An RGB-D vision synchronization position and map construction method based on line feature reliability evaluation and depth consistency optimization

Abstract

The application discloses a kind of based on line feature reliability evaluation and depth consistency optimization RGB-D vision synchronization position and map construction method, belong to computer vision and robot autonomous positioning technical field.This method includes: acquisition RGB-D vision sensor data and extract point feature and two-dimensional line segment;Based on inverse depth lateral support method, RGB-D line depth optimization is carried out;Adaptive screening based on local structure context is carried out to the line segment after optimization, and line segment verification threshold is dynamically adjusted according to local structure density;Single line segment observation quality evaluation model is established and quality factor is generated;Local bundle adjustment optimization is carried out based on quality factor;The position and three-dimensional map after optimization are output.The application carries out line feature extraction, depth reliability analysis, line segment quality evaluation and weighted optimization to the acquisition data, and effectively improves the positioning accuracy and robustness of vision synchronization positioning and map construction system in weak texture, structure repetition and complex indoor environment.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and robot autonomous localization technology, specifically to an RGB-D visual synchronous localization and map building method based on line feature reliability assessment and depth consistency optimization. Background Technology

[0002] Simultaneous Localization and Mapping (SLAM) is a key technology in robot autonomous navigation, augmented reality, and intelligent perception systems. It acquires environmental information through visual sensors and combines feature extraction, data association, and backend optimization methods to achieve device pose estimation and environmental map construction. In recent years, visual feature-based SLAM methods have become an important research direction in the field of mobile robot localization and 3D environment reconstruction due to their low cost, rich information, and wide applicability.

[0003] Existing visual SLAM methods primarily rely on point features in images to represent the environment. For example, the ORB-SLAM method proposed by Mur-Artal et al. utilizes ORB features for real-time feature matching and combines local map construction and loop closure detection to achieve high-precision monocular visual localization, becoming one of the classic feature-point SLAM frameworks. Subsequently, Campos et al. further proposed ORB-SLAM3, which integrates visual, visual-inertial, and multi-map mechanisms, improving the adaptability of visual SLAM systems in complex environments. However, point features typically depend on local texture information. When applied to large-area, weakly textured, or repetitive structural environments such as walls, corridors, and tabletops, problems such as insufficient feature quantity, unstable matching, and localization drift can easily occur.

[0004] To improve the ability of visual SLAM systems to represent structured environments, researchers have gradually introduced line features as auxiliary constraints, forming point-line fusion visual SLAM methods. Compared to point features, line features can describe long-range structural information in the environment, such as wall boundaries, door frames, and object outlines, and can still provide effective geometric constraints when texture is insufficient. Pumarola et al. proposed the PL-SLAM method, which uses point features and line features together for visual localization and map building, improving the system's stability in structured scenes. In addition, Zhang et al. proposed Line Flow SLAM, which improves long-term tracking capabilities by associating line features between consecutive frames, further demonstrating the application value of line features in visual SLAM.

[0005] With the development of RGB-D (Red-Green-Blue-Depth) sensors, visual SLAM methods incorporating depth information can directly acquire the spatial structure of the environment, providing richer information for 3D map construction. RGB-D SLAM systems typically use depth data to reconstruct 3D point or line structures and improve positioning accuracy through geometric constraints. However, the depth information acquired by RGB-D sensors is easily affected by object boundaries, occlusion, and sensor noise. Foreground and background depth mixing may occur in areas where different objects meet, leading to deviations in the reconstruction of 2D line features in 3D space. Related research shows that RGB-D sensors have significant measurement errors in areas of depth discontinuity, requiring uncertainty modeling to improve the reliability of depth constraints.

[0006] To address the line feature processing problem in RGB-D environments, existing research has explored aspects such as line detection, depth estimation, and uncertainty optimization. For example, Grompone von Gioi et al. proposed the LSD line segment detection algorithm, achieving high-precision two-dimensional line segment extraction; ELSED (Efficient Line Segment Detector) further optimized the line segment extraction efficiency, improving real-time detection capabilities through anchor point generation and line segment growth strategies. However, existing line detection methods typically rely on fixed thresholds or local geometric conditions for filtering, which can easily generate a large number of unstable short lines and may miss effective line segments in weakly structured regions.

