the scene through curvature-based feature point selection and local geometric feature extraction, improve the capture ability of fine object edges, and significantly improve the segmentation accuracy; at the same time, the present application introduces multi-scale local geometric features and adaptive neighborhood definition, and adjusts the neighborhood radius adaptively through curvature change, thereby performing excellently in non-uniform point cloud scenes; the present application significantly reduces the computational complexity by simplifying the feature extraction process and using a lightweight multi-layer perceptron as a classifier. Combined with the optimization strategies such as farthest point sampling (FPS) to reduce the number of point cloud processing points, KD tree to accelerate neighborhood search, parallel processing of feature extraction and classification, etc., the present method can maintain real-time performance under limited computing resources and meet the needs of low-power devices.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of point cloud data processing, in particular to an indoor scene point cloud semantic segmentation method and system for wearable devices. BACKGROUND

[0002] The indoor scene point cloud semantic segmentation technology in wearable devices is an important direction of three-dimensional data processing, and its goal is to extract meaningful semantic information from sparse point cloud data. The core of current technology is how to effectively process and optimize these data to adapt to hardware limitations and real-time requirements. In recent years, methods that directly process point cloud data have been widely used. PointNet is the first neural network model that directly processes unordered point cloud data, which can extract features for each point and aggregate global features through symmetry functions (such as max-pooling). This method avoids converting point cloud data into voxels or grids, thereby reducing information loss and computational overhead. PointNet++ improves on this basis by using a hierarchical sampling mechanism to enhance the ability to capture local geometric features, making it more suitable for processing complex indoor environments. In addition, various fusion and optimization strategies have emerged in recent years. Models such as JSMNet introduce self-attention mechanisms and multi-scale feature fusion to further improve the accuracy of semantic segmentation, especially suitable for dense indoor point cloud environments. At the same time, some methods combine geometric features with neural networks for joint optimization, effectively improving the segmentation accuracy in complex scenes.

[0003] Although existing indoor scene point cloud semantic segmentation techniques have made significant progress in terms of accuracy and real-time performance, there are still some objective shortcomings:

[0004] 1. Information loss and lack of detail: Although models like PointNet and PointNet++ can directly process point clouds, they still have limitations in capturing local details and complex geometric structures, especially when dealing with dense and detailed indoor scenes. This can result in low segmentation accuracy for fine objects.

[0005] 2. Poor adaptability to non-uniform point cloud data: Many existing methods perform poorly when faced with non-uniform point cloud density, such as areas in the scene far from the sensor where point cloud data becomes sparse, making it difficult for the model to effectively capture features in these areas.

[0006] 3. High computational overhead and resource consumption: Existing deep learning models often require a large amount of computational resources and memory for high-precision segmentation, which is difficult to meet real-time requirements for resource-limited application scenarios such as wearable devices. Our goal is to design a more lightweight network structure to reduce the computational complexity of the model while maintaining high segmentation accuracy to meet the needs of low-power devices. SUMMARY

[0007] Therefore, the present application provides an indoor scene point cloud semantic segmentation method and system for wearable devices, which can reduce information loss, improve the adaptability of processing uneven data, and significantly reduce computational overhead while maintaining accuracy, and is particularly suitable for real-time indoor scene point cloud segmentation tasks in wearable devices.

[0008] To achieve the above-mentioned purpose, an indoor scene point cloud semantic segmentation method for wearable devices is provided, comprising the following steps:

[0009] S1: performing a downsampling preprocessing operation on point cloud data to obtain a set of downsampled point clouds;

[0010] S2: selecting feature points from the set of downsampled point clouds to obtain a set of key points;

[0011] S3: extracting features from the set of key points;

[0012] S4: performing feature fusion and classification on the features extracted in S3;

[0013] S5: performing real-time processing optimization on S1-S4.

