摘要
提出了一种单幅图像上轮廓信息的获取方法,在一段时间内发生的动作可以通过采样获得多幅图像来表示该动作特征。这些图像可以看成是轮廓形状变化序列,通过比较不同动作的轮廓形状序列,使得动作能够较容易地识别出来。实验证明,轮廓信息能够较好地反映对象的形状,采用形状序列和DTW相结合的方法能够提高动作识别的准确性。
The object shape from a single image is represented by using silhouette information. Several images compose a sequence of the shape to represent an activity for a limited time. An activity can be recognized compared with different shape sequences. Experimental results show that the silhouette information can characterize the object shape. Thus the recognition accuracy can be improved by combining the shape sequence with dynamic time warping（DTW）.
出处
《数据采集与处理》
CSCD
北大核心
2009年第5期615-620,共6页
Journal of Data Acquisition and Processing
关键词
动作识别
动态时间规整
形状序列
activity recognition
dynamic time warping（DTW）
shape sequence
