基於時空興趣點的籃球運動行為辨識

Basketball Sport Action Recognition based on Spatio-Temporal Interest Points

李佳星(Jia-Hsing Li)

指導教授 ： 陳淑媛

元智大學/資訊學院/資訊工程學系/碩士(2013年)

https://doi.org/10.6838/YZU.2013.00146

摘要

視訊影片中人體行為辨識在電腦視覺領域是一個重要的問題，其可廣泛應用於視頻檢索、安全監控、人機互動，以及遠端照護等。本論文提出一基於時空興趣點(Spatio-Temporal Interest Points, STIP)演算法，辨識視訊影片中籃球運動行為。首先在籃球視訊影片中偵測時空興趣點，並將偵測到興趣點的特性，以方向梯度 (Histograms of Oriented Gradient, HOG)及光流直方圖(Histogram of Optical Flow, HOF)等特徵描述。接著針對測試籃球視訊影片的時空特徵點，就訓練集中籃球視訊影片，找尋最鄰近(最相似)的時空特徵點，並以相似度高於門檻值之對應方視為有效對應。最後定義對應數的多寡為兩籃球視訊影片的相似度，以進行後續的籃球運動行為分類。為達以上目的，本研究使用最近鄰居(Nearest Neighbor, NN)找尋最相似的時空特徵點，同時利用時間與空間上的一致性，修正有效特徵對應，並以最近鄰居分類法(Nearest Neighbor rule, NN rule)進行籃球運動行為辨識。本論文所採用的視訊影片是由公開的Action Similarity LAbeliNg (ASLAN) 資料庫挑選出來之籃球運動視訊影片，實驗結果證實所提出方法確實可行。

關鍵字

行為辨識 ； 籃球視訊影片 ； 時空興趣點 ； 時間一致性 ； 空間一致性 ； 方向梯度直方圖 ； 光流直方圖

並列摘要

Identification of human behavior in video clips is an important issue in the field of computer vision, which can be widely used in video retrieval, security surveillance, human-computer interaction, remote care and so on. This study presents a classification algorithm based on Spatio-Temporal Interest Points (STIP) to identify basketball behavior in video clips. First, the STIPs are detected in basketball video and their characteristics are described in terms of the direction of the gradient, Histograms of Oriented Gradient (HOG), and optical flow, Histogram of Optical Flow (HOF).
Matching scores are then calculated using constraints on matching pairs and consistency. More specifically, the match (or correspondence) of a STIP point in the test video is defined as the nearest neighbor (NN) of all the STIP points from the model video based on the Euclidean distance of the STIP descriptor vectors in terms of HOG and HOF. A matching is regarded valid only if the matching distance is less a threshold. Matching score is then computed according to the number of valid matching pairs and the degree of consistency among valid matching. The matching scores are used to classify basketball behavior.
The proposed method uses publicly library Action Similarity LAbeliNg (ASLAN) and selects video clips of basketball from the database to prove the feasibility of our method. Experimental results confirmed that the proposed method is promising.

並列關鍵字

Action recognition ； Basketball video ； Spatio-temporal interest points ； Matching consistency ； Histogram of oriented gradient ； Histogram of optical flow

參考文獻

[1] J. K. Aggarwal and M. S. Ryoo, “Human activity analysis: a review,” ACM Computing Surveys, vol. 43, no. 3, 2011.

[2] A. Bobick and J. Davis, “The recognition of human movement using temporal templates,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 23, no. 3, pp. 257–267, 2001.

[4] C. Rao and M. Shah, “View-invariance in action recognition,” in Proc. Int. Conf. Computer Vision and Pattern Recognition, 2001.

[5] Y. Sheikh M. Sheikh, and M. Shah, “Exploring the space of a human action,” in Proc. Int. Conf. International Conference Computer Vision, 2005.

[6] A. Yilmaz and M. Shah, “Recognizing human actions in videos acquired by uncalibrated moving cameras,” in Proc. Int. Conf. International Conference Computer Vision, 2005.