[0007] On the other hand, existing point-line SLAM systems typically treat all line observations as constraints with similar reliability and primarily set optimization weights based on image scale or uniform covariance, making it difficult to distinguish the stability differences between different line observations within the same scale level. Previous research has indicated that the quality of feature observations in visual SLAM directly affects the backend optimization results; therefore, it is necessary to establish more reasonable uncertainty evaluation models for different visual constraints. Some studies have further explored the uncertainty propagation of point-line observations and the optimization of line constraints. For example, SPLODE uses a probabilistic model to describe the uncertainty of RGB-D point-line measurements, improving the robustness of visual odometry; PLVS further considers the uncertainty related to line feature scale and depth to optimize line reprojection constraints.

[0008] In summary, while existing RGB-D point-line SLAM methods can enhance the geometric representation of the environment by utilizing line structures, they still have shortcomings in line depth reliability, adaptive selection of line segments based on context, and modeling of differences in observation quality among different lines. Therefore, there is an urgent need to propose an RGB-D visual SLAM method that can evaluate the entire process from line feature acquisition and 3D reconstruction to backend optimization and enhance the reliability of line observations, in order to improve the accuracy and stability of localization and map construction in complex indoor environments. Summary of the Invention

[0009] In view of the shortcomings of the prior art, the present invention aims to solve the following technical problems: 1. This paper proposes a 3D reconstruction method for RGB-D line features based on depth consistency analysis, which solves the problem of inaccurate spatial positioning of line features caused by depth mixing in boundary regions in existing methods. By analyzing the depth information at different locations around the 2D line segment, an inverse depth consistency evaluation mechanism is introduced to select depth support regions that better conform to the actual spatial structure, thereby improving the accuracy of 3D line segment initialization and verification and reducing the impact of depth error on visual positioning results.

[0010] 2. This paper proposes a line feature extraction method based on local structural context adaptation, which addresses the problems of missed detection of valid line segments and incorrect retention of low-quality line segments caused by the use of fixed thresholds in existing line detection algorithms. By analyzing the structural support information in the neighborhood of candidate line segments, the line segment direction verification conditions are dynamically adjusted, enabling the line feature extraction process to adapt to different scene structural complexities. This method retains valid structural lines in weakly textured environments while suppressing fragmented and unstable line segments in complex textured regions.

[0011] 3. This paper proposes a line constraint optimization method based on single-segment observation quality evaluation to address the problem that existing visual SLAM systems cannot distinguish the reliability differences of different line observations. By comprehensively analyzing geometric features such as line segment direction consistency and fitting stability, a line observation quality evaluation model is established. Quality information is then incorporated into the local bundle adjustment optimization process, enabling line features with different reliability levels to obtain constraint weights that match their actual stability. This improves the positioning accuracy and grounding performance of the visual SLAM system in complex environments. Figure 1 To the point of being responsive.

[0012] To address the problems of inaccurate online feature 3D reconstruction, lack of scene adaptability in line segment extraction, and inability to effectively distinguish the reliability of different line observations in existing RGB-D point-line visual SLAM systems, this invention proposes an RGB-D visual simultaneous localization and map construction method based on line feature reliability assessment and depth consistency optimization. The specific technical solution is as follows: A method for constructing RGB-D visual synchronization bitmaps based on line feature reliability assessment and depth consistency optimization includes the following steps: Step 1: Acquire RGB-D vision sensor data and extract point feature sets and two-dimensional line segment sets from it; Step 2: For each two-dimensional line segment in the set of two-dimensional line segments, perform depth optimization based on the inverse depth lateral support method to obtain the depth support region with the best depth consistency, and back-project the two-dimensional line segment to the three-dimensional space based on the depth support region to generate three-dimensional line features; Step 3: Perform adaptive line segment filtering based on local structural context on the depth-optimized 2D line segments, count the local structural density in the neighborhood of the candidate line segments, and dynamically adjust the line segment direction verification threshold according to the local structural density to filter the line segments. Step 4: Establish a single-segment observation quality evaluation model for each selected line segment and generate the line observation quality factor; Step 5: Introduce the line observation quality factor into the local bundle adjustment optimization process, dynamically adjust the optimization weight of the line constraint according to the line observation quality factor, and perform pose and map joint optimization by combining point feature constraints and line feature constraints; Step 6: Output the optimized camera pose, 3D point map, and 3D line map.

[0013] Compared with the prior art, the present invention discloses the following technical effects: 1. This invention introduces a lateral support strategy based on inverse depth consistency to analyze and filter the depth information of the corresponding region of a 2D line segment, avoiding the direct use of depth data with mixing errors for 3D reconstruction. This method can effectively reduce spatial positioning errors caused by the mixing of foreground and background depths in the object boundary region, improve the accuracy of 3D line feature construction, and provide more reliable geometric constraints for subsequent visual optimization.

