運用電腦視覺技術完成室內盲人避障及導引之系統

An Indoor Obstacle Avoidance and Guidance System for the Blinds Using Computer Vision Techniques

王尋玉(Xun-Yu Wang)

指導教授 ： 蘇義明

共同指導教授 ： 林文寬(Wen-Kuan Lin)

義守大學/電機資訊學院/電子工程學系/碩士(2016年)

摘要

為了實現室內盲人避障及導引系統之建立，本系統首先擷取視訊中的圖像，使用均值偏移(Mean-Shift)群聚演算法作為圖像前處理，再與洪水填充(Flood Fill)演算法進行顏色分割，接著擷取出地面區域顏色，作為篩選其他非地面區域的條件。當獲得的地面區域後，從此區域擷取出左右邊緣的兩條直線，再計算圖像中兩條直線的交點作為消失點(Vanishing Point)特性，並依消失點找出這兩條直線所夾的角平分線作為導引的方向。當地面上有障礙物如行人或物體時，將使用一個具有近距離限制的路線定位方法找出規避的方向。由於地面區域的輪廓會受到障礙物的影響，因此進行障礙物的偵測並找出障礙物的區域範圍，去除圖像中非地面顏色得到初歩地面，並找出障礙物輪廓影響的興趣區域，接著使用前景特性所產生的顯著顏色，校正此興趣區域得到更精確的障礙物位置，從地面區域的輪廓所佔據的範圍，取得的中心點座標可規劃出一個導引路線。而不同障礙物的影響下產生的地面區域範圍，也會產生不同的地面導引路線，因此達到動態的規避功能。最後，實驗結果中擷取的地面區域與物體偵測都有近88%的成功率，亦可在近距離的障礙物之下完成盲人避障和導引等功能。

關鍵字

Mean-Shift群聚演算法 ； Flood Fill演算法 ； 物體偵測 ； 室內導引

並列摘要

In indoor environment color are basic on building fabric, and its distributed also related on building construct, which make floor and non-floor region can easily understand. The floor region has walkable condition, and space extend conception, that can seem as the guidance direction, which make the floor region become the walkable path.
In this thesis, we present an obstacle avoidance and guidance system for the blind in the indoor environment. In the system, we capture images from the video, use Mean-Shift algorithm for image preprocessing, and the Flood-Fill algorithm for color segmentation. After the color segmentation, the floor color is segmented to filtering non-floor colors, which extract two lines from floor edge on right and left side of the corridor to calculate their intersection point regarded as a vanishing point. A guidance line is obtained from the position of the blind to the vanishing point. Furthermore, there are some pedestrians or objects existed on the floor. The avoidance direction is decided to use a near distance decision to locate path. Due to the non-floor region and obstacle to change the contour of floor region, we can use this specific floor contour to obtain first coordinate position, which can locate region of obstacle in the image. This region will have foreground saliency color, which we can use to adjust region position by compare with background color. The continuous center points of floor region can connect to be a guidance line to achieve the object avoidance. Finally, the floor region and obstacle detection has the successful rate of the 88% in the experiment. It also has good avoidance and guidance performance in the near-distance obstacles.

並列關鍵字

Mean-Shift Algorithm ； Flood-Fill Algorithm ； Obstacle Detection ； Indoor Guidance

參考文獻

[31]鄭綱,"行走區域標示及危險狀況判別之盲人輔助系統"台灣國立交通大學,電子工程學系電子研究所碩士班,碩士學位資格論文,2011,九月

[2]Bruno Andò, Salvatore Graziani “Multisensor strategies to assist blind people: a clear-path indicator”, IEEE Transactions on Instrumentation and Measurement,On April 2009, Page(s)2488-2494.

[3]D. Conrad and G. N. DeSouza, ” Homography-Based Ground Plane Detection for Mobile Robot Navigation Using a Modified EM Algorithm ” Robotics and Automation (ICRA), 2010 IEEE International Conference , On May 2010 Page(s)910-915.

[4]J. Josè, J. M. H. du Buf and J. M. F. Rodrigues " Visual Navigation for the Blind Path and Obstacle Detection" Vision Laboratory, Institute for Systems and Robotics (ISR/IST),University of the Algarve (FCT and ISE), 8005-139 Faro, Portugal.

[6]Loay Ismail, Lubna Eliyan, Razan Younes, and Reem Ahmed, “Monocular Vision-Based Collision Avoidance System using Shadow Detection ” GCC Conference and Exhibition (GCC),Nov 2013 7th IEEE, On Nov 2013, Page(s)589-594.