- 學位論文
創新式機器人三維視覺技術與雛型系統之研發
Development of an Innovative 3-D Robot Vision Technique and its Prototype System
摘要
本論文旨在發展一創新式機器人三維視覺技術與雛形系統，利用近紅外線雷射二極體光源，以及微小型光學分光鏡與高精度平面反射鏡，建構在麥克森干涉儀系統上，投射出高調變性之正弦式週期性條紋結構光，並以每秒30張之高速進行影像擷取，此外針對各個光學元件，設計出一機器人三維視覺技術之雛型系統。系統所產生出的週期性條紋結構光，在安全範圍下，可投影至五公尺遠之參考平面上，並且在此範圍中，進行物體三維形貌之量測，除此之外，搭配創新式之傅立葉轉換輪廓術進行高速三維表面重建，僅需單張取像即可進行三維表面影像重建，可有效避免因振動問題所產生之量測問題，系統在不同深度位置量測平均誤差度在5%內，故此發展可應用在高速三維機器視覺上。 目前發展出之三維視覺量測雛形系統，使用近紅外線雷射投影週型性條紋結構光，可克服量測黑色表面物體之光強不足問題，但對於吸光性質表面之物體略為不足，未來系統可針對此部分進行改進。
並列摘要
This work presents a new compact 3-D vision system for in-situ robots. A Michelson interferometric optical system equipped with a He-Ne laser diode, a set of miniaturized optical lenses and a tilting reference mirror was developed to generate interferometric fringes with a flexible projecting fringe period. The generated structured fringes can be projected within a large detection range up to 5 meters with a reasonable image contrast to be retrieved for satisfying demanding needs in accommodating surface reflectance variances of objects under 3-D surveillance. Meanwhile, employing the Fourier transform profilometry (FTP) principle with a bandpass filtering algorithm, the developed vision system can be employed to obtain high-speed 3-D robot vision by using single one-shot imaging. This work presents a new compact 3-D vision system for in-situ robots. A Michelson interferometric optical system equipped with a He-Ne laser diode, a set of miniaturized optical lenses and a tilting reference mirror was developed to generate interferometric fringes with a flexible projecting fringe period. The generated structured fringes can be projected within a large detection range up to 5 meters with a reasonable image contrast to be retrieved for satisfying demanding needs in accommodating surface reflectance variances of objects under 3-D surveillance. Meanwhile, employing the Fourier transform profilometry (FTP) principle with a bandpass filtering algorithm, the developed vision system can be employed to obtain high-speed 3-D robot vision by using single one-shot imaging. In the research, a miniaturized real-time 3-D vision system employing the proposed interferometric structured fringe projection and the elliptical bandpass filtering algorithms was successfully developed for 3-D machine vision with measurement accuracy reaching 5.6% of the overall detection range at measurement speed of 30 frames of 3-D maps per second. The high bandwidth of dynamic 3-D vision provided by the developed system can avoid potential in-situ environmental disturbances such as vibration from robot working space. The miniaturized probe can be further integrated easily with autonomous robots for achieving real-time dynamic 3-D local mapping and object recognition in the near future. The experimental results indicate that the developed system is capable of capturing 3-D maps of objects with an adequate level of reconstruction accuracy for precision robot manipulation. The work provides a feasible possibility in developing robust in-situ 3-D robot vision in the near future.