發展自成長類神經網路嵌入式低成本MEMS INS/GPS整合式定位定向演算法

The Developmen of Self-Growing Neural Network Embedded POS Determination Scheme for MEMS IMU/GPS Integrated Systems

數位式移動測繪系統是一種整合數位影像與直接定位定向的技術。利用多張動態數位影像直接定位定向以直接計算該影像於拍攝時間點其影像製造平台的空間軌跡之時變位置與時變姿態參數。現今最通用之技術為使用全球定位系統作為位置感測的主要元件同時使用慣性導航系統作為姿態感測的主要元件。並使用卡曼濾波器作為即時性INS/GPS整合系統的最佳預估工具。該演算法可提供具有高計算效率及可預估狀態的連續過程。而後處理模式的RTS又具有能使用整條軌跡資料去估計軌跡的優點，如今商業化的移動式測繪系統即是使用RTS後向平滑演算法來提供高精度的位置與姿態解以運行直接定位定向之解算程序。本研究開發與分析整合微機電系統之慣性量測元件與GPS接收機，所提出的架構運用自成長式類神經網路之階層式最大關聯神經網路（CCN）以克服傳統技術KF所存在之限制。使用CCN於INS/GPS整合位置與姿態系統POS較多層前饋式網路（MFNN）更具有彈性位相關係並可從新軌跡所獲得之知識自行成長神經元已吸納新資訊及提供較穩定的誤差補償等之優點，將對此領域之自動化控制之研究相當有幫助。 The technique that integrates digital imaging with direct-georeferencing is known as aDMMS(Digital Mobile Mapping System). This is a new alternative to directly decide thetime-variable position and orientation parameters for orienting the images in space, trajectory andattitude of the imaging platform by using different images. The most common technologies usedfor this purpose today are using GPS as a major position sensor, while the INS is the majororientation sensor. KF has been considered as the optimal estimation tool for real-time INS/GPS.Post-mission processing, compared to real-time filtering, has the advantage that data of the wholemission can be used to estimate the trajectory. Most of the commercial mobile mapping systemsuse an optimal smoothing algorithms to provide accurate positioning and developing an alternative data fusion scheme that integrates the outputs of a low cost orientation for directgeo-referencing. This study exploits and analyzes the idea of MEMS IMU and GPS receivers. Theproposed scheme is implemented using a constructive neural network CCN to overcome thelimitations of conventional techniques that are based on the KF. The CCN applied in this research hasthe advantage of having a flexible topology, growing neurons from new data sets in different trajectoryand stable outputs if compared to the recently utilized MFNN for INS/GPS integrated POS. Theseattractive factors can make the proposed schemes have more automatically characteristic.

The technique that integrates digital imaging with direct-georeferencing is known as aDMMS(Digital Mobile Mapping System). This is a new alternative to directly decide thetime-variable position and orientation parameters for orienting the images in space, trajectory andattitude of the imaging platform by using different images. The most common technologies usedfor this purpose today are using GPS as a major position sensor, while the INS is the majororientation sensor. KF has been considered as the optimal estimation tool for real-time INS/GPS.Post-mission processing, compared to real-time filtering, has the advantage that data of the wholemission can be used to estimate the trajectory. Most of the commercial mobile mapping systemsuse an optimal smoothing algorithms to provide accurate positioning and developing an alternative data fusion scheme that integrates the outputs of a low cost orientation for directgeo-referencing. This study exploits and analyzes the idea of MEMS IMU and GPS receivers. Theproposed scheme is implemented using a constructive neural network CCN to overcome thelimitations of conventional techniques that are based on the KF. The CCN applied in this research hasthe advantage of having a flexible topology, growing neurons from new data sets in different trajectoryand stable outputs if compared to the recently utilized MFNN for INS/GPS integrated POS. Theseattractive factors can make the proposed schemes have more automatically characteristic.