图像和惯性传感器相结合的摄像机定位和物体三维位置估计  

Estimation of camera location and objects' 3D position combining image and inertial sensor

作　　者：郭恩特 陈志峰[1] 范振嘉 黄立勤[1] 潘林[1] GUO Ente;CHEN Zhifeng;FAN Zhenjia;HUANG Liqin;PAN Lin(College of Physics and Information Engineering,Fuzhou University,Fuzhou,Fujian 350116,China)

机构地区：[1]福州大学物理与信息工程学院,福建福州350116

出　　处：《福州大学学报（自然科学版）》2018年第4期481-489,共9页Journal of Fuzhou University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61471124);福建省高校产学合作基金资助项目(2017H6009);福州大学人才基金资助项目(XRC-1601);赛尔网络下一代互联网技术创新基金资助项目(NGII20160208)

摘　　要：为解决无人驾驶中车辆定位与周围场景中物体三维位置估计,采用卷积神经网络(CNN)检测图像中的物体,用扩展卡尔曼滤波(EKF)方法融合惯性传感器测量得到的加速度和角速度,同时估计摄像机位置和物理世界中物体三维位置.图像结合惯性传感器(IMU)信息,克服了单目摄像机估计得到的摄像机位置和物体三维位置的尺度不确定性;结合卷积神经网络检测物体提高特征点匹配准确度,实现对物体在三维世界中的位置通用的估计.在实验部分用Matlab分别模拟仿真场景和现实场景的数据库KITTI,有效估计摄像机运动和场景中物体三维位置估计.In order to estimate the three-dimensional position of the objects in the surrounding scene and camera location,we use the convolution neural network to detect the objects in the image combining the acceleration and angular velocity measured by the inertial sensor with the extended Kalman filter method. The image combined with the information from inertial sensor overcomes the scale uncertainty of the position of camera and objects. Combining CNN can improve the accuracy of feature point matching,meanwhile estimating the position of the object in the three-dimensional world. In the experimental,we use Matlab to simulate a scene and use KITTI database,which can effectively estimate the three-dimensional position camera and objects in scene.

关 键 词：卷积神经网络 扩展卡尔曼滤波 摄像机定位 物体三维位置估计 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]