【摘要】：當(dāng)下很多機(jī)器人需要在工作場景中移動完成任務(wù),而地圖是移動機(jī)器人理解周圍環(huán)境并作出行動的基礎(chǔ)和關(guān)鍵。本文敘述了一種基于2D激光的環(huán)境三維地圖構(gòu)建方法,使機(jī)器人系統(tǒng)可以實現(xiàn)在動態(tài)運動過程中采集激光數(shù)據(jù)并構(gòu)建三維點云地圖。本文針對移動地圖構(gòu)建,以通過二維旋轉(zhuǎn)激光傳感器構(gòu)建三維點云地圖為目的,在系統(tǒng)架構(gòu)、傳感器數(shù)據(jù)處理、位姿估計等方面展開研究。主要研究內(nèi)容和成果如下:(1)設(shè)計了環(huán)境感知單元的軟硬件架構(gòu),并針對多傳感器原始數(shù)據(jù)時間戳不一致問題,提出了基于樣條平滑的多傳感器時間同步。對不同的傳感器采用了不同的時間同步方法,一般傳感器數(shù)據(jù)直接使用線性插值法,而對于存在時間不確定性的視覺里程計數(shù)據(jù),則利用重投影誤差和關(guān)鍵幀信息,采用樣條平滑方法實現(xiàn)時間同步。(2)構(gòu)建了 JDL 模型(Joint Directors of Laboratories Model)和 EKF(Extended Kalman Filter)相結(jié)合的多傳感器融合位姿估計方法。該方法采用JDL模型作為多傳感器融合架構(gòu),在JDL模型的態(tài)勢推斷層級,采用EKF方法融合基于ORB-SLAM2的雙目視覺里程信息和慣性傳感單元數(shù)據(jù)。通過兩者在位置變換估計和角度變換估計的優(yōu)勢互補(bǔ)提高了位姿估計精度。(3)實現(xiàn)了基于ICP(Iterative Closest Point)方法的在線和離線多幀地圖融合優(yōu)化。在線模式下,本文將多個點云局部地圖使用ICP的方法拼接融合在一起,得到最終的激光地圖。在保證基本的實時性的同時,最大可能地消除了實時姿態(tài)估計的累計誤差。離線模式下,基于在線模式的基礎(chǔ),添加了基于ICP方法的單個局部點云內(nèi)部優(yōu)化。
[Abstract]:Nowadays, many robots need to move in the work environment to complete the task, and map is the basis and key for the mobile robot to understand the surrounding environment and take action. In this paper, a 3D map construction method based on 2D laser is described, which enables the robot system to collect laser data and construct 3D point cloud map in the course of dynamic motion. In order to construct 3D point cloud map by two-dimensional rotating laser sensor, this paper studies the system architecture, sensor data processing, position and pose estimation and so on. The main research contents and results are as follows: (1) the hardware and software architecture of the environment sensing unit is designed, and the multi-sensor time synchronization based on spline smoothing is proposed to solve the problem of multi-sensor original data timestamp inconsistency. Different time synchronization methods are used for different sensors. Generally, linear interpolation is used directly for sensor data, while for visual odometer data with time uncertainty, reprojection error and key frame information are used. The spline smoothing method is used to realize time synchronization. (2) A multi-sensor fusion position and attitude estimation method based on JDL (Joint Directors of Laboratories Model) and extended Kalman filter (EKF) is proposed. In this method, the JDL model is used as the multi-sensor fusion framework, and the binocular visual mileage information and the inertial sensor unit data based on ORB-SLAM2 are fused by the EKF method at the situation inference level of the JDL model. The accuracy of position and pose estimation is improved by the complementary advantages of the two methods. (3) the online and offline multi-frame map fusion optimization based on ICP (iterative closed Point) method is implemented. In online mode, the local maps of multiple point clouds are fused together using ICP method to obtain the final laser map. At the same time, the accumulative error of real-time attitude estimation is eliminated as much as possible. In offline mode, based on the online mode, a single local point cloud internal optimization based on ICP method is added.
【學(xué)位授予單位】：浙江大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP242

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 蘇麗穎,譚民;移動機(jī)器人構(gòu)建地圖的研究與發(fā)展[J];中國科學(xué)院研究生院學(xué)報;2002年02期

，

本文編號：2127807