[0014] Preferably, in S1, the farthest point sampling method is used to perform downsampling preprocessing on the input point cloud data,

[0015] Preferably, the farthest point sampling method (for downsampling preprocessing on the input point cloud data) is as follows:

[0016] The original point cloud data is where is the three-dimensional coordinate of the i-th point, a random number is set, and from the random number, the point farthest from the existing set of sampled points is selected in turn until the number of sampled points reaches the set threshold m, thereby obtaining a set of downsampled point clouds where m < n.

[0017] Preferably, in S2, curvature-based feature point selection is used to select feature points from the set of downsampled point clouds, thereby obtaining key points, which are selected through local geometric information of curvature.

[0018] Preferably, the curvature of the set of point clouds is calculated according to the following formula:

[0019]

[0020] where represents the local geometric change rate of point in the neighborhood of each point and The distance between points represents the geometric difference between that point and its surrounding points;

[0021] If the curvature at a certain point exceeds a preset threshold If the point is selected as a key point, it forms the key point set. Among them, the key point set The expression is:

[0022] .

[0023] Preferably, in step S3, feature extraction is performed on the key point set to extract local geometric features and global features of the key point set; the global features include geometric center, color information, and global feature formula.

[0024] Preferably, the process of extracting the local geometric features of the key point set is as follows:

[0025] For each key point Based on its neighborhood Extract local geometric features by using neighborhood points and the center point. Geometric relationships capture key points Local features, the key points The local geometric features are calculated in the following way:

[0026]

[0027] In the formula, This is the key point. and key points The Euclidean distance represents the straight-line distance between two points;

[0028] Among them, in the key point set In, each key point neighborhood Defined by radius r, that is, within a radius r... A set of points within a radius r centered at:

[0029] .

[0030] Preferably, the process of extracting the global features of the keypoint set is as follows:

[0031] a. Obtain the geometric center of the key point set;

[0032] The geometric center of the keypoint set is the average position of all points in the neighborhood; it represents the central position of the entire neighborhood. The formula is as follows:

[0033]

[0034] In the formula, is the geometric center of the key point set, is the number of points in the neighborhood;

[0035] b, obtain the color information of the key point set:

[0036] For each point , extract the color information of it;

[0037] c, obtain the global feature formula of the key point set:

[0038] Wherein, the global feature of the key point set is the combination of the geometric center and the color information:

[0039] The formula of the global feature of the key point set is:

[0040] .

[0041] Preferably, the feature fusion of the features extracted by the S3 is specifically:

[0042] The local geometric feature and the global feature obtained by the S3 are spliced to form the final fusion feature ; The formula is:

[0043] ;

[0044] Wherein, the feature classification of the features extracted by the S3 is specifically:

[0045] The lightweight multilayer perception (MLP) is used as a classifier to classify the features extracted by the S3.

[0046] According to another aspect of the present application, there is provided an indoor scene point cloud semantic segmentation system for wearable devices, which adopts the above-mentioned indoor scene point cloud semantic segmentation method for wearable devices, and the system comprises:

[0047] A preprocessing module is configured to perform a downsampling preprocessing operation on the point cloud data to obtain a set of downsampled point clouds.

[0048] A key point set acquisition module is configured to select feature points from the set of downsampled point clouds to obtain a key point set.

[0049] A feature extraction module is configured to extract features from the key point set.

[0050] A feature fusion and classification module is configured to perform feature fusion and classification on the features extracted by the S3.

[0051] A real-time processing optimization module is configured to perform real-time processing optimization on the point cloud semantic segmentation process.

[0052] The present application has the following advantages and benefits:

[0053] The present application screens out key points with significant geometric features through curvature calculation, and dynamically adjusts the neighborhood size based on the curvature change of each point, so that the model can effectively extract features in dense and sparse areas. This adaptive neighborhood division method not only improves the sensitivity of the model to areas with rich details, but also maintains the integrity of the features when processing sparse point clouds. Through curvature-based feature point selection and local geometric feature extraction, the model can better capture the fine details in the scene. The extraction process of local geometric features no longer relies on simple global operations, but through the calculation of the geometric change rate within the point cloud neighborhood, the ability to capture the edges of fine objects is improved, thereby significantly improving the segmentation accuracy.

