- 學位論文
考慮攝影機FOV的條件下利用模糊推論的避障模擬系統
Obstacle Avoidance Simulation Using Fuzzy Inference Under Considering FOV of Camera
摘要
我們提出了路徑規劃與障礙回避的短程策略，一般而言，取得一個已知的地圖對於路徑規劃來說是必須的。然而，為了路徑規劃而即時繪製地圖，是不切實際的應用。我們採用的方式是只在車輛前方的裝置攝影機，並由攝影機取得駕駛的路況影像。因此，攝影機的可視範圍field-of-view (FOV)必須被列入考慮，當FOV越窄，可獲得的訊息就越少。另一方面，對於真實世界中車輛駕駛，進行障礙物閃避的決策時，仍有許多影響力必須應考慮到：道路的資訊、障礙物與車輛之間的距離、以及車輛的寬度。為了同時考慮這些影響力，採用了具有negative-rule的fuzzy。本研究中，由模糊推理系統（FIS）決策車輛的前進方向，並透過演進式的演算法（EA）之中的基因演算法(GA)，進行自動化的Membership Function參數調整。為了簡化訓練與測試所需的繁瑣過程，我們建立了一個測試與訓練的虛擬平台。並且在這個平台下，利用模擬的方式進行FIS的自動訓練。
並列摘要
We present a short term guidance strategy (aka local pilot): for path planning and obstacle avoidance. In general, a known map (global view) is necessary for path planning, however, to obtain the map is impractical in some application. While we consider that only the scene in front of vehicle is captured by camera, and hence the limitation of field-of-view (FOV) of a camera must be considered, the narrower FOV, the fewer information. In addition, for a real world driving application, there are some forces considered: the road information, distance between obstacle and vehicle, and vehicle width. In order to balance those forces, therefore, the negative-rule fuzzy is adopted. In this research, to substitute a driving expert, the fuzzy inference system (FIS) is training by evolutionary algorithm (EA). For simulation and verification of FIS, we build a testing and training platform in virtual scenes.