雙模影像式車輛偵測系統

Abstract

本論文提出一套雙模影像式車輛偵測系統，可適用於高速公路與十字路口的應用。首先我們提出一套新的即時數位影像穩定單元，採用修改過的比例積分控制器來移除攝影機錄影畫面的震動使畫面穩定。透過本論文提出修改過的比例積分控制系統估測的補償向量可以有效地移除震動。 在高速公路的應用上，我們提出一套新的且完整的架構。我們採用高斯混和模型的方法濾除掉背景並在測試畫面上偵測移動陰影。接著定義兩個非陰影的特徵，一個是物體內部的線條，另一個是用高斯模型去模擬修改過的陰影變暗因子。藉由多種特徵的結合定位出移動陰影的位置，便可以很輕易地抓出前景的物件。 在十字路口的應用上，繁忙的交通環境中偵測車輛是一大挑戰。我們提出一個新的自動車輛偵測架構。首先採用簡單的統計與對稱性的線索，找出可能有車輛的位置，降低後面搜尋比對的運算量。接著採用AdaBoost與統計決策神經網路兩種分類器分別由局部與全域特徵來驗證是否真的有車輛存在。 由實驗結果顯示本論文提出的方法可適用在不同的環境下。且經過與背景為基礎的方法比較亦顯示本論文提出的方法是有相當大的改進。

In this dissertation, a dual-mode video vehicle detection system and its applications for freeway and intersection are proposed. First, we presents a novel, real-time digital image stabilization (DIS) unit using a modified proportional integral (MPI) controller to remove stably unwanted shaking from an image sequence that is captured by a outdoors video camera. A compensating motion vector (CMV) estimation method with a modified PI control unit is proposed to remove the unwanted jitter. For freeway applications, we present a novel and complete structure in feature combination as well as analysis for orientating and labeling moving shadows so as to extract the defined objects in foregrounds more easily. Moreover, we make use of Gaussian Mixture Model (GMM) for background removal and detection of moving shadows in our tested images, and define two indices for characterizing non-shadowed regions where one indicates the characteristics of lines and the other index can be characterized by the information in gray scales of images which helps us to build a newly defined set of darkening ratios (modified darkening factors) based on Gaussian models. For intersection applications, to detect vehicles in a scene with heavy traffic is still a challenging problem. We present a novel automatic vehicle detection system. It first hypothesize potential locations of vehicles to reduce the computational costs by statistic of edge intensity and symmetry, then verify the correctness of the hypotheses using AdaBoost and Probabilistic Decision-Based Neural Network (PDBNN) classifiers, which exploits local and global features of vehicles respectively. Experimental results show that the proposed methods of this dissertation adapt to different conditions of image sequencing. The proposed methods also show effective improvements in different conditions of image sequence through in comparison with the background-based approaches.