全天候嵌入式車輛盲點警示系統

Abstract

駕駛者在道路駕駛進行車道變換時，會利用側照後鏡來判斷鄰近車道的路況，但是因視線死角或駕者者的疏忽，常會造成意外事故的發生。因此本篇論文利用數位信號處理器來實現車輛盲點警示系統，當鄰近車道的車輛過近時，系統會發出警告提醒駕駛者注意變換車道的風險。本系統在高速公路與市區道路之下均可運作，並且克服各種天候因素，包括晴天、陰天、雨天和夜晚，並藉由影像式日夜切換演算法，分辨日間與夜間影像。在TI DM642平台上經過最佳化後，我們的系統可以同時偵測左右方盲點影像的車輛，並估測其距離，系統執行時間達到每秒20 張CIF的影像。 本文主要是利用影像處理與電腦視覺的技術，針對駕駛者的盲點區域做車輛障礙物的偵測，將兩隻攝影機分別架設於自車左右側照後鏡下方，利用攝影機所拍攝的盲點區域影像，在日間與夜間採用不同的車輛特性來偵測，在日間以車輛底部陰影作為盲點區域車輛的主要特徵，而夜間偵測成對車頭燈，並採用道路模型來設定盲點車輛偵測區間與估算自車至盲點車輛的距離。

Although side view mirrors can be used to assist the driver to determine if there is enough safe distance for lane change maneuvers, side collision accidents might still occur due to the existence of the blind spot or the inattention of the driver. In this thesis, a multipurpose digital signal processor (DSP) based blind spot warning system is implemented. The major research topic in this system is the blind spot vehicle detection warning systems by TI DM642 platform. The system has been applied in highways and urban roadways, and works well in sunny, cloudy, and rainy conditions for day and night. The day and night switch algorithm is used to determine time. The system performance can reach 20 frames per second with CIF format through the optimization technologies which are specially designed for the DSP platform. When the adjacent vehicle approaches to the host vehicle, the system will generate a warning signal to alert the drivers avoiding the collision.