近年來，行車安全愈來愈受到人們重視。為了改善行車安全，本論文使用一部架設在車上的相機擷取前方道路影像以偵測道路及行車狀況。本研究所包含的偵測工作有：車道線偵測、多車道位置估計、車道線分類及前車距離估計。 車道線是藉由在影像中搜尋車道線模型中的最佳參數來求得，同時藉由限制車輛的偏轉角度來減少搜尋空間以達到即時偵測的需求，並利用累積邊緣點的方式將車道線分類。多車道位置估計則是利用一限制的公式，不需額外地搜尋就可估計出相鄰車道的位置。相機相對於空間座標系統的傾斜角度和偏轉角度可以由車道線的消失點推論得知，因此前車距離可藉由得到傾斜角估計求出。 實驗結果顯示本研究所提出的方法可以有效地偵側出本車道與相鄰車道的車道線位置，前車距離估計與車道線分類可提供更多有用資訊給駕駛人。在 Pentium 4 2.4GHz 的一般個人電腦上執行車道線偵測、虛實車道線分類、多車道位置估計、與偏離車道警示所需的平均處理時間約為0.027秒。 Recently, people pay much attention on driving safety. To improve the driving safety, we used a camera mounted on the vehicle to capture road scenes for road and driving situation detection. The detection tasks in this study include lane marking detection, multi-lane estimation, lane marking classification, and front-vehicle distance estimation. The lane markings are detected by searching the optimal parameters of a defined lane model on the images. By restricting the yaw angle of the vehicles, the searching space can be reduced to achieve the real time requirement. The multiple lane estimation method estimates the adjacent lanes based on a restricted formula without any search. The lane markings are classified based on a proposed method of accumulating edge pixels. The tilt and pan angles of camera related to the world coordinate system can be inferred from the vanishing point of lane markings. The distance to a front vehicle is then estimated from the acquired tilt angle. The experimental results show that the proposed method can detect the current lane markings and adjacent lanes efficiently. The estimated distance and lane marking classification are good enough to provide more information to the drivers. The whole processing time of the proposed system is about 0.027 seconds in average on a 2.4 GHz Pentium-based personal computer.
