輔助視障者於枕木紋行人穿越道線行走安全系統之研製

Abstract

視障者一般在行經交通道路口時，要如何獲取行人穿越道上的資訊來達成安全行走的目的是很重要的問題。所以本研究所研製的辨識系統可以用來輔助視障者準確地偵測出枕木紋行人穿越道線。針對動態的影像進行辨識處理。透過使用可攜式的攝影設備，代替視障者的雙眼以偵測出枕木紋行人穿越道線，輔助其能安全地穿越枕木紋行人穿越道路口。讓視障者在穿越馬路時取得前方行人穿越道的資訊，可以有更安全的行走資訊。 本研究主要是以影像處理的方式來達成，整個演算法架構為：載入影像分序圖、彩圖轉灰階、灰階轉黑白、圖形濾波、最後標出圖框。偵測枕木紋行人穿越道線部分，本論文採用了連通物件標籤法(Connected Component Labeling)及面積濾波(Size Filter)，而這兩種方法應用上容易因影像雜訊太多而產生誤差，為了解決這個問題，本論文嘗試以中值濾波器(Median Filter)與形態學濾波器(Morphological Filter)來抑制影像雜訊之影響。 實驗內容是對於不同天候條件下和多種道路場景來作偵測，由實驗結果得知，皆能有效的偵測並判別結果。整體辨識率可達99%，處理速度約為每張0.03秒。When the visually impaired walk across the roads, how do they obtain the necessary information to pass through safely? This is a serious issue concerning people’s life. Therefore, the auxiliary visual identification system developed by the Institute can be very helpful to assist the visually impaired to detect and notify where the pedestrian crossing is. By wearing portable video gears, the visually impaired can easily acquire effective and accurate information with coordination when a pedestrian’s crossing lying ahead. Our system could provide the visually impaired with more secure and trusted notification of the environment. This study is based on image processing approach to reach its goal. The procedures of this research include loading serial images, gray scale converting, binarization, graphic filtering and targeting on the objects in the long run. The paper uses Connected Component Labeling Algorithm and Size Filter theory together to accomplish the task of targeting on the zebra crossing. It is found that by applying these two methods in the process would cause certain inaccuracies due to much noise in an image. In order to solve this problem, this paper attempts to adopt Median Filter approach and Morphological Filter theory to suppress and minimize the effect of image noise. Experimental contents are likely to do with different weather conditions and a variety of road scenes. After realistic involving with the system, we gladly obtain the satisfying results of overall recognition rate of 99% and the processing speed is about 0.03 seconds per frame.