基於視覺的多車輛速度估算以及交通資訊蒐集;Vision-Based Multiple-Vehicle Tracking and Speed Estimation for Traffic Information Collection

[[abstract]]交通監控影像中包含相當多的交通資訊，比如道路是否正在塞車或者是否有事故發生，若能知道這些資訊，交通管理單位就能即時應變處理，用路人也能提早改道，不過這都必須建立於能利用影像技術從交通監控影像分析出這些資訊的前提下。本論文主要目的在於透過影像處理與電腦視覺的相關技術追蹤交通監控影像中的車輛並蒐集車速、車流等重要交通資訊。我們在白天和夜間採用不同的方式偵測影像中的車輛，白天是透過替影像建立一可靠的背景模型，以背景相減的方式分割出影像中的前景；夜間場景是以車輛的車燈作為偵測對象。車輛在連續影像裡位置的連續性與外形的相似性會被用於車輛追蹤，藉此了解車輛在一段時間內的移動情形。並以相機校正把原本影像平面上的訊息轉換成更符合實際場景的訊息。在實驗部分對拍攝的交通車流影像利用所提出的系統進行處理，實驗結果證明系統可以有效取得交通資訊。 Traffic surveillance videos contain a lot of traffic information, such as whether the road is congested or whether an accident occurred. Once this information is obtained, the traffic management department will be able to react immediately, and the drivers can change the route earlier.The main purpose of this thesis is to trace the vehicles in a traffic surveillance video via image processing techniques and to collect important traffic information such as speed and traffic volume. Different methods are used to detect vehicles in daytime and nighttime images. In the daytime, a reliable background model is created for a road scene, and the foreground in an image is divided by background subtraction. In night scenes, the headlights of each vehicle are selected as detection targets. The position continuity and shape similarity of a vehicle in successive images are used in vehicle tracking to understand the movement of the vehicle over a period of time. The camera calibration step then converts the information from the original image plane into realistic information. Experimental results demonstrate that the system can effectively obtain the desired traffic information