即時影像處理技術應用於智慧型運輸系統之研究

Abstract

智慧型運輸系統依照服務的對象，共提出了九項服務，大致可歸類為道路、車輛以及用路人服務項目。因為電機相關產業的蓬勃發展，近年來已分別投入了許多研究能量進行研究。其中，雷達微波技術、影像處理技術、紅外線技術、GPS定位與無線網路技術等，皆已從各自原本的應用領域被帶入智慧型運輸系統領域。然而，大多數研究仍舊專注於單項服務之研究，對於跨項目的整體服務串聯架構，缺乏一個具有整合性的規劃。本論文的精神，首先在於一個整合這三大類服務項目的架構，基於這個架構，而發展出多項以影像處理為基礎之相關研究，預期能成為實現這個架構的基礎。 首先，本論文針對智慧型車輛項目，提出以單一攝影機之車輛前方車道線偵測技術，針對不同的天候與夜晚皆適用的方法。以這個演算法為基礎，本論文完成了多項相關的成果，包括結合了車輛橫向控制系統，實現自動車道保持系統；移植演算法於嵌入式平台，實現車道偏離警示系統。 這個前方影像偵測的方法，同時結合一個高效能的影像壓縮技術，完成了一個創新的影像式行車紀錄器，國外雖有相關的想法提出，整合即時影像車道偏離偵測技術仍尚未被提出。這個系統，針對事故發生後的現場重建與責任釐清有很大的幫助。 接著，我們針對交通參數蒐集項目，提出一個即時車輛偵測技術，應用於影像式車輛偵測器。這個偵測器的成本遠遠低於目前國外相關系統，增加了我國交通部廣佈車輛偵測器構想的可行性。這個技術，分為兩個版本，一個是針對通用性的狀況，適合架設於道路中段，進行較多項目的交通資訊蒐集，另一個是針對路口紅綠燈前，車輛因為壅塞或是等候燈號而慢速前進，甚至停住不動的狀況。經過實路的測試，也經過交通部運輸研究所相關人員驗證，有97%的偵測正確率。 最後，我們將車輛偵測技術，直接整合至目前已架設在各道路口與高快速道路的閉路電視系統(CCTV)，基於節省國家建置成本的考量下，將目前的CCTV系統進行車輛偵測加值服務。以測試結果看來，演算法在不經過修改的情況下，可直接應用於高快速道路上的CCTV，同時保有相同的偵測準確率。

With the rapid development of information and communication technologies, it provides the opportunity of applying these technologies to intelligent transportation system (ITS). ITS related services have been developed in developed countries such as those in Europe, America and Japan for the past thirty years. Although the development of ITS in Taiwan is a little bit late, various technology developments and practical validation projects have been conducted by government agencies, private industries and academic institutions in view of ITS development potentials. From the master plan of intelligent transportation systems in Taiwan, nine service areas are announced to construct the national ITS framework. In the decade, the sensing technologies are brought to ITS related investigations, e.g. microwave, image processing, infra-red, global position system, wireless network, etc. However, these projects belong to the individual studies and lack for the integration consideration. The innovative service plan namely VERI system consists of Vehicle detection, Embedded service unit, Road side server and Information center is addressed in the beginning of this dissertation, which is an integration structure of AVCSS and ATIS service areas. The following real-time image processing approaches are presented for the applying to VERI system. The first part focuses on AVCSS topic. The real-time robust lane detection approach presented for the integration with lane keeping and lane departure warning systems. In the lane keeping system, the lane detection approach provides the necessary roadway information to the lateral control unit of our TAIWAN iTS-1 smart car to accomplish the automatic lane keeping. The lane detection approach is also implemented on the low power DSP system to be a lane departure warning system to remind the drivers when the vehicle departure from the lane. Furthermore, a wavelet based video compression approach is addressed to combine with the lane detection approach to be a novel vision-based driving recorder. The front-view image will be recorded when the departure event occurs. These compressed vide data can be the evidence of the accident identification. The second part of this investigation focuses on the traffic information collection for ATIS. A real-time vehicle detection approach is proposed for the image-based vehicle detector. A color background is extracted to segment the moving vehicles with the luminance adaptive. Moreover, an advanced approach for the vehicle detection in traffic jam condition is presented. The vehicles move slowly or even stop on the roadway for traffic jams or waiting for the traffic lights. Finally, our vehicle detection algorithms are combined with the closed-circuit television (CCTV). The CCTV video streams are compressed for the limited network bandwidth so that the distorted images are acquired for detection. The experiments show that our algorithm is also robust for the CCTV application with good performance.