为了准确提取人体动作特征,提出了一种新的基于二维Gabor滤波器的时空兴趣点检测器,该检测器对遮挡,光照变化以及镜头缩放等具有较强的鲁棒性.基于80面体模型在一定大小的时空邻城内提取精细的时空梯度信息进一步刻画人体动作在时空上的视觉特征.采用最大似然估计得到对每段动作视频的权重直方图估计,使算法更有效率且权重直方图描述特征更具区分度.将低层次的权重直方图特征和高层次的动作语义属性融合,采用隐支持向量机求解最终动作识别模型的局部最优解.在几种典型的数据库上对算法进行了验证,与现有方法相比较,识别率有了较大的提高.%A new spatio-temporal interest point detector using 2D Gabor filters is presented to extract features of human action accurately, which is robust to occlusion, lighting changes and camera zooming. A polyhedron with eighty faces model-based spatio-temporal gradient descriptor is created to illustrate the spatio-temporal visual features of human action. A weight histogram is adopted as the action representation based on maximum likelihood estimation making the algorithm more efficient while the weight histogram is more discriminative. The low-level weight histogram and high-level semantic attributes are fused together and the latent Support Vector Machine (SVM) is adopted to find the local optimum of the prediction model. Experiments using some kinds of typical datasets demonstrated that approach achieves a higher recognition rate compared to existing methods.
展开▼
