CN-120125755-B - Building model reconstruction method and system based on four-foot robot and machine vision

CN120125755BCN 120125755 BCN120125755 BCN 120125755BCN-120125755-B

Abstract

The invention discloses a building model reconstruction method and system based on a quadruped robot and machine vision, based on the quadruped robot carrying the laser scanner, a global map of the area to be reconstructed is generated by utilizing a synchronous positioning and map construction technology. And determining a target scanning point by combining three-dimensional ray tracing and an A-path searching algorithm and performing global path planning. Based on a high-precision sensor carried by the quadruped robot, the accurate positioning and obstacle avoidance of the quadruped robot in a complex environment are realized through the self-adaptive Monte Carlo positioning and improved dynamic window algorithm. After moving to the target scanning point, the robot is stopped and started to perform high-precision point cloud data acquisition until all the sampling is completed. And after preprocessing the acquired point cloud data, generating a three-dimensional building model through a point cloud registration and poisson surface reconstruction technology. The invention can realize comprehensive, efficient and accurate three-dimensional reconstruction in complex building environments.

Inventors

WANG JIAN
ZHANG SHUOLIN
Yang tianqin
Yue Hongzhe
ZENG NINGSHUANG

Assignees

东南大学

Dates

Publication Date: 20260508
Application Date: 20250310

Claims (6)

1. A building model reconstruction method based on a quadruped robot and machine vision is characterized by comprising the following steps: step S1, three-dimensional point cloud data of a reconstruction area and IMU pose data of the quadruped robot are obtained, and environment sensing and mapping are carried out on the reconstruction area by adopting a synchronous positioning and map building algorithm to obtain a global map; Step S2, determining a target scanning point and a corresponding scanning path thereof based on a global map by combining a three-dimensional ray tracing algorithm and an A-path searching algorithm, and generating a global path plan; Step S3, planning the quadruped robot to start walking according to the generated global path, and realizing real-time global positioning of the quadruped robot based on a self-adaptive Monte Carlo positioning algorithm and combined with sensor data in the walking process of the quadruped robot; Step S4, stopping the four-legged robot after reaching a target scanning point, starting a laser scanner to acquire point cloud data and RGB data, and after the acquisition is completed, planning and repeating the step S3 according to the global path in the step S2 to move to the next target scanning point until the acquisition of all original point cloud data is completed; s5, preprocessing all the acquired original point cloud data to obtain preprocessed point cloud data; s6, adopting a multi-mode data fusion registration method to the preprocessed point cloud data to obtain a reconstructed three-dimensional building model; The step S2 specifically includes: Extracting geometric midlines of navigable spaces from a global map of a two-dimensional grid to generate a topological skeleton map, and converting discretized geometric midline nodes V j into a three-dimensional candidate scanning point set P j by utilizing an improved medial axis transformation algorithm; Initializing a target point set T as an empty set, selecting a point P max covering the maximum value of the score from the candidate scanning point set P j , and adding the point P max into the T; Setting a minimum distance d min between two adjacent target scanning points to avoid redundancy of scanning data, synchronously eliminating neighborhood nodes with P max as a center and a radius d min , continuously updating a target point set T, optimizing global coverage rate C, calculating increment delta C, and stopping iteration until delta C is less than 1%; Obtaining a target scanning point set T; After determining a target scanning point set T, performing path planning by using an A algorithm, calculating the shortest path and avoiding collision with an obstacle to generate a preliminary path; After the preliminary path is generated, the path is optimized by taking the kinematic constraint of the robot into consideration, and the global path planning is obtained.
2. The building model reconstruction method based on the quadruped robot and the machine vision, which is characterized in that the step S1 specifically comprises the following steps: adopting a tight coupling-based graph optimization framework to fuse three-dimensional point cloud data and IMU pose data, and generating optimized pose estimation on the fused data by utilizing a pre-integration technology and a dynamic weight distribution mechanism; Performing loop detection by combining laser radar information and visual information, and generating a globally consistent pose-map structure through a sparse pose map and an LM algorithm; Generating an incremental three-dimensional occupied grid map based on a pose-map structure with optimized pose estimation consistent with the global state; The three-dimensional occupancy grid map is converted to a global map of a two-dimensional grid for location initialization and navigation.
3. The building model reconstruction method based on the quadruped robot and the machine vision, which is characterized in that the local path planning comprises the following steps: When the four-foot robot executes a scanning path planned by a global path, a self-adaptive Monte Carlo positioning algorithm is adopted to realize real-time positioning; based on the current speed and acceleration limit of the quadruped robot, a dynamic speed window is defined by adopting an improved dynamic window algorithm, each possible speed vector is evaluated in the speed window, the distance between the quadruped robot and an obstacle and the possibility of approaching the target point when the quadruped robot moves with the speed vector are calculated, the optimal speed vector is selected, and the quadruped robot is guided to move.
4. The building model reconstruction method based on the quadruped robot and the machine vision, which is characterized in that the preprocessed point cloud data acquisition method specifically comprises the following steps: Downsampling the original point cloud data Q by adopting an octree-based downsampling method to generate downsampled point cloud data Q'; using a conditional filtering algorithm to remove noise from the processed point cloud data Q ', and filtering abnormal points and outliers to obtain processed point cloud data Q ' '; And removing point cloud data irrelevant to the target area according to the boundary condition of the reconstruction target area to obtain point cloud data Q ' ' ', and taking the point cloud data Q ' ' ' of the reconstruction area as preprocessed point cloud data Q ' ' '.
5. The building model reconstruction method based on the quadruped robot and the machine vision, which is characterized in that the reconstructed three-dimensional building model acquisition method specifically comprises the following steps: using an iterative nearest point registration algorithm to align the preprocessed point cloud data Q ' ' '; Constructing a visual feature-assisted Gaussian mixture model registration model, and further aligning point cloud data; Performing missing region complementation of the point cloud data to generate a registered and complemented point cloud set Q aligned ; And inputting the point cloud set Q aligned into a Poisson surface reconstruction algorithm to reconstruct the surface, and generating a reconstructed three-dimensional building model.
6. The building model reconstruction method based on the quadruped robot and the machine vision, which is characterized in that the Poisson surface reconstruction algorithm specifically comprises the following steps: Carrying out normal vector estimation by adopting a fitting method based on a triangular mesh structure, and estimating a normal vector n i for each point q i in the registered point cloud, wherein the normal vector n i is used for defining the local geometric characteristics of the point cloud surface; poisson surface reconstruction is carried out by constructing a global optimization problem based on poisson equation, taking a normal vector in point cloud as a source term, and solving a continuous scalar field phi; and extracting an isosurface from the scalar field phi through a volume rendering algorithm, namely the reconstructed three-dimensional building model.