[0014] 2. This invention adaptively adjusts the line segment verification process by incorporating local structural context information, enabling the line feature extraction process to dynamically change according to the structural characteristics of different scenes. Compared to traditional fixed threshold detection methods, this method can reduce invalid fragmented lines in complex textured environments while improving the retention of effective structural lines in weakly textured regions, thereby further enhancing the stability of the visual SLAM system in the online feature input stage.

[0015] 3. This invention establishes a single-segment observation quality evaluation model, transforming the line segment's directional stability and geometric fitting ability into quality information usable for optimization, and further adjusting the line constraint weights during local bundle adjustment. This method avoids the excessive impact of low-reliability line features on camera pose estimation and map optimization, allowing line observations of different quality levels to play a role commensurate with their reliability, effectively improving the system's positioning accuracy and map optimization. Figure 1 To the point of being responsive.

[0016] 4. This invention does not require changes to the overall framework of traditional RGB-D visual SLAM systems; it only adds reliability enhancement modules during the online feature processing and local optimization stages, thus ensuring compatibility with existing point-line fusion SLAM architectures. This method is suitable for structured indoor environments such as walls, furniture edges, and door frames, and can effectively improve positioning stability in applications such as robot navigation, augmented reality, and 3D reconstruction.

[0017] 5. This invention adopts a phased reliability enhancement strategy, which optimizes only the line observation-related process and does not require the introduction of complex scenario prior models or large-scale computing modules. Therefore, it can effectively improve the quality of line constraints and the overall system performance while maintaining the real-time performance of traditional RGB-D SLAM. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the RGB-D visual synchronization bit and map construction method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the overall process framework of the RGB-D visual synchronization bit and map construction method provided in this embodiment of the invention. Figure 3 This is a flowchart of the adaptive line segment filtering method provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of quality-aware local bundle adjustment optimization provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides a method for RGB-D visual synchronization bit and map construction based on line feature reliability assessment and depth consistency optimization, such as... Figure 1-2 As shown, it includes the following steps: Step 1: Acquire RGB-D vision sensor data and extract point and line observations.

[0022] To achieve simultaneous localization and map building in complex environments, this invention employs an RGB-D visual sensor to acquire environmental image data, including color images. and depth images .

[0023] First, feature extraction is performed on the input image to obtain a set of point features: ; And a set of two-dimensional line segments: ; Point features are used to provide local texture constraints, while line features are used to provide environmental structure constraints. For the detected two-dimensional line segments: ; in: and These represent the coordinates of the starting and ending points of the line segment, respectively.

[0024] This invention enhances the reliability of line observations based on the traditional point-line SLAM framework before they are used in 3D map construction and pose optimization.

[0025] Step 2: RGB-D line depth optimization based on inverse depth lateral support.

[0026] Because RGB-D sensors suffer from discontinuity in depth measurement at object boundaries, this invention first performs a reliability analysis on the depth information corresponding to two-dimensional line segments.

[0027] For example, when a line segment in an image is located at the edge of a table, the boundary of a wall, or the outline of an object, the two sides of the line segment may correspond to different spatial depths. If the depth at the center of the line segment is used directly for 3D reconstruction, the reconstructed 3D line segment will deviate from the true spatial structure.

[0028] Therefore, this invention proposes a depth optimization method based on inverse depth lateral support. Inverse depth lateral support is a line depth optimization method proposed to address the depth discontinuity problem in the boundary regions of RGB-D images. By comparing the inverse depth consistency corresponding to different candidate positions, it selects a depth support region that better conforms to the real spatial structure for the 3D reconstruction of 2D line segments, thereby improving the accuracy of 3D line features.

[0029] For two-dimensional line segments First, calculate its unit direction vector: ; And calculate the corresponding normal direction: ; in, and Unit direction vectors Components in the x-axis and y-axis directions of the image coordinate system; Subsequently, multiple depth sampling locations are generated in the line segment direction and the normal direction: ; in, The sampling ratio in the direction of the line segment. k=1,2,…,K, where K is the number of sampling points. This represents the offset distance along the normal direction.

