作者: Yu Kong , Xiaoqin Zhang , Qingdi Wei , Weiming Hu , Yunde Jia
DOI: 10.1109/ICPR.2008.4761001
关键词:
摘要: Group action recognition in soccer videos is a challenging problem due to the difficulties of group representation and camera motion estimation. This paper presents novel approach for recognizing with moving camera. In our approach, ego-motion estimated by Kanade-Lucas-Tomasi feature sets on successive frames. The optical flow then computed compensated Due inaccurate estimation, can not reflect accurate objects. this paper, we propose new descriptor which treats as spatial patterns extracts global from noisy flow. latent-dynamic conditional random field model employed recognize action. Experimental results show that promising.