Description

Building model reconstruction method and system based on four-foot robot and machine vision Technical Field The invention relates to a building model reconstruction method and system based on a quadruped robot and machine vision, and belongs to the technical field of civil engineering and automation interaction. Background The traditional existing building three-dimensional reconstruction method based on laser point cloud mainly depends on manual frame station scanning, and the data acquisition efficiency and accuracy are improved. However, the method is complex in operation, time-consuming and easily affected by human factors, and still faces practical problems such as insufficient scanning coverage in a complex environment, so that the requirements of large scale and high precision are difficult to meet. In recent years, four-legged robots are increasingly used in complex terrain by virtue of their excellent obstacle-surmounting capability and flexibility. The quadruped robot can autonomously move in rugged, narrow and complex building environments, and the data acquisition capacity of difficult-to-reach areas is remarkably improved. However, in the prior art, the application of combining the quadruped robot and the laser point cloud technology to the three-dimensional reconstruction of the existing building is not common, and the related research and the practical application are few, so that the efficient, comprehensive and accurate three-dimensional modeling is difficult to realize in the practical application. The prior art has the technical challenges that firstly, scanning points are selected through artificial subjective factors, so that the scanning efficiency is low and the global coverage rate is poor, and secondly, the track and the gesture of a robot in the moving process cannot be accurately controlled, the stability, the safety and the accuracy of the scanning process are poor, and high-quality point cloud data cannot be acquired. Therefore, developing a high-efficiency and accurate three-dimensional reconstruction method and system for an existing building model based on a quadruped robot and machine vision becomes a technical problem to be solved. Disclosure of Invention The purpose is as follows: in order to overcome the defects in the prior art, the invention provides a building model reconstruction method and a system based on a quadruped robot and machine vision, the invention can fully exert the mobility of the quadruped robot and the high precision of the laser point cloud, and realize comprehensive, efficient and accurate three-dimensional reconstruction in a complex building environment. The technical scheme adopted by the invention is as follows: in a first aspect, a building model reconstruction method based on a quadruped robot and machine vision specifically includes: and S1, acquiring three-dimensional point cloud data of a reconstruction region and IMU pose data of the quadruped robot, and performing environment sensing and mapping on the reconstruction region by adopting a synchronous positioning and map building algorithm to obtain a global map. And S2, determining target scanning points and corresponding scanning paths thereof based on a global map by combining a three-dimensional ray tracing algorithm and an A-path searching algorithm, and generating a global path plan. And step S3, planning the quadruped robot to start walking according to the generated global path, and realizing real-time global positioning of the quadruped robot based on a self-adaptive Monte Carlo positioning algorithm and combined with sensor data in the walking process of the quadruped robot. And adopting an improved DWA algorithm to conduct local path planning. And S4, stopping the four-foot robot after reaching a target scanning point, and starting a laser scanner to acquire point cloud data and RGB data. After the acquisition is completed, the quadruped robot moves to the next target scanning point according to the global path planning in the step S2 and repeats the step S3 until the acquisition of all original point cloud data is completed. And S5, preprocessing all the acquired original point cloud data to obtain preprocessed point cloud data. And S6, adopting a multi-mode data fusion registration method to the preprocessed point cloud data to obtain a reconstructed three-dimensional building model. Preferably, the step S1 specifically includes: and fusing three-dimensional point cloud data and IMU pose data by adopting a tightly coupled graph optimization framework, and generating optimized pose estimation on the fused data by utilizing a pre-integration technology and a dynamic weight distribution mechanism. And performing loop detection by combining laser radar information and visual information, and generating a globally consistent pose-map structure through a sparse pose map and an LM algorithm. And generating an incremental three-dimensional occupied grid map based on the pose-map structu