利用車道資訊與停等特徵達成預先交疊偵測與解析

Abstract

道路車輛資訊的蒐集在先進交通控制系統中是不可或缺的重要環節；而車輛偵測系統為收集交通流量與行車速度等路況資訊之重要基礎設備。本論文提出一套利用車道資訊與停等特徵，達成預先交疊偵測與解析之即時多車輛偵測與追蹤系統；將前視(Forward-Looking)的彩色影像輸入動態前背景切割處理後，可將移動物體從影像中切割出，接著利用車輛追蹤技術，紀錄車輛在偵測範圍中的路徑；為了提高追蹤的正確率，我們利用自動車道偵測所得到之車道線資訊，提出了以車道為基礎的去陰影方法以及交疊偵測與切割。在得到車輛的追蹤軌跡之後，此系統可計算並提供行車速度、車流量統計等交通參數，並可分別偵測與辨識出大車、小車及機車，以及得到車輛的移動方向。除此之外，本研究亦提出了一套停等車輛偵測及分割演算法，能有效地解決車輛在停紅綠燈或是塞車的情況中，因為交疊而無法保留追蹤路徑問題，使得車輛追蹤能夠持續且正確地進行，以便後續交通參數的計算。

A multiple-vehicle detection and tracking (MVDT) system with prior occlusion detection and resolution by lane information and queue features has various applications – tracking and classifying vehicles, determining traffic parameters, moreover, detecting violations of vehicles. The characteristics of MVDT system are real-time operation, robustness, precision, and ease of setup which are all important consideration for a vehicle detector. In this study, the proposed tracking reasoning is applied to track vehicles after the dynamic segmentation firstly. Next, some functional methods, such as lane-based run-length shadow suppression and prior lane-based occlusion detection and resolution, are proposed to enhance the accuracy of the tracking processing. Then, the edge-based queue detection and resolution is exploited to keep tracking trajectories when vehicles are waiting at the traffic lights or at traffic jam. Furthermore, tracking trajectories and the lane mask are applied to derive the values of traffic parameters and the direction of vehicle movement. Finally, for ease of setup, the adaptation of the system parameters is proposed.