| 摘要: | 在本論文中，我們提出能跟隨特定物體移動的自走車系統。利用即時影像及目標物偵測與定位技術，使自走車具有主動影像追蹤與跟隨目標物的移動控制。
本系統的設計規劃是由硬體架構與軟體流程兩個部份整合而成。硬體架構可以細分為四個次系統：微電腦系統、影像處理系統、動力驅動系統、及通訊傳輸系統。每個次系統代表著不同的硬體與周邊介面。系統的運作是利用微電腦系統進行所有影像處理與演算法程序的實現與計算，並統合控制所有周邊介面；例如，相機模組的影像擷取控制、馬達正反轉控制的訊號傳輸、遠端個人電腦的通訊等。
軟體流程則可以分成三個部份：目標物的偵測與定位、目標物的追蹤、與車體運動系統的控制。目標物的偵測與定位，是運用色彩偵測法的方式，先是以高斯混合模型觀察出色彩範圍的定義並選用對光線變化較不敏感的YCbCr色彩空間來使用。後續再經過二值化、形態學雜訊去除、再利用輪廓搜尋法來取得目標物影像區塊的特徵資訊。而目標物追蹤的處理流程，則是事先以校正法建立一個影像像素與真實世界距離之間的轉換關係式查找表(look-up table)，用來給之前取得的特徵資訊進行影像中像素變化與真實世界距離遠近的轉換計算。此距離可以結合目標物在影像中的左右位置估計目標物相對於相機中心線的左右方向。後續在利用三角函數來計算求出直角三角形其餘的斜邊與銳角角度，之後就可以根據此距離與方向進行車體運動系統的直行、旋轉、及停止的驅動控制。
最後本研究將軟體演算法與硬體裝置整合，完成一個利用影像中目標物特徵資訊的測距方法，並且實現目標物跟隨的自走車控制系統。
;In this thesis, we propose a automatic vehicle system, which can track the specified moving object. Use the real-time video and target detection and location technology, then the automatic vehicle can achieve active video tracking by image and follow this moving object control.
This main system is integrated by two parts. One is the architecture of the hardware and the other is software flow control with image processing algorithms. The architecture of the hardware could be implemented by the four sub-systems: There are MCU system, image processing system, the power driving system, and the communication system. Each sub-system has individual hardware, interface, and peripheral. The main system operates by all sub-systems, performs image processing algorithm of MCU, and controls all peripherals by interface. For example, the MCU controls the camera module of image acquisition, send direction signals to the motor controller, and communicates with remote PC, etc.
Software flow control and algorithms can be implemented by three parts: The first one is Object detection and location, the second is object tracking, and the last is the motion control of automatic vehicle. The Object detection and location utilizes color detection, which define the range of colors by Gaussian Mixture Model (GMM) and transform to YCbCr color-space for processing, the YCbCr color-space is lighting-insensitivity. After the YCbCr color-space transformation, we can get the feature from target object image by thresholding, morphology noise removal, and contour searching method, and furthermore the object tracking obtains a look-up table by calibration method. This calibration method establishes a transform between image’s pixels and real world’s distance. This distance can be used to estimate the target is in left or right relative to camera center line. Following that, we use trigonometric function to evaluate the hypotenuse and acute angle of the right triangle, then to control the driver of the automatic vehicle going forward, rotation, and stop.
Finally, in this research, by the integration of software algorithm and hardware device that we use the image feature distance measurement method to approach controlling the automatic vehicle that can follow the specified object. |
