Action Sensing Based on Computed Volume

Abstract

The local features based on interest point have achieved much success in action sensing recently. The interest point is not only limited to 2D space, but also extended to 3D space. We apply the 3D interest point to action sensing. A classic method to use 3D interest point is through creating a feature using histogram vector based on bag of words; some better methods take advantage of the position of each interest point besides the local feature; however, it’s difficult to position these points due to the complexity of an action. We propose a simple method to position each interest point, and create a new feature for action sensing. Evaluation of the approach on two sets of videos suggests its effectiveness.