[0030] For each sampling point, calculate its depth. and depth Convert to inverse depth: ; Subsequently, linear fitting was performed on the support regions at different candidate depths: ; in, and These represent the slope coefficient and intercept coefficient in the inverse depth linear fitting model, respectively, used to describe the inverse depth variation relationship within the candidate depth support region; Calculate the inverse depth fitting error: ; in, Indicates the first The line segment at the offset The number of corresponding effective depth sampling points; Indicates the first The line segment in the th Each sampling position, offset is The corresponding inverse depth observation value at that time; Choose the depth support position with the smallest error: ; Recover the depth of the line segment endpoints based on the optimal inverse depth model: ; ; Using camera intrinsic matrix K 1. Back-project the two-dimensional endpoints to obtain the three-dimensional line endpoints: ; Specifically, for the starting point of a line segment, the three-dimensional endpoints obtained by projection are: ; For the endpoint of the line segment, the 3D endpoint obtained by projection is: .

[0031] By employing the inverse depth lateral support method to optimize the RGB-D line depth, two-dimensional line segments can be matched with a more consistent spatial depth structure.

[0032] Step 3: Adaptive line segment filtering based on local structural context.

[0033] To avoid the problems of missed valid lines and retained invalid broken lines caused by traditional fixed threshold line detection methods, this invention proposes an adaptive line segment verification method based on local structure. (Refer to...) Figure 3 .

[0034] For candidate line segments The number of structural anchor points within its neighborhood is counted.

[0035] Define local structural density: ; in, This represents the number of valid structural points within the regions on both sides of the line segment. The length of the line segment. , For neighborhood range parameters, m Indicates the endpoint extension range.

[0036] Based on the structural density distribution of all candidate line segments in the current image, the region is divided into: Low-structure region: ; Medium structural region: ; High-structure areas: ; Further adjust the direction verification threshold according to the region type: ; in: The initial line segment direction verification threshold is set. and These are the 30th and 75th percentiles of the structural density distribution, used to delineate regions with different levels of structural complexity.

[0037] This method can reduce verification restrictions and improve the retention rate of effective structural lines in weakly textured areas, while increasing verification requirements and reducing fragmented line interference in complexly textured areas.

[0038] Step 4: Establish a single-segment observation quality evaluation model.

[0039] Since different line segments may have different degrees of stability even at the same scale, this invention further establishes a single line segment observation quality evaluation model.

[0040] First, calculate the line segment direction consistency error: ; in, Indicates the first i The line segment in the th j The first frame of the image mLocal orientation estimation at each sampling location, where M is the number of sampling locations. This is the overall fitting direction for the line segment; Simultaneously calculate the line segment fitting error: ; in, For the first m The two-dimensional coordinates of each sampling location on the line segment Sampling position to fitted line segment The vertical distance; The direction consistency error and the line segment fitting error are normalized separately: ; ; in, and These are the mean and standard deviation of the orientation consistency error, respectively. and These are the mean and standard deviation of the fitting error, respectively. Establish a line segment quality evaluation model: ; in, For the intercept term, This is the weighting coefficient for the direction consistency error. These are the weighting coefficients for the fitting error; Further, the line observation quality factor was obtained: ; In the formula, the line observation quality factor Used to indicate the relative reliability of the line observation during the local BA optimization process; This represents the original line observation quality value calculated based on the consistency of line segment direction and the stability of straight line fitting. This represents a truncation function used to limit the input value to a preset range. When the input value is less than the lower limit, it outputs the lower limit value; when the input value is greater than the upper limit, it outputs the upper limit value, in order to ensure that the quality factor is within a stable range.

[0041] The observation quality factor is used to represent the reliability of constraints on different line segments.

[0042] Step 5: Quality-aware local bundle adjustment optimization.

[0043] In traditional SLAM backend optimization, line observations typically use uniform weights: ; This invention introduces a line observation quality factor: ; Among them, weight This represents the information matrix of the i-th map line in the j-th observation, used to describe the constraint strength of the line observation in the local BA optimization; This represents the basic weighting coefficient for line observations; Indicates the i-th The reciprocal of the variance of the line reprojection error corresponding to the j-th frame image is used to describe the observation noise level; Indicates the first The map line at the 1st Two-dimensional observations in keyframes; Represents the identity matrix.

[0044] The corresponding line constraint optimization objective is: ; in, This is the line reprojection error. (·) is the robust loss function, used to reduce the impact of outlier observations on the optimization process.

[0045] Final joint optimization: ; in: E P For point feature constraint error, These are the weighting coefficients for the line constraints.

[0046] By adjusting the quality-perceived weights, the stable line features can provide stronger constraints, further reducing the impact of low-reliability line observations on pose estimation. Figure 4 .

[0047] Step 6: Output the optimized pose and 3D map.

