Computation strategies for volume local binary patterns applied to action recognition

Abstract

Volume Local Binary Patterns are a well-known feature type to describe object characteristics in the spatiotemporal domain. Apart from the computation of a binary pattern further steps are required to create a discriminative feature. In this paper we propose different computation methods for Volume Local Binary Patterns. These methods are evaluated in detail and the best strategy is shown. A Random Forest is used to find discriminative patterns. The proposed methods are applied to the well-known and publicly available KTH dataset and Weizman dataset for singleview action recognition and to the IXMAS dataset for multiview action recognition. Furthermore, a comparison of the proposed framework to state-of-the-art methods is given.

Feature Fusion of Deep Spatial Features and Handcrafted Spatiotemporal Features for Human Action Recognition

“…Baumann et al [1] introduced Volume Local Binary Pattern (VLBP) for action recognition, which was first proposed in [5] for recognizing the facial expressions. VLBP describes the dynamic texture feature by comparing the intensity values of the neighboring voxels in the spatiotemporal domain.…”

“…Human action recognition is an attractive research topic in the area of computer vision due to its wide range of applications in video surveillance, sports video analysis, movie search, etc. Action recognition is challenging due to different viewpoint, occlusions, clothing, and the subject’s appearance, personal style, action length, and complex background motion [1,2,3,4]. Despite extensive research done on this topic, several issues still need to be resolved.…”

“…Feature extraction is an essential and core step for image and video analysis [1,2,4,5,6,7,8,9,10]. Basically, there are two different kinds of feature descriptor for the video representation, one is hand-crafted descriptors and another one is deep-learning based descriptors.…”

“…Previously, in [1,2,3,7,14] texture based spatiotemporal features are introduced to resolve the problem of action recognition due to their notable achievements and computational efficiency. However, these approaches are variants of Local Binary Pattern (LBP) [6] and suffer from similar kinds of issues that LBP faces, which include sensitivity to noise and limited capability to obtain more discriminative information.…”

A framework for cardio-pulmonary resuscitation (CPR) scene retrieval from medical simulation videos based on object and activity