[0054] The present application introduces a lightweight feature extraction strategy that combines local geometric features with global geometric center and color information, and innovatively fuses local and global information together. This fusion not only maintains efficiency, but also balances between global structure and local details, improving the accuracy and robustness of segmentation.

[0055] The present application introduces multi-scale local geometric features and adaptive neighborhood definition, and adaptively adjusts the neighborhood radius through curvature change to ensure that effective features can be captured in both sparse and dense areas, thereby performing well in non-uniform point cloud scenes. The adaptive mechanism not only improves the sensitivity of the model to sparse areas of point cloud data, but also effectively enhances the overall segmentation performance.

[0056] The present application significantly reduces computational complexity by simplifying the feature extraction process and using a lightweight multi-layer perceptron (MLP) as a classifier. Combined with farthest point sampling (FPS) to reduce the number of points processed by the point cloud, KD-tree accelerated neighborhood search, parallel processing of feature extraction and classification, and other optimization strategies, the present method can maintain real-time performance under limited computing resources, meeting the needs of low-power devices.

[0057] The present application is designed for real-time processing requirements of wearable devices, and specially designed efficient parallel processing and KD-tree accelerated neighborhood search. Compared with traditional linear neighborhood search, the introduction of KD-tree reduces the search time complexity to , combined with parallel processing, the overall processing efficiency is significantly improved, and can perform well in real-time segmentation tasks. This efficient optimization makes the method particularly suitable for wearable device application scenarios that require fast response. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the description of the present application or the prior art will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those of ordinary skill in the art without creative effort based on these drawings.

[0059] Figure 1 A flowchart of an indoor scene point cloud semantic segmentation method for a wearable device is provided for an embodiment of the present application.

[0060] Figure 2 A process schematic diagram from data processing to output of semantic segmentation results of an indoor scene point cloud semantic segmentation for a wearable device is provided for an embodiment of the present application.

[0061] Figure 3 A schematic diagram of an indoor scene point cloud semantic segmentation system for a wearable device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the description of the present application or the prior art will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those of ordinary skill in the art without creative effort based on these drawings.

[0063] ATTACHMENT Figure 1 A flowchart of an indoor scene point cloud semantic segmentation method for a wearable device is provided for an embodiment of the present application. Figure 2 A process schematic diagram from data processing to output of semantic segmentation results of an indoor scene point cloud semantic segmentation for a wearable device is provided for an embodiment of the present application. Figure 1 Figure 2 As shown in the accompanying drawings,

[0064] S1: performing a downsampling preprocessing operation on point cloud data to obtain a set of downsampled point clouds;

[0065] ​In this step, the input point cloud data is preprocessed by farthest point sampling (FPS). Farthest point sampling (FPS) is a commonly used sampling algorithm, mainly used for processing point cloud data. Its core idea is to make the distance between all sampling points as far as possible, so as to realize uniform discrete sampling of data. The principle of this algorithm is as follows: initial point selection: first, select a starting point from the point cloud, usually select the point farthest from the center of the point cloud as the first sampling point. Iterative sampling: for the remaining points, calculate the distance from each point to the current sampling point, select the point farthest from the current sampling point as the next sampling point, and update the sampling point set. Distance measurement: in point cloud data, the Euclidean distance is generally used to measure the distance between points. This algorithm can reduce the number of points while maintaining the geometric distribution characteristics of the point cloud.

[0066] Specifically, the farthest point sampling (FPS) preprocessing of the input point cloud data is as follows:

[0067] The original point cloud data is where is the three-dimensional coordinate of the ith point, a random number is set, and from the random number, the points farthest from the existing sampling point set are selected in turn until the number of sampling points reaches the set threshold m, thereby obtaining the down-sampled point cloud point set where m < n; by using farthest point sampling, it is ensured that the down-sampled point set has a uniform distribution and can effectively represent the overall structure of the point cloud.

