利用影像處理分析及比對人體肢體運動的研究

Analysis and comparison of body limb motion by image processing

指導教授 : 黃榮堂、陳正光   研究生 : 陳信生  機電整合研究所 91年


摘要

  本研究將視覺影像辨識應用在人體運動比對方面,建立一套以一般視訊攝影機為影像回饋裝置的運動比對系統,可廣泛應用於各類的運動家教系統。為了本系統日後可普及化,而採用一般平價視訊攝影機,因此在辨識的難度上大為提高,所以我們簡化前處理部份,以特殊的方式代替複雜又缺乏準確性的前處理部分。此特殊的方式是使用一套特定的服裝,以服裝上的特殊顏色來區分每一個肢體,當作辨識的目標,此方式可容易且準確的定出肢體的位置。
本研究使用Hough Transform將肢體的角度估計出,包括雙手及雙腳的角度,再利用象限的規則計算出正確的角度。在身體方面:利用身體上的色彩比例關係,計算出轉身的角度。透過使用這些角度與標準資料做比對,即可知道使用者的動作正確與否。本研究以主從模仿動作的影像做比對學習實驗,結果證實本研究的可行性。

ABSTRACT

  This study attempts to compare different body postures. A motion comparing system that uses the web CCD camera as a feedback device has been developed. The system can be widely applied to many kinds of exercise teaching system.
For the purpose of popularization, we use the low cost web camera as the device to capture the images. However, the image quality captured by the web camera is not good and may make pattern recognition more difficult. In order to enhance the recognition capability of this low-end camera and avoid complex preprocessing algorithms, the user has to put on special clothes which are designed to have different colors for every limb to enhance recognition. In this way, the position of limb can be easily and rapidly found.
This research uses Hough Transform to calculate the angle of each limb, including arms and legs. Then, use the rule of the quadrant to calculate the corresponding real angle. The trunk part of the clothes has two different colors which can be used to calculate the turning angle of body by the area of these two colors. We use these angles to compare with the standard data provided by a model, and we can know whether the user’s motion is correct. In the experiments, we use two similar series image of motion for comparison. The results show the proposed method is feasible.