[0048] After the above processing, a reliable set of two-dimensional line observations can be obtained: The point features are combined to participate in SLAM optimization, and the final output is: camera pose, 3D point map, and 3D line map.

[0049] This invention improves the reliability of line constraints through three stages: line depth recovery, line feature selection, and backend optimization, effectively enhancing the positioning accuracy and map stability of the RGB-D visual SLAM system in complex indoor environments.

[0050] This invention also provides an RGB-D visual synchronous localization and mapping system based on line feature reliability assessment and depth consistency optimization, applied to the above-mentioned RGB-D visual synchronous localization and mapping method, including: The data acquisition and feature extraction module is used to acquire RGB-D vision sensor data and extract point feature sets and two-dimensional line segment sets from it. The line feature depth optimization module is used to perform depth optimization on each two-dimensional line segment in the set of two-dimensional line segments based on inverse depth lateral support, obtain the depth support region with optimal depth consistency, and back-project the two-dimensional line segment to three-dimensional space based on the depth support region to generate three-dimensional line features. The adaptive line segment filtering module is used to adaptively filter candidate line segments based on the local structural context. It dynamically adjusts the line segment verification threshold according to the structural density within the neighborhood of the candidate line segment in order to filter the line segments. The line observation quality evaluation module is used to establish a single-segment observation quality evaluation model for each selected line observation and generate line observation quality factors. The joint optimization module is used to introduce the line observation quality factor into the local bundle adjustment optimization process, dynamically adjust the optimization weight of the line constraint according to the line observation quality factor, and perform pose and map joint optimization by combining point feature constraints and line feature constraints. The output module is used to output optimized camera pose, 3D point map, and 3D line map.

[0051] The RGB-D visual synchronous localization and mapping method and system based on line feature reliability assessment and depth consistency optimization provided by this invention can be widely applied in fields such as autonomous navigation of mobile robots, intelligent inspection, augmented reality, 3D spatial reconstruction, and environmental perception of unmanned systems. By performing line feature extraction, depth reliability analysis, line segment quality evaluation, and weighted optimization on image data acquired by RGB-D vision sensors, the localization accuracy and robustness of the visual synchronous localization and mapping system can be effectively improved in environments with weak textures, repetitive structures, and complex indoor environments. It also has high compatibility and practicality, making it suitable for large-scale industrial applications and promotion.

[0052] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for constructing RGB-D visual synchronization bitmaps based on line feature reliability assessment and depth consistency optimization, characterized in that, Includes the following steps: Step 1: Acquire RGB-D vision sensor data and extract point feature sets and two-dimensional line segment sets from it; Step 2: For each two-dimensional line segment in the set of two-dimensional line segments, perform depth optimization based on the inverse depth lateral support method to obtain the depth support region with the best depth consistency, and back-project the two-dimensional line segment to the three-dimensional space based on the depth support region to generate three-dimensional line features; Step 3: Perform adaptive line segment filtering based on local structural context on the depth-optimized 2D line segments, count the local structural density in the neighborhood of the candidate line segments, and dynamically adjust the line segment direction verification threshold according to the local structural density to filter the line segments. Step 4: Establish a single-segment observation quality evaluation model for each selected line segment and generate the line observation quality factor; Step 5: Introduce the line observation quality factor into the local bundle adjustment optimization process, dynamically adjust the optimization weight of the line constraint according to the line observation quality factor, and perform pose and map joint optimization by combining point feature constraints and line feature constraints; Step 6: Output the optimized camera pose, 3D point map, and 3D line map.

2. The RGB-D visual synchronization bit and map construction method according to claim 1, characterized in that, In step 1, the RGB-D vision sensor data includes color images and depth images.

3. The RGB-D visual synchronization bit and map construction method according to claim 1, characterized in that, Step 2 specifically includes: For two-dimensional line segments Calculate its unit direction vector: ； in, Indicates the coordinates of the starting point of the line segment. Indicate the coordinates of the endpoint of the line segment; Then calculate the corresponding normal direction: ； in, and Unit direction vectors Components in the x-axis and y-axis directions of the image coordinate system; Generate multiple depth sampling positions along the line segment direction and the normal direction: ； in, The sampling ratio in the direction of the line segment. k =1,2,…,K, where K is the number of sampling points. This represents the offset distance along the normal direction; For each sampling point, calculate its depth. and depth Convert to inverse depth: ； Linear fitting was performed on support regions of different candidate depths: ； in, and These represent the slope coefficient and intercept coefficient in the inverse depth linear fitting model, respectively. Calculate the inverse depth fitting error: ； in, Indicates the first The line segment at the offset The number of corresponding effective depth sampling points; Indicates the first The line segment in the th Each sampling position, offset is The corresponding inverse depth observation value at that time; Choose the depth support position with the smallest error: ； Recover the depth of the line segment endpoints based on the optimal inverse depth model: ； ； Using camera intrinsic matrix K 1. Back-project the two-dimensional endpoints to obtain the three-dimensional line endpoints: For the starting point of the line segment, the three-dimensional endpoint obtained by projection is: ; For the endpoint of the line segment, the 3D endpoint obtained by projection is: .

