機車駕駛安全輔助系統

Abstract

本論文之主要目的在發展一機車駕駛安全輔助系統。由於台灣歷年來因機車事故而死亡和受傷的人數一直沒有降低的傾向，而交通事故中因駕駛者未注意車前狀態或未保持行車間距等肇事原因佔肇事比例的三成左右。因此本系統將著重於偵測機車本車與前方障礙物間距，並利用當前時速計算應保持的安全間距，將二者進行比較就可以發出訊號提醒駕駛者是否處於安全狀態。本研究使用智慧型手機做為研究設備，主要因為智慧型手機已具備多種感應器且擁有運算能力可視為行動電腦。智慧型手機上的攝影機可用來捕捉前方景物，而其GPS可用來取得機車當前位置和使用方向感應器取得攝影機拍攝的角度。利用上述的資訊就可以將攝影機拍攝的影像透過轉換矩陣轉成鳥瞰影像，以便計算本車與障礙物之間的距離。 本研究分兩部份，一部份為事前處理，即計算各種拍攝角度的轉換矩陣並建立轉換矩陣資料庫；另一部份為系統設計與運作，即將拍攝影像轉換鳥瞰影像，並偵測障礙物和計算本車與障礙物之間的間距，以及計算應該保持的安全間距。在事前處理的部份，本研究設計自動產生轉換矩陣的方法，僅需將拍攝影像和拍攝時的方向感應器感應到的角度資訊輸入本系統，就可以自動計算轉換矩陣。轉換矩陣的計算方式為透過已知的四組角點座標，將角點座標帶入聯立方程式並解之就可以得到轉換矩陣。而系統設計與運作的部份，利用輸入影像中的縱向直線數量和初始化的角度資訊，就可擷取合適的轉換矩陣將輸入影像轉換成鳥瞰影像。另外，透過自動調整門檻值的方式，系統可擷取出鳥瞰影像中機車前方的障礙物，來計算本車與前方障礙物之間的間距。同時，利用GPS的定位資訊來計算當前車速，就可以換算應保持的安全間距。最後比較兩間距，就可以決定是否發出訊號讓駕駛者做出相對應的動作。 根據實驗結果，本系統具即時性及可用性。值得一提的是，本系統不僅僅可供機車駕駛者使用，也可供汽車駕駛使用。The main objective of this thesis is to develop a driver safety assistance system for motorcycles. In Taiwan, the number of casualties of motorcycle accidents did not decrease recently. According to car accident statistics, distracting driving and tailgating caused about 30 percent of motorcycle accidents. The proposed driver safety assistance system for motorcycles is used to detect the safety distance between the owner’s vehicle and obstacle ahead, and gives the driver a warning signal if necessary. In this study, a smartphone is used as a research tool, while modern smartphones are sophisticated computing platforms with complex sensors. The phone camera can capture and record videos; the orientation sensor is used to obtain the pitch and roll angles, and GPS can help to get the precise coordinates of the motorcycles. The study can be divided into two parts. Firstly, the transformation matrices of different camera angles are calculated and stored in a transformation matrix database. The transformation matrix database is constructed automatically. All the information the system needs includes input images, pitch angles, and roll angles. The system obtains the transformation matrix by solving the simultaneous equations about the corner information detected from the input image. Secondly, a kernel system is developed, including the following stages: turn each image into a top view, detect the distance between the owner’s vehicle and obstacle, and calculate the safe distance. The system can create the top view images using a suitable transformation matrix in the database. Moreover, the system can detect the obstacles in the top view images. Then the distance between owner’s vehicle and obstacles can be calculated. At the same time, the GPS embedded in the smartphone can help to obtain the speed of the vehicles to calculate the safe distance. Finally, the system will give the driver suitable warning signal after comparing two distances. The experimental results show that the proposed real-time system is robust and efficient. Moreover, the system can work very well not only on motorcycles but also on cars.