本論文中利用Kinect感應器，來建立盲人行走輔助系統。系統提供四種模式，行走模式、探索模式、定位模式和過馬路模式。(1) 行走模式: 用感應器提供的深度資訊，藉由路面偵測演算法，並將深度轉為三維空間資訊，建立出環境資訊，偵測障礙物。(2) 探索模式：由感應器的色彩資訊，用SURF (Speeded-Up Robust Feature) 偵測特徵點和追蹤，使用最小平方誤差法計算出影像之間特徵點空間座標轉換矩陣，並建立平面地圖和地標特徵資訊。(3) 定位模式：從地圖中搜尋目前所在的位置座標和方向。(4) 過馬路模式：利用彩色影像來偵測行人穿越道的位置和路寬。 本論文提出用路面偵測演算法將使用者行走之路面切割出來，並即時的告知使用者障礙物的位置。提供室內地圖資料庫的建立、管理和定位的方式，讓視障人士可以知道目前在於室內的位置和方向。我們也提供行人穿越道偵測，協助視障人士安全的通過馬路。並且使用語音辨識來控制系統模式的切換，和直覺的語音提示告知使用者環境資訊。希望藉由此系統和搭配白手杖的使用，即能令目前導盲的輔具化被動為主動，讓視障人士的行動更自由。In this thesis, we use the Kinect sensor to establish a navigation system for the blind. The systems provide four modes, walk mode, exploring mode, positioning mode and cross road mode. (1) Walking mode: The system uses depth information and floor detection algorithms to build environmental information and detect obstacles. (2) Exploring modes: The system detects SURF (Speeded-Up Robust Feature) feature points and tracking feature points by the color information from the sensor. We calculate the coordinate transformation matrix between two images and create a map by the method of least-square. (3) Positioning mode: Search the current location coordinates and direction in the map. (4) Crossing road mode: The system uses color images to detect the crosswalk location and width of the road. In this thesis, we proposed that the floor detection algorithm to segment the floor region from depth information, and real-time tell the blind obstacle position. And we provide indoor map database to establish, manage and position method. The blind people can know the position and orientation in indoor. We also provide a crosswalk detection to help the blind safety cross road. And we use speech recognition to control the system mode. The system will use the voice to tell who the blind surrounding environment information. With the information about the environment the blind will have less fear in walking through unfamiliar environments via white canes.