4. The RGB-D visual synchronization bit and map construction method according to claim 3, characterized in that, In step 3, the formula for calculating the local structural density is: ； in, This represents the number of valid structural points within the regions on both sides of the line segment. The length of the line segment. , For neighborhood range parameters, m Indicates the endpoint extension range.

5. The RGB-D visual synchronization bit and map construction method according to claim 4, characterized in that, In step 3, the line segment direction verification threshold is dynamically adjusted according to the structural density, specifically as follows: ； in, The initial line segment direction verification threshold is set. and These are the 30th and 75th percentiles of the structure density distribution, respectively.

6. The RGB-D visual synchronization bit and map construction method according to claim 1, characterized in that, In step 4, a single-segment observation quality evaluation model is established for each selected line segment, specifically including: first, calculating the line segment direction consistency error: ； in, Indicates the first i The line segment in the th j The local orientation estimation at the m-th sampling position of a frame image, where M is the number of sampling positions. Indicates the overall fitting direction of the line segment; Then calculate the line segment fitting error: ； in, Let m be the two-dimensional coordinates of the m-th sampling position on the line segment. Sampling position to fitted line segment The vertical distance; The direction consistency error and the line segment fitting error are normalized separately: ； ； in, and These are the mean and standard deviation of the orientation consistency error, respectively. and These are the mean and standard deviation of the fitting error, respectively. Establish a line segment quality evaluation model: ； in, For the intercept term, This is the weighting coefficient for the direction consistency error. These are the weighting coefficients for the fitting error; Finally, the line observation quality factor is generated: ； in, This represents the original line observation quality value calculated based on the consistency of line segment direction and the stability of straight line fitting. This represents the truncation function.

7. The RGB-D visual synchronization bit and map construction method according to claim 6, characterized in that, In step 5, the optimization weights of the line constraints are dynamically adjusted based on the line observation quality factor. Specifically, the weights of the line constraints are expressed as follows: ； in, This represents the basic weighting coefficient for line observations; Indicates the first i The line segment in the th j The reciprocal of the variance of the line reprojection error corresponding to the frame image is used to describe the observation noise level; Indicates the first i The line segment in the th j Two-dimensional observations in frame images; Represents the identity matrix; The objective of the line constraint optimization is defined as: ； in, This is the line reprojection error. (·) represents the robust loss function.

8. The RGB-D visual synchronization bit and map construction method according to claim 7, characterized in that, In step 5, the pose and map are jointly optimized by combining point feature constraints and line feature constraints. Specifically, the joint objective optimization function is: ； in, E P For point feature constraint error, These are the weighting coefficients for the line constraints.

9. An RGB-D visual synchronous localization and mapping system based on line feature reliability assessment and depth consistency optimization, applied to the RGB-D visual synchronous localization and mapping method according to any one of claims 1-8, characterized in that, include: The data acquisition and feature extraction module is used to acquire RGB-D vision sensor data and extract point feature sets and two-dimensional line segment sets from it. The line feature depth optimization module is used to perform depth optimization on each two-dimensional line segment in the set of two-dimensional line segments based on inverse depth lateral support, obtain the depth support region with optimal depth consistency, and back-project the two-dimensional line segment to three-dimensional space based on the depth support region to generate three-dimensional line features. The adaptive line segment filtering module is used to adaptively filter candidate line segments based on the local structural context. It dynamically adjusts the line segment verification threshold according to the structural density within the neighborhood of the candidate line segment in order to filter the line segments. The line observation quality evaluation module is used to establish a single-segment observation quality evaluation model for each selected line observation and generate line observation quality factors. The joint optimization module is used to introduce the line observation quality factor into the local bundle adjustment optimization process, dynamically adjust the optimization weight of the line constraint according to the line observation quality factor, and perform pose and map joint optimization by combining point feature constraints and line feature constraints. The output module is used to output optimized camera pose, 3D point map, and 3D line map.

Images