Actions as Space-Time Shapes

摘要: Human action in video sequences can be seen as silhouettes of a moving torso and protruding limbs undergoing articulated motion. We regard human actions three-dimensional shapes induced by the space-time volume. adopt recent approach [14] for analyzing 2D generalize it to deal with volumetric shapes. Our method utilizes properties solution Poisson equation extract features such local saliency, dynamics, shape structure, orientation. show that these are useful recognition, detection, clustering. The is fast, does not require alignment, applicable (but limited to) many scenarios where background known. Moreover, we demonstrate robustness our partial occlusions, nonrigid deformations, significant changes scale viewpoint, high irregularities performance an action, low-quality video.