CAJ | 학술논문

提出一种单目视觉里程计/捷联惯性组合导航定位算法.与视觉里程计估计相机姿态不同,惯导系统连续提供相机拍摄时刻对应的三维姿态,克服了单纯由视觉估计相机姿态精度低造成的长距离导航误差大的问题.通过配准和时间同步,用惯导系统解算的速度和视觉里程计计算的速度之差作为组合导航的观测量,采用Kalman滤波修正组合导航系统的误差,同时估计视觉里程计标度因数误差.分别在室内外不同环境下进行了22 m的推车实验和1412m的跑车实验,定位误差分别为3.2%和4.0%.与Clark采用姿态传感器定期更新相机姿态估计结果的方法相比,单目视觉里程计/惯性组合导航定位精度更高,定位误差随距离增长率低,适合步行机器人或轮式移动机器人在复杂地形环境下车轮严重打滑时的自主定位导航.
제출일충단목시각리정계/첩련관성조합도항정위산법.여시각리정계고계상궤자태불동,관도계통련속제공상궤박섭시각대응적삼유자태,극복료단순유시각고계상궤자태정도저조성적장거리도항오차대적문제.통과배준화시간동보,용관도계통해산적속도화시각리정계계산적속도지차작위조합도항적관측량,채용Kalman려파수정조합도항계통적오차,동시고계시각리정계표도인수오차.분별재실내외불동배경하진행료22 m적추차실험화1412m적포차실험,정위오차분별위3.2%화4.0%.여Clark채용자태전감기정기경신상궤자태고계결과적방법상비,단목시각리정계/관성조합도항정위정도경고,정위오차수거리증장솔저,괄합보행궤기인혹륜식이동궤기인재복잡지형배경하차륜엄중타활시적자주정위도항.
A new algorithm is presented for monocular visual odometry/SINS integrated navigation,in which the camera attitude is provided by the SINS other than visual estimation.The low accuracy of visual attitude estimation which would cause much error for long-range navigation is avoided.After registration and time synchronization,the velocity computation difference between SINS and visual odometry is chosen as observation of integrated navigation.A Kalman filter is used to correct the integrated navigation error including the visual odometry scale factor error.The 22 m indoor and 1412 m outdoor navigation experiments show that the position errors are 3.2％ and 4.0％ respectively.It can be seen that,compared with Clark's method in which camera attitude estimation is updated periodically with attitude sensors,the proposed algorithm is more accurate and robust with low rate of error growth.The proposed algorithm can be applied to the autonomous navigation of walking robots or wheeled mobile robots in case of serious wheel slip in complex terrain.