A Variable Size Block Matching Based Descriptor for Human Action Recognition

Abstract

Translate Article

In this work we present an ongoing research which intends to integrate the Block Matching technique into the Human Action Recognition field. We propose a method to create a video descriptor composed of three main steps: Computation of displacement vectors between frames with VSBMA, histogram representation of these vectors, and orientation tensor generation from the histogram. It is a promising approach for human action recognition since it is a simple method with low computational cost which achieves comparable results with state-of-the-art techniques.