基于立体视觉和扩展卡尔曼滤波器的运动估计  

Motion Estimation Using Stereovision and Extended Kalman Filter

作　　者：胡福志 黄影平[1] 魏宏建 胡兴 HU Fuzhi;HUANG Yingping;WEI Hongjian;HU Xing(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《信息与控制》2020年第5期570-577,共8页Information and Control

摘　　要：针对智能汽车避障问题,提出了一种相机自身运动情况下,基于双目立体视觉和扩展卡尔曼滤波器(EKF)的运动估计方法.检测阶段,借助立体视觉的三维重建能力依据目标位置分割目标获得感兴趣区域(ROI).跟踪阶段,通过光流法跟踪ROI内的边缘点,建立EKF预测与测量模型将自车运动、光流及视差融合在一起,更新获得优化的目标位置和速度信息.该方法针对自车运动平台,建立自车和目标相对运动模型,采用边缘点跟踪约束和随机采样一致性算法(RANSAC)剔除不可靠的跟踪点.以KITTI公共数据集提供的交通场景为测试对象,验证了方法的有效性.To address the issue of collision avoidance in intelligent cars,in this paper,we present a motion estimation method that uses binocular stereovision and extended Kalman filtering( EKF). In the detection stage,the region of interest is identified by segmenting obstacles based on their position by the three-dimensional reconstruction capability of stereovision. In the tracking stage,the optical flow is used to track the edge points within the target area. EKF is used to make the prediction and enable the measurement models to fuse the vehicle motion,optical flow,and disparity,and thereby obtain the optimized target position and velocity. We established a relative motion model with respect to the self-driving vehicle and objects. This method employs edge-point constraints and the random sample consensus algorithm to eliminate unreliable tracking points. We tested and verified the performance of the proposed method using traffic scenarios provided by the KITTI public data set.

关 键 词：目标检测 目标跟踪 运动估计 立体视觉 扩展卡尔曼滤波器 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]