- 學位論文
應用立體視覺之自動停車系統設計與實現
Design and Experiment of an Automatic Parking System with Stereo Vision
摘要
本研究中，旨在於設計出一套可裝置在車輛上的立體視覺自動停車系統，藉此系統偵測停車環境中的資訊，並利用車輛動態數學模型以及依循規劃的停車路徑執行自動停車於目標停車格之中，以提升停車時的效率。 本研究分為二個主要部分。首先是影像感測方面，傳統影像系統主要是使用單眼視覺，其優點在於系統較為簡單，但是無法正確的得知目標與攝影機的距離，故利用雙CCD(Charge Coupled Device，電荷耦合元件)相機來擷取含有停車格之左右成對影像，經由影像前處理及型態學等方法分割出僅只有目標停車格線的影像，再來藉由立體視覺演算法估算出目標停車格線與攝影機的三維空間資訊。第二部份則是停車控制系統，藉由影像處理後得知的停車格幾何關係，判別停車模式為路邊停車或倒車入庫模式，並再以車輛與停車格的相對位置、角度以及停車模式設計停車路徑，且利用模糊邏輯控制器控制車輛跟隨參考路徑完成停車任務，最後並以實驗方式驗證之可行性。
並列摘要
In this thesis, it is aimed at designing an automatic parking system with stereo vision for vehicles. The system can collect information about parking space. By using the vehicle models and palnned parking trajectories, this system can drive a vehicle automatically into a target parking space with high parking efficiency. This research divides into two main parts. The first part is about vision system. The conventional vision systems used in automatic parking control are mainly monocular vision. The monocular vision is simple in stucuture and processing, but hard to get correct distance between the target and the camera. The dual-CCD system (charge-coupled device) can take a pairs of pictures of target parking space. Through image preprocessing and morphology, the parking space can be separated from the background. And the correct dimensions of parking space can then be obtained from the stereo vision information by stereo vision algorithms. The second part is the fuzzy parking control system. This system determines parallel or garage parking model by geometric relation between the target space and the camera. According to the position and angles between vehicle and parking space, the control system controls vehicle to follow planned trajectory and to complete the parking task. Finally, experiments are performed to verify the simulation results.