請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/55934
標題: | 應用影像式線雷射障礙物偵測系統於防跌智慧鞋之研究 Intelligent fall prevention shoes using camera based line-laser obstacle detection system |
作者: | Chi-Yun Yang 楊啟雲 |
指導教授: | 施文彬(Wen-Pin Shih) |
關鍵字: | 跌倒,障礙物偵測,線雷射,homography,disparity,影像分割, Obstacle detection,line-laser,homography,disparity,image segmentation, |
出版年 : | 2014 |
學位: | 碩士 |
摘要: | 在現今的高齡化社會中,老人跌倒是一個非常嚴重的社會問題。根據美國疾病控制與預防中心的資料,在意外事故死亡案例中,有一半是因為跌倒而死亡,另外在美國,每年因跌倒所產生的醫療花費高達三百億美元。而老人因各種疾病以及身體能力、反應速度的下降,又屬於特別容易跌倒的族群。因此設計一個自動化的系統來預防跌倒的發生或在跌倒傷害發生後產生一個警示的作用,就成為當代科技中一個刻不容緩的議題。
科學家和工程師已經致力於發展防跌倒的科技很多年了,包含安裝於穿戴式裝置上或是安裝於環境中的裝置,這些方法各有不同的應用以及優缺點。而我們在此論文中提出了一個影像式線雷射障礙物偵測系統,在老人踢到危險障礙物跌倒之前,能夠發出警訊提醒老人,達到防止跌倒、遏止跌倒對老人身心靈的損害,以及為社會節省醫療成本的效果。由於老人大多較常待在家中或照護中心,跌倒的案例也以室內較多,因此本系統也是針對室內環境而設計。 影像式線雷射障礙物偵測系統將雷射光水平投射在距離地面一個固定高度的平面上,並利用和此平面呈一夾角的相機來取出線雷射特徵。在這樣的硬體設置下,當系統前方的障礙物深度產生變化時,其在相機影像中的特徵會呈現上下偏移的現象,系統即利用此一特性,來偵測障礙物的距離。 在實際應用中,系統會在腳踩在地面上或是腳抬到最高點的時候被觸發,以偵測的地面上的障礙物或是高度較高的障礙物。本系統利用動態影像中各個影格強度值的差異,來判斷系統被驅動的時間。當系統被驅動後,擷取的影像會先通過中值濾波器濾除雜訊,接著線雷射特徵會被擷取出來。在擷取線雷射特徵的同在擷取線雷射特徵的同時,系統運用一個強度閥值來將線雷射特徵和背景做分離,得到一個更佳的線雷射特徵擷取結果。 我們利用同一個線雷射特徵線段中每個像素和前一個像素的位置偏移關係,設定閥值對線雷射特徵上的每個像素做歸類,達到影像分割的效果,然後利用histogram上的切線斜率來對屬於障礙物的類別和屬於雜訊的類別做分離,就可以判斷是否有障礙物。最後利用homography矩陣來校正影像中座標和線雷射偵測平面上點的座標關係,即可計算出障礙物的深度和寬度,判斷障礙物的危險性,進而發出警訊提醒老人預防其跌倒。 在系統的演算法中,系統觸發時間的決定、線雷射特徵擷取、影像分割、histogram去除雜訊等都有參數或閥值需要被決定。因此本論文中也實作了一些相關的實驗,進而提出建議的參數和閥值。並且我們也對homography矩陣計算障礙物深度進行驗證,其誤差被證明能夠滿足障礙物偵測應用中準確度的需求。 關鍵字: 跌倒、障礙物偵測、線雷射、homography、disparity、影像分割 In aging society nowadays, fall down of elders is a serious social problem. According to the data of CDC (Centers for Disease Control and Prevention), in death cases of injuries, half death cases result from unintentional falls. Besides, the medical costs with regard to falls reach up to 30 billion US dollars per year in the US. Also, because the decline of physical and reaction abilities of elders, elders tend to fall especially. As a result, designing an automated system to prevent falls from happening or send alarms for help after falls happened becomes a critical problem in the development of technologies recent years. Scientists and Engineers have dedicated to develop technologies to prevent falls, including wearable devices and ambience devices, these devices has various application and their pros and cons. In this thesis, a camera-based line-laser obstacle detection system is proposed. Before elders kick on obstacles to fall, this system can emit alarm messages to mention them and then reach the goal to prevent falls, to prevent the damage on the bodies and the negative impacts to the mind of elders, finally to save medical costs of the whole society. Because elders spend most of their time at home or in nursing homes, most fall cases occur indoors. Therefore, the proposed line-laser obstacle detection system is designed mainly for indoor applications. The camera based line-laser obstacle detection system cast its line-laser on a horizontal plane which has a specific height to the ground. And then the system uses a camera which has an angle to the plane to extract the line-laser pattern. Under this system configuration, when the depth of obstacles in front of the system changes, the extracted line-laser pattern in the image will have disparities. This system uses the characteristic of disparities to detect the distance of obstacles from the system. In real implementation, system will be triggered when users step on the ground or when users raise the foot at max height during a gait cycle. This enables the system to detect obstacles on the ground or higher obstacles. In our method, differences between frames are utilized to determine the trigger time of our line-laser obstacle detection system. After the system was triggered, the captured frame will be filtered with a median filter first, and then the line-laser pattern on the image will be extracted. At the same time, an intensity threshold is used to separate the line-laser pattern from background in order to get a better line-laser pattern extraction result. On the extracted line-laser pattern, the deviation between the present pixel and the last pixel is utilized to classify each pixel to the cluster which denotes a specific obstacle, and a deviation threshold is set to implement image segmentation. And then a tangent line with a specific threshold on histogram is used to distinguish clusters which denote obstacles from clusters which come from noise. Last, we implement homography matrix to transform the coordinate on the image into the coordinate on the detection plane. Therefore, the depth and the width of each obstacle can be calculated to judge how dangerous this obstacle is. If any obstacle is too dangerous, an alarm message will be emitted to prevent elders from falling. Among algorithms of line-laser obstacle detection system, some parameters and thresholds of the system need to be determined. As a result, some experiments are designed to suggest suitable parameters and thresholds for system operation. Furthermore, a validation of homography which computes the distance between obstacles and the system is carried out. The result shows that the error of homography distance computation can fulfill the need of obstacle detection. Keywords: Obstacle detection, line-laser, homography, disparity, image segmentation |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/55934 |
全文授權: | 有償授權 |
顯示於系所單位: | 機械工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-103-1.pdf 目前未授權公開取用 | 2.39 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。