具磁力計自我校正功能之磁性輔助視覺慣性測程器

Abstract

由於直接將慣性量測裝置的量測資訊進行積分以估測行進軌跡的方法存在著誤差累積的問題，多數研究以感測器融合的技術為架構提出了許多方法來解決前述問題，以期望得到更可靠的估測軌跡。將慣性量測裝置與攝影機進行感測器融合即為熱門的方式之一。慣性量測裝置一方面可利用攝影機豐富的二維量測資訊抑制誤差的累積，另一方面亦可提供攝影機缺乏的地圖尺度資訊，進而得到更可靠的估測結果。 然而攝影機於特定的環境或運動下，例如當攝影機行經存在多個相似度高的物件聚集的環境或處於快速運動的狀態導致影像模糊時，此時影像所提供的量測資訊是不夠可靠的。因此本論文將可用來估測移動方向的磁力計與單一攝影機、慣性量測裝置進行感測器融合，提出了一套測程器以得到更可靠的估測軌跡。 磁力計提供的量測資訊除了可用來進行姿態估測的地球磁場外，亦存在著影響估測結果的非地球磁場，因此本論文同時提出了以擴展型卡爾曼濾波器進行磁力計自我校正的方法。相較於一般測程器，本論文所提出的具磁力計線上校正功能之磁場輔助視覺慣性測程器具有無需事先針對磁力計偏移量進行校正、無需假設地球磁場在參考座標系下的分量為已知條件的優點。甚至於當載體行經有非地球磁場擾動的環境時，本論文所提出的方法依舊可以使用地球磁場進行自我軌跡的估測。最後，本論文以一套整合了磁力計、單一攝影機與慣性量測裝置的硬體設備錄取真實環境下的資料來驗證本論文所提出的演算法。

Due to error accumulation problem, it is hard to get reliable results of ego-motion estimation by using IMU only. Many methods based on sensor fusion have been proposed to solve this problem. Combining cameras and IMU together is a good choice to get reliable result. By using the information of cameras, the accumulation error from IMU can be constrained. On the other hand, IMU can provide real scale which cameras lack. When cameras pass by the environment where contains similar things, or moves too fast to blur the image, the information from cameras is not robust enough. Therefore, this thesis presents an odometer which combines a magnetometer, monocular camera and IMU to get more reliable results of ego-motion estimation. Because the information from the magnetometer contains the earth's magnetic field and interference, this thesis also presents a magnetometer self-calibration method based on extended kalman filter (EKF). Compared with traditional methods, the proposed method does neither need to pre-calibrate the bias of magnetometer, nor assume the earth's magnetic field vector in reference frame is known. The proposed method is capable of operating even if the disturbance of the earth's magnetic field occurs. Finally, a hardware which integrates with a IMU which has 10 degree of freedom, monocular camera and GPS is used to validate the proposed method in real environment.