結合局部與全域特徵之新型車輛偵測系統

Abstract

近年來，基於影像式的車輛偵測技術在智慧型運輸系統中受到廣泛的重視與研究。然而，在交通流量大的場景偵測車輛仍然是一個困難且具有挑戰性的問題。在本研究中，我們提出了一個創新與可靠的自動化偵測系統。本系統先以車輛的邊緣強度與對稱性統計方法來假設車輛的位置以減少運算成本。接下來，以AdaBoost與決策型類神經網路(Probabilistic Decision-Based Neural Network)分類器分別對車輛的局部與全域特徵來進行確認車輛的位置，我們相信由兩個不同性質的分類器在特徵擷取有互補的效果，可以同時降低誤報率與達到高偵測率的效果，由我們蒐集的樣本資料庫測試結果，我們的系統可達到96.12%的偵測率，同時只產生0.0153%的誤報率；另外對公開的MIT CBCL資料庫測試，我們的系統可達到96.3%的偵測率，同時只有0.0013%的誤報率。 本研究的目標是在局部與全域上抽取出車輛的特徵，由此系統所得到的結果可以在車輛偵測的技術中，針對以背景為基礎的車輛之偵測系統所遭遇到的問題提供更好的解決方案。實驗的結果證實我們提出的系統可以不依賴背景資訊，偵測出影像中的車輛，我們所實作的系統也能為後續處理，e.g. 車輛追蹤、計數、分類、識別等應用，提供有用的資訊。

Vehicle detection techniques in visual-based Intelligent Transportation System (ITS) have been studied for years. However, to detect vehicles in a scene with heavy traffic is still a challenging problem. In this study, 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. The combination of two classifiers can learn the complementary relationship among local and global features, and it gains the extremely low false positive rate while still keeps high detection rate. For proprietary database, a 96.12% detection rate leads to a false-positive rate of approximately 0.0153%. For the MIT CBCL database, a 96.3% detection rate leads to a false-positive rate of approximately 0.0013%. The objective of this study is to extract the characteristic of vehicles in both local- and global-orientation, and model the implicit invariance of vehicles. This novel approach would provide a better solution to handle the problems encountered by conventional background-based detection systems. The experimental results proved the proposed system achieved a good performance of detecting vehicles without background information. The implemented system also extracted useful traffic information that can be used for further processing, like tracking, counting, classification and recognition.