基於雙核心嵌入式平台之駕駛安全監控系統

Abstract

近年來，隨著現代人對於行車安全的日益關注，以及汽車電子相關技術的快速發展和相關硬體設備價格越來越親民，駕駛安全輔助系統已儼然發展成汽車的標準配備。然而駕駛安全輔助系統往往只提供瞬時的輔助，只在事件發生時警示駕駛。本研究提出一駕駛安全監控系統，提供危險係數給使用者做為駕駛安全的指標。以駕駛安全輔助系統為基礎，進一步分析與監控駕駛行為，長時間監控駕駛，預防潛在的危險。本系統使用攝影機搭配車道偏移警示演算法、前車防碰撞警示演算法再加上一三軸加速度計與GPS做為感測器，以一個二階段的演算法來推估駕駛行為的危險程度，除了偵測危險之外，更可知為何而危險。第一個階段為駕駛行為偵測，利用隱性馬可夫模型辨識正常駕駛、加速、減速、飄移、變換車道、逼近前車這六種行為。後推算出各行為之程度，第二階段則是利用模糊推論系統推估出危險係數。在監控的同時也將行車的影像以及相關狀態加以記錄，搭配一人性化之事件檢視器，提供方便的調閱記錄影像服務。本系統實做於Devkit8500D雙核心嵌入式平台上，並於高快速道路及平面道路進行實車測試，駕駛行為辨識演算法可達到99%以上之偵測率。

In recent years, with the growing concern for driving safety and the rapid development of automotive electronic technology, driving assistance systems have become essential equipment in vehicles. However, driving assistance systems only warn drivers on the occurrence of certain events to provide instant assistance. The main goal of this thesis is to develop a driving behavior monitoring system providing an intuitive indicator of the danger-level for driving safety. The proposed system is based on driving assistance systems, including a lane departure warning, and a forward collision warning system. In further, it also integrates information from an inertial measurement unit and a GPS module to analyze and monitor driving behaviors in the long term for preventing drivers from hidden dangers. A two-staged danger-level reasoning algorithm is proposed. The first stage uses Hidden Markov models to distinguish between normal driving, acceleration, deceleration, lane change, drifting, and approaching the front car. Then the degrees of each behavior are estimated. The second stage uses the fuzzy inference system to evaluate the danger-level. Moreover, drivers can easily access the interested part of recording videos with the customized video viewer. The proposed system was successfully implemented on the Devkit8500D embedded platform, and fully tested on highways, expressways, and urban roads. The experimental results show the accuracy ratio of 99% for the event recognition and can be used for driving safety.