[0068] S2: Selecting feature points from the down-sampled point cloud point set to obtain a key point set;

[0069] In order to reduce unnecessary calculation, this embodiment does not process all point cloud point sets, but selects key points with obvious geometric features for feature extraction. Therefore, in this step, curvature-based feature point selection is used to select feature points from the down-sampled point cloud point set, thereby obtaining key points. Key points are selected by local geometric information of curvature;

[0070] wherein the curvature of the point cloud point set is calculated by the following formula:

[0071]

[0072] wherein, represents the local geometric change rate of point Each point in the neighborhood and The distance between the points represents the geometric difference between the point and the surrounding points;

[0073] If the curvature of a point exceeds a preset threshold , the point is selected as a key point to form a key point set ; wherein the expression of the key point set is:

[0074] .

[0075] S3: Feature extraction is performed on the key point set;

[0076] In this step, feature extraction is performed on the key point set to extract local geometric features and global features of the key point set;

[0077] The global features include geometric center, color information, and global feature formula;

[0078] wherein the process of extracting the local geometric features of the key point set is:

[0079] For each key point , based on its neighborhood , the local geometric features are extracted, and the geometric relationship between the points in the neighborhood and the center point is used to capture the local features of the key point , and the local geometric features of the key point are calculated by the following method:

[0080]

[0081] wherein, is the Euclidean distance between the key point and the key point , representing the straight-line distance between the two points, and by calculating the square of the distance, the influence of the far points can be amplified, making the changes in the local geometric structure more obvious;

[0082] wherein, in the key point set , the neighborhood of each key point is defined by a radius r, i.e. the point set within the range of the center with a radius of r:

[0083]

[0084] In this step, the local geometric features reflect the distribution of points in the neighborhood. If all the neighborhood points are close to If the points in the neighborhood are more scattered, the eigenvalue will increase. This information is very useful for distinguishing different geometric structures and helps the model to understand the local geometric changes.

[0085] When facing non-uniform point cloud data, traditional methods often have difficulty in processing, for example, when the point cloud density is low (such as the far distance area of the sensor), the model has difficulty in effectively segmenting these areas. This step introduces multi-scale local geometric features and adaptive neighborhood definition, which adaptively adjusts the neighborhood radius according to the curvature change, ensuring that effective features can be captured in both sparse and dense areas, thereby performing well in non-uniform point cloud scenes. The adaptive mechanism not only improves the sensitivity of the model to sparse areas of point cloud data, but also effectively enhances the overall segmentation performance.

[0086] wherein the process of extracting the global feature of the key point set is:

[0087] a. Obtain the geometric center of the key point set;

[0088] The geometric center of the key point set is the average position of all points in the neighborhood, which represents the center position of the entire neighborhood. The geometric center The formula is as follows:

[0089]

[0090] In the formula, is the geometric center of the key point set, reflecting the spatial position of the entire neighborhood, is the number of points in the neighborhood, ensuring that the geometric center is the weighted average of all neighborhood points;

[0091] b. Obtain the color information of the key point set:

[0092] For each point , extract its color information Color can be an important semantic clue in indoor scenes, such as different furniture or walls with different colors, helping to identify different object categories;

[0093] c. Obtain the global feature formula of the key point set:

[0094] wherein the global feature of the key point set is a combination of the geometric center and color information:

[0095] The formula for the global feature of the key point set is as follows:

[0096]

[0097] The global feature is simple and efficient, suitable for use in resource-limited environments, while providing additional semantic information through color and geometric position.

[0098] S4: feature fusion and classification are performed on the S3 extracted features;

[0099] The feature fusion of the S3 extracted features is specifically:

[0100] The local geometric features and global features obtained in S3 are spliced to form the final fusion features ; The formula is:

[0101] ;

[0102] The feature classification of the S3 extracted features is specifically:

[0103] A lightweight multilayer perceptron (MLP) is used as a classifier to classify the features extracted in S3;

[0104] ‌Multilayer Perceptron (MLP) is a feedforward artificial neural network model, which is connected between network layers composed of multiple neurons in a fully connected manner. MLP is composed of multiple layers of neurons, including input layer, hidden layer and output layer. The input layer receives input data and passes the data to the hidden layer, the hidden layer converts the input value to the output value through the activation function and passes it to the output layer, and the output layer gives the final prediction result. MLP optimizes the weights and biases of neurons through training, so that the network can learn the mapping relationship between input and output. During training, the error between the predicted value and the actual value is calculated using the backpropagation algorithm, and the error is propagated back to each neuron in the network to update the weights and biases, so that the network can better approximate the target function.

[0105] In this step, the multilayer perceptron contains two hidden layers, each using a ReLU activation function, and the input of the multilayer perceptron is the fusion feature , and the output is the semantic category prediction of the point ; ;

[0106] The formula is:

[0107]

[0108] In this step, the final semantic segmentation result of the point cloud is output through the lightweight classifier.

[0109] As a preferred embodiment, the following steps are also included:

[0110] S5: real-time processing optimization is performed on S1-S4;

[0111] The S5 specifically includes KD number neighborhood search and neighborhood feature extraction and classification parallel processing.

[0112] In order to speed up the process of neighborhood search, KD tree structure is adopted. KD tree (K-Dimensional Tree) is a data structure for efficient processing of multi-dimensional data, which is a special binary tree structure used to organize points in k-dimensional space. KD tree divides high-dimensional space into a series of hyper-rectangular regions by recursion, and each node represents a hyperplane that divides the space into two subspaces. This structure can complete the nearest neighbor search in logarithmic time complexity , which can significantly improve the processing efficiency.

[0113] The neighborhood feature extraction and classification parallel processing specifically includes that for different key points , neighborhood feature extraction and classification can be processed in parallel to further improve the computing efficiency, so that the entire process can run in a computing resource limited environment such as a wearable device.

[0114] Embodiment two includes an indoor scene point cloud semantic segmentation system for wearable devices, which Figure 3 shows a structural diagram of an indoor scene point cloud semantic segmentation system for wearable devices, as shown in the accompanying Figure 3 , the system adopts the indoor scene point cloud semantic segmentation method for wearable devices in embodiment one, and the system includes:

[0115] A preprocessing module is configured to perform a downsampling preprocessing operation on the point cloud data to obtain a set of downsampled point clouds.

[0116] A key point set acquisition module is configured to select feature points from the set of downsampled point clouds to obtain a key point set.

[0117] A feature extraction module is configured to extract features from the key point set.

[0118] A feature fusion and classification module is configured to perform feature fusion and classification on the features extracted by S3.

[0119] A real-time processing optimization module is configured to perform real-time processing optimization on the point cloud semantic segmentation process.

[0120] Embodiment three includes a computer readable storage medium, and a data processing program is stored on the computer readable storage medium. The data processing program is executed by a processor to implement the indoor scene point cloud semantic segmentation method for wearable devices in embodiment one.

[0121] Those skilled in the art will appreciate that embodiments of the present disclosure can be readily used as a method, apparatus (device) or computer program product. Accordingly, embodiments of the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. The embodiments of the present disclosure can be implemented as program code (e.g., in the form of software or firmware) for use with a programmable computer-based system, computer-based system or other programmable processing apparatus, or devices. Program code, such as program code implemented as program modules, can be stored on any type of non-transitory computer-readable medium or memory, such as RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Further, those skilled in the art will appreciate that the communication media can typically embody computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and can include any information delivery media.

[0122] The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks.

[0123] The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer, other programmable processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks.

