基于Realsense传感器的机器人视觉里程计研究

Research on Robot Visual Odometry Based on Realsense Sensor

基于RGB-D传感器能直接获取空间环境的深度信息和图像信息的能力,采用最新的RGB-D传感器(Realsense),实现了室内移动机器人的视觉里程计。首先利用张正友相机标定法对Realsense传感器的彩色摄像头进行标定;接着采用ORB算法实现图像特征点的匹配,并用PROSAC算法进行外点剔除;最后采用PnP算法实现相机姿态的估计,利用g2o库对相机的姿态进行光束法平差优化,并绘出相机的运动轨迹,实现了基于RGB-D传感器的机器人视觉里程计。实验结果显示该方法能快速、准确地估计视觉传感器的位姿信息,实现移动机器人的精准定位和导航。

Visual odometer is one of the key technologies to realize robot’s autonomous localization and navigation. It can provide robot with real-time pose estimation and navigation through the vision sensor of mobile robot. However, the approach of monocular visual odometry can not directly obtain depth information and quickly confirm sensors’ pose. Furthermore, the approach of binocular visual odometry need complicated geometric operation for obtaining depth information. Thus, we proposed a method to realize robot’s visual odometry based on Intel’s Realsense through obtaining depth image and image of indoor scenes. Firstly, we used ZHANG Zhengyou’s camera calibration algorithm to calibrate RGB camera of Realsense sensors. Then, we used the ORB algorithm to match the image feature points, and used PROSAC algorithm to delete outlines. Finally, we used the Pn P algorithm to estimate the position and orientation of the camera, and constrained and adjusted the location and pose of the camera optically by using the g2 o library to draw the motion trajectory of the camera, thereby realizing the robot vision odometer based on the Realsense sensor. Experimental results show that this method can quickly and accurately estimate the location and pose information of the vision sensor, and realize the accurate localization and navigation of the mobile robot.