發生道路交通事故的部分因素是因為在車輛行進中，駕駛沒有注意到障礙物而造成碰撞意外。其中原因包含車體結構與後照鏡角度造成的盲點區域，使得駕駛無法充分了解車輛週遭環境導致人員傷亡與車輛損壞。為了避免因為盲點區域而造成的交通意外，並提高停車與低速會車時的安全性。本論文提出一套全周俯瞰監視與障礙物偵測系統。整個系統共包含兩大部份：一是全周俯瞰監視用於輔助駕駛監視車輛周遭的狀況，二是主動偵測車輛周遭的障礙物並提醒駕駛注意。 全周俯瞰監視系統在車輛四周架設四台廣角相機以拍攝車輛週遭影像，經過離線計算相機內部參數、處理扭曲校正、暗角消除。再利用大型校正版，根據特徵點對應求得四周俯瞰影像的相對關係，將俯瞰影像快速對位為一張俯視車輛週遭的全周俯瞰影像，最後將各項參數建立一張查找表。在線上處理階段根據查找表資訊，內插產生即時的全周俯瞰影像，並根據重疊區域的亮度差調整整張影像的亮度均勻性。 俯瞰影像式障礙物偵測系統針對不同的環境特性，選擇適合的偵測方式。在路面紋理複雜時，以單張俯瞰影像估計光流，估計車體自我運動向量，藉由光流濾除及群聚後，擷取障礙物。在單純紋理的路面則以靜態顏色資訊為基礎，計算路面顏色的分佈情形，濾除路面與地面標誌後擷取障礙物區域並提醒駕駛注意。 全周俯瞰監視系統可在影像的解析度為720×480的情況下，於Intel@ Core?2 Duo 2.93GHz及4.00GB RAM的個人電腦上執行可達每秒24張的處理速度。而障礙物偵測程序在同樣的硬體設備上可達平均每秒16張，平均障礙物偵測正確率可達88%。 A lot of traffic accidents are caused by driver's incomplete understanding of the whole vehicle surroundings. To reduce the accidents caused by collision with surrounding obstacles, we mount four wide-angle cameras at the front, rear, and both sides of the vehicle to capture consecutive images then we present a real-time surrounding top-view monitor and obstacle detection system for slow driving and parking assistance. In offline steps of surrounding top-view monitor system, we first calibrate camera intrinsic parameters, distortion of lens, and vignetting effects of four wide-angle cameras. Then we calibrate the geometric relationships (extrinsic parameters) of four cameras using a proposed multi-camera calibration method. Third, we calculate the feathering weights of pixels on overlapped image areas to produce a seamless surrounding top-view image. At last, we build look-up tables for the mapping between the captured images and the surrounding synthesized image to speed up the processing. In online procedure, the proposed system interpolates and generates the surrounding synthesized image by those look-up tables directly. In obstacle detection system, we utilize different algorithms for driving environments of different texture complexity. If texture of road surface is complicated, we can generally detect enough features for estimating the optical flow from captured images. After estimating ego-motion of vehicle, we can distinct the non-ground features and ground-features. Obstacle detection is performed based on static color information if the texture of road surface is simple, and no features for detection was found, then we can use color of road region to separate obstacles from road. In our experiment, the detection accuracy is about 88%.
