以SOPC為基礎利用單一攝影機傾斜攝影之移動式機器人物體追蹤與定位系統
No Thumbnail Available
Date
2011-07-31
Authors
許陳鑑
Journal Title
Journal ISSN
Volume Title
Publisher
行政院國家科學委員會
Abstract
本計畫擬建置一台以可程式規劃系統晶片(System On Program Chip, SOPC)為基礎,具備 影像物體追蹤(object tracking)、影像式定位(image-based localization)、以及避障功能 (obstacle avoidance)之全方位移動式機器人。作法上係以單一CCD 作傾斜式之攝影(tilt photographing),據以發展出多種以傾斜式攝影為基礎之影像式距離及角度量測系統,藉 由與物體追蹤技術相結合,達到對於目標物之定位(localization),並將相關演算法實現 在可程式規劃系統晶片(System On Program Chip, SOPC)上,達到以SOC 晶片系統即時 處理及控制移動式機器人的目的。三年期之計畫共包含5 個研究子題,分別是:(1)全方 位機器人系統架構與建置,(2) 混合型物體影像追蹤系統,(3)影像式物體定位系統,(4) 模糊避障系統,以及(5) SOPC 系統之實現。 機器人系統架構上,本計畫擬設計一台具有全方位移動能力之機器人,視覺系統則 僅配置單一CCD 作傾斜式攝影,作為影像式距離量測、影像物體追蹤、以及影像式物 體定位等相關系統之基礎;機器人的週邊則配置有超音波感測器、紅外線感測器,利用 多重感測器資料融合技術,建立機器人之模糊避障(fuzzy obstacle avidance)功能。在影像 物體追蹤方面,本計畫擬以粒子濾波器演算法(particle filter, PF)、粒子群最佳化法(particle swarm optimization, PSO),以及單體搜尋法(Nelder-Mead Simplex Search Method, NM)為 基礎,擷取其長處予以整合,開發兩種混合型(Hybrid)目標物體追蹤法,配合特徵為基 礎之物體辨識技術(feature-based methods for object recognition),快速偵測目標物體,再 輔以所開發出來之多種影像式物體定位系統,以獲取所偵測到的目標物體相對於量測系 統的座標位置。在機器人避障系統方面,本計畫擬設計一模糊系統,將多種感測器資訊 融合成障礙物的距離與方向資訊,並透過模糊避障機制來決定機器人的移動方向與旋轉 角度,最後再以SOPC 實現各種相關演算法,利用FPGA 的硬體電路優勢,以軟硬體協 同設計(Hardware/Software Co-design)的技術設計各種硬體加速器電路,提升相關演算法 的執行效率,實現以SOC 晶片系統即時處理及控制移動式機器人的目的。
The objective of this project is to investigate the design and implementation of a SOPC-based object tracking and localization system via tilt photographing of a single CCD camera, providing image-based localization and obstacle avoidance for omni-directional mobile robots. Based on tilt photographing of a single CCD camera, two imaged-based distance and angle measurement methods will be investigated for integration with object tracking techniques to develop an image-based localization scheme suitable for use in mobile robots. Algorithms developed in this project will be implemented on a System On Program Chip, SOPC, to allow real-time processing and control of the mobile robot. To achieve the above-mentioned objectives, the three-year project will investigate the following major items: (1) Design and implementation of an omni-directional mobile robot, (2) Development of hybrid algorithms for object tracking, (3) development of an image-based localization system, (4) Design of a fuzzy controller for obstacle avoidance, and (5) Implementation of the developed algorithms on a SOPC. The architecture of the mobile robot is equipped with omnidirectional motion capabilities, which allows it to move toward any directions. A single CCD camera with the help of two laser projectors are only required in this project to provide tilt photographing, under which various image-based localization schemes will be developed. Multi-sensor data fusion will be investigated to construct an obstacle avidance fuzzy controller for the mobile robot. To improve the performance of object tracking, this project considers the development of a memetic algorithm incorporating an enhanced particle filter, NM simplex search method, and particle swarm optimization. As a result, an image-based localization system for obtaining the coordinate of concerned objects relative to the measuring system is expected to be developed with the help of feature-based methods for object recognition and the object tracking methods. To allow safe navigation of the robot for preventing possible collision, multi-sensor fusion of the sensing data will be considered to design a fuzzy obstacle avoidance controller. To further improve the performance of the system, algorithms developed in the project will be practically implemented on a SOPC by designing various hardware accelerators to achieve the objective of real-time processing and control of the mobile robot.
The objective of this project is to investigate the design and implementation of a SOPC-based object tracking and localization system via tilt photographing of a single CCD camera, providing image-based localization and obstacle avoidance for omni-directional mobile robots. Based on tilt photographing of a single CCD camera, two imaged-based distance and angle measurement methods will be investigated for integration with object tracking techniques to develop an image-based localization scheme suitable for use in mobile robots. Algorithms developed in this project will be implemented on a System On Program Chip, SOPC, to allow real-time processing and control of the mobile robot. To achieve the above-mentioned objectives, the three-year project will investigate the following major items: (1) Design and implementation of an omni-directional mobile robot, (2) Development of hybrid algorithms for object tracking, (3) development of an image-based localization system, (4) Design of a fuzzy controller for obstacle avoidance, and (5) Implementation of the developed algorithms on a SOPC. The architecture of the mobile robot is equipped with omnidirectional motion capabilities, which allows it to move toward any directions. A single CCD camera with the help of two laser projectors are only required in this project to provide tilt photographing, under which various image-based localization schemes will be developed. Multi-sensor data fusion will be investigated to construct an obstacle avidance fuzzy controller for the mobile robot. To improve the performance of object tracking, this project considers the development of a memetic algorithm incorporating an enhanced particle filter, NM simplex search method, and particle swarm optimization. As a result, an image-based localization system for obtaining the coordinate of concerned objects relative to the measuring system is expected to be developed with the help of feature-based methods for object recognition and the object tracking methods. To allow safe navigation of the robot for preventing possible collision, multi-sensor fusion of the sensing data will be considered to design a fuzzy obstacle avoidance controller. To further improve the performance of the system, algorithms developed in the project will be practically implemented on a SOPC by designing various hardware accelerators to achieve the objective of real-time processing and control of the mobile robot.