[0124] The above-described embodiments and / or implementations are merely intended for describing the preferred embodiments and / or implementations of the present technology and are not intended to limit the present technology in any form, and any person skilled in the art can make some changes or modifications to other equivalent embodiments without departing from the scope of the technology disclosed in the present disclosure, and such changes or modifications should be considered as substantially the same technology or embodiments. Note that the above is only the preferred embodiment of the present application and the technical principles used. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and various obvious changes, reconfigurations, and substitutions can be made without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for semantic segmentation of indoor scene point cloud for wearable devices, characterized in that, The method comprises the following steps: S1: performing a downsampling preprocessing operation on point cloud data to obtain a set of downsampled point cloud points; S2: selecting feature points from the set of downsampled point cloud points to obtain a set of key points; in S2, the set of downsampled point cloud points is selected for feature points based on curvature, so as to obtain key points, and the key points are screened by local geometric information of curvature; a calculation formula of the curvature of the set of point cloud points is: wherein the local geometry variation rate of a point each point in the neighborhood the distance between indicates the geometric difference of the point with the surrounding points; If the curvature of a point exceeds a preset threshold , the point is selected as a key point to form a key point set ; wherein the expression of the key point set is: ; S3: extracting features from the set of key points; in S3, local geometric features and global features of the set of key points are extracted by extracting features from the set of key points; the global features include a geometric center, color information, and a global feature formula; The process of extracting the local geometric features of the set of key points is: For each key point , a local geometric feature is extracted based on its neighborhood The local feature of the key point is captured by using the geometric relationship between the points in the neighborhood and the center point The local geometric feature of the key point is calculated by wherein is a key point and a key point Euclidean distance between two points; wherein, in the key point set each key point has a neighborhood defined by a radius r, i.e. a set of points within a range centered at with radius r. S4: fusing and classifying the features extracted in S3; S5: performing real-time processing optimization on S1-S4.

2. The method of claim 1, wherein, In S1, the furthest point sampling method is used to perform downsampling preprocessing on the input point cloud data.

3. The method of claim 2, wherein, The process of using the furthest point sampling method to perform downsampling preprocessing on the input point cloud data is: The original point cloud data is wherein is the three-dimensional coordinate of the i-th point, a random number is set, and from the random number, points farthest from the existing sampling point set are selected in turn until the number of sampling points reaches the set threshold m, thereby obtaining the down-sampled point cloud point set wherein .

4. The method of claim 1, wherein, The process of extracting the global features of the set of key points is: a. obtaining a geometric center of the set of key points; The geometric center of the key point set is the average position of all points in the neighborhood, which represents the central position of the entire neighborhood. The geometric center The formula is as follows: wherein is the geometric center of the set of key points, is the number of points in the neighborhood; b. obtaining color information of the set of key points: For each point , extract its color information ; c. obtaining a global feature formula of the set of key points: The global features of the set of key points are a combination of the geometric center and the color information: global feature of the key point set The formula is: 。 5. The indoor scene point cloud semantic segmentation method for wearable devices according to claim 4, wherein The process of fusing the features extracted in S3 is: The local geometric features obtained by the S3 are spliced with the global features to form final fusion features ; the formula is: ； The process of classifying the features extracted in S3 is: A lightweight multi-layer perception (MLP) is used as a classifier to classify the features extracted in S3. 6.A system for indoor scene point cloud semantic segmentation for wearable devices, characterized in that, The system adopts the indoor scene point cloud semantic segmentation method for wearable devices according to any one of claims 1-5, and the system comprises: A preprocessing module for performing a downsampling preprocessing operation on point cloud data to obtain a set of downsampled point cloud points; A key point set acquisition module for selecting feature points from the set of downsampled point cloud points to obtain a set of key points; A feature extraction module for extracting features from the set of key points; A feature fusion and classification module for fusing and classifying the features extracted in S3; A real-time processing optimization module for performing real-time processing optimization on the point cloud semantic segmentation process.

Images