Publication:

以等位函式法改善動作捕捉系統之效能

桌球教練工作者時常遇到的問題，則是選手動作調節適應的問題，有時選手因為動作上極小的角度差異造成練習狀況不佳，而往往這些極小差異用肉眼並無法判斷出來，也因此造成訓練上之盲點。

一般動作捕捉系統其成效往往取決於系統中所用之影像分割方法之好壞。本研究目的,嘗試以數學中的等位函式法,應用在動作捕捉系統中的影像分割技術上,並且和一般動作捕捉系統中常用的影像分割方法進行比較。本研究之資料收集﹐採用台南桌球館錄製之蔣澎龍選手練習之影片﹐經輸入影像後﹐應用一般影像分割方法(canny法)與等位涵式法進行手臂擊球之瞬間角度的影像分析並加以比較。並以(分割結果之角度/實際角度)作為單位﹐結果發現等位函式法所分割之結果在10行中有5行誤差較肯尼法小。如加上比值為1的3行,則等位函式法所分割之結果在10行中有8行誤差較肯尼法相對為小。此研究結果證實運用此法所處理之運動影像分割之結果較一般常見的影像分割方法來的精確。本研究之成果可作為未來在改良動作捕捉系統中有關於三維影像重建,影像追蹤及影像比對之基礎研究。

Abstract

The problem which table tennis coach worker often meets, is it player's movements that regulate the problem met. Sometimes, a small angle difference on movements causes that player exercise state is not very good. The small difference that cause blind spot of training usually can not be found by eyes. Generally the Motion Capture System its result usually depends on the system in the good and bad of the images segmentation used. In this paper, try with the Level Set Method that apply in the images segmentation in mathematics in the Motion Capture System. And compare with images segmentation that the traditional Motion Capture System is convenient. Our research use the Chiang Peng-Lung’s exercise image which is recorded by table tennis hall of Tainan. After inputting images, we use traditional images segmentation method (canny method) and Level Set Method to analyze the images that arm hits producing moment angle. After analyzing, we compare this result of images segmentation. We use (angle of segmentation result / real angle) as the unit. The result of the Level Set Method segmentation at 10 items has 5 items that error are smaller than the result of the canny method. If we add 3 items that ratio is 1, then the result of the Level Set Method segmentation at 10 items has 8 items that error are smaller than the result of the canny method. The segmentation results certificate makes use of the more traditional precision of the result of the images segmentation that this method handles. The achievement of this research can rebuild the three-dimensional image in improving the Motion Capture System as the future, the image is followed the trail of and the image, than to its basic research.