運動影片中人物動作的自動偵測與分析：以立定跳遠為例

Automatic Detection and Analysis of Human Motion in Sports Video: A Case Study in Standing Long Jump

謝勝文(Sheng-Wen Hsieh)

指導教授 ： 許輝煌

淡江大學/工學院/資訊工程學系碩士班/碩士(2006年)

https://doi.org/10.6846/TKU.2006.00108

摘要

關於各種的運動分析的研究越來越多，如分析網球、籃球、高爾夫球等，都有人在研究分析資料以做為訓練之用，而我們這篇論文所要做的是分析小朋友跳遠的影片，希望可以藉由電腦的幫助，將分析影片的難度降低，更進而讓系統可以提供受測者意見，使之可以了解自己動作的問題而不需要專業的老師在側協助。 在這邊我們的論文將分為兩大部分來介紹我們所使用的方法，第一部分是將人從影片中擷取出來，而在這部分中，我們將分為五個細項來處理，第一個項目是還原背景，第二個項目是取出前景物件，第三個項目是去除雜訊點，第四個項目是修補輪廓破損，第五個項目是去除陰影。 第二部分是預測骨架，在這邊我們是透過基因演算法的方法來預測骨架，透過基因演算法的特性“適者生存，不適者淘汰＂的規則來預測骨架，因此在產生數代的骨架組合後，我們將可以從這些組合中得到一組最適合輪廓的骨架。 而當預測出來的骨架都能對應到影片中受測者的輪廓時，我們就可以得到一連串的動作分解，進而可以做到分析受測者的動作是否正確，將來更可以將評分的動作也加入到系統中，使這系統可以給受測者一點回饋，使受測者知道說自己的動作還有那邊可以加強，如此一來就可以讓受測者在沒有專業的體育老師在旁邊時，依然可以做自我的練習。

關鍵字

動作分析 ； 人的偵測與追蹤 ； 姿勢預估 ； 基因演算法

並列摘要

Analyses and researches toward various sport activities are broadly put in use, and these sports involve tennis, basketball, and golf, whose data are collected and analyzed as the training data. In this thesis, we examine the video of children doing long jumps, aiming to reduce the difficulty in analyzing video sequences of children who are doing long jumps with the assistance of the computer. Furthermore, the developed system can also give
comments on the testers about the problems of their movements and gestures without a professional instructor.
The proposed method consists of two parts. The first one is to retrieve human objects from the video and it can be divided into five steps -background restoration, foreground objects retrieval, noise reduction, breakage repainting, and shadow removal.
The second part is to do stick model prediction, which utilizes the characteristics of survival of the fittest in Genetic Algorithm (GA). As a result, we can obtain a stick model that is fittest to the silhouette after several generations of evolutions.
While the predicted a stick model correspond to the silhouette of testers in the video, a succession of action decomposition is then taken place to analyze the correctness of the movements. Also, a scoring function can be integrated into the system so that the testers can receive feedbacks from the system. They can understand which parts of movements should be improved and the system makes it possible that testers can do self practice even without a professional physical education instructor.

並列關鍵字

Motion analysis ； human detection and tracking ； shadow removal ； pose estimation ； genetic algorithms

參考文獻

[1] Hwann-Tzong Chen, Horng-Horng Lin, Tyng-Luh Liu,"Multi-object tracking using dynamical graph matching", Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Volume: 2 , 8-14 Dec. 2001, Pages:II-210 - II-217 vol.2

[3] Fengliang Xu and Kikuo Fujimura, “Human Detection Using Depth and Gray Images,” Proc. IEEE Conference on Advanced Video and Signal Based Surveillance, 115 – 121, July 21-22, 2003.

[7] Wen-Nung Lie and Ruey-Lung Chen, “Tracking moving objects in MPEG-compressed videos,” IEEE International Conference on Multimedia and Expo, 2001. ICME 2001. 22-25 Aug. 2001 Page(s):965 – 968

[8] Thakoor, N and Gao, J. “Automatic Video Object Shape Extraction and its Classification with Camera in Motion,” IEEE International Conference on Image Processing, 2005. ICIP 2005. Volume 3, 11-14 Sept. 2005 Page(s):437 - 440

[10] Yilmaz, A, Xin Li, and Shah, M, “Contour-based object tracking with occlusion handling in video acquired using mobile cameras,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume 26, Issue 11, Nov. 2004 Page(s):1531 - 1536