智慧車側影像駕駛人輔助安全系統

Abstract

最近幾年，車輛安全是個非常重要的社會、經濟問題。一般來說，很多交通意外都來自於駕駛者精神狀態不佳、注意力不集中、未保持安全距離或甚至打瞌睡，在很多情形是駕駛者打瞌睡偏離車道而造成交通意外。而高速公路上，也很多意外是車輛側邊或是後方的直接碰撞。因為這些原因造成許多的意外，所以車道偵測與車側碰撞預防系統在智慧型交通傳輸計畫內扮演著相當重要的地位。以價格與效果為考量，本論文以裝置於車輛照後鏡的車側攝影機為感應器，且此攝影機搭配魚眼鏡頭，廣角的鏡頭，可能得到更多車輛附近的相關資訊。 因為攝影機位於車輛上，所以在車輛行進中，攝影機所取得的影像是晃動的，故在進行分析之前，必須先做影像穩定的修正，雖然只是前處理，但是可以簡化後續之演算法。本論文提出了車道偏移與車側碰撞警告系統，此系統不僅可以提醒駕駛人在偏移車道的時候是否會發生危險，透過車側碰撞警告系統更幫助駕駛者監看盲點區域是否有車輛進入，此外，我們亦借助交大腦科學中心對駕駛者的分析資料來推測駕駛者是否有打瞌睡或其他異常狀態。因為本論文只用車側攝影機，所以比較容易整合 車道偵測 與 車側碰撞警告系統，一般拍攝車前的影像式車道偵測系統，主要目的是提醒駕駛者不要偏離目前車道。而車側碰撞警告為了當有車輛進入盲點區而發出警告，但其整合比較麻煩且必須要兩套系統。故本論文的架構能更有效的提供駕駛者在變換車道時，是否會發生危險，且這個架構是非常特別且有用的，進而達到輔助駕駛人安全的目的。

In recent years, an important social and economic problem is traffic safety. In general, a considerable fraction of these accidents is due to driver’s fatigue, inattentive driving and driving without keeping proper distance. In many cases, the driver falls asleep will make the vehicle to leave its designated lane and possibly cause an accident. On the highway, the most important cause of traffic accidents is the lateral and same direction collision. Due to the inattentive driving, the driver may deviate from the correct lane orientation, which induces the traffic accidents. As a result, the lane detection system and vehicle lateral collision warning system play a significant role in improving the driver’s safety in a moving vehicle. For cost and performance consideration, a lateral fish-eye camera mounted under the rear-view mirror is chosen as our sensing device. The robust in-car DIS technique offers all major algorithms a stable image source. It is a minor pre-processing, but the following processing can be simplified massively. We propose an integrated system for lane departure warning and lateral collision warning. The lateral collision warning system aims at detecting the image in driver’s blind spot region and exporting signal to remind driver in the realistic driving environment. In addition, driver’s drowsiness will also be estimated by integrating the EEG-based analysis approach developed by the Brain Research Center, NCTU, into our lane departure warning system. In this thesis, for the lane detection, we develop a method for automatic region of interesting extraction only by analyzing the image contents captured by the lateral fish-eye camera without knowing the related camera parameters in advance. The lane-based stable system is integrated into the blind-spot lateral collision warning system to increase the better detection rate and provide more adaptive performance. Besides, by constructing the mechanism for drowsiness estimation in the dynamic driving environments, we can collect more data to further analyze other inattentive behavior of drivers so that the safety driving system can consider all possible risks caused by the internal or external factors of drivers as much as possible. Achieving these algorithms simultaneously by lateral camera is very novel and useful. Warning from this kind of system reflects real hazard and is really worth noticing, it does achieve the goal of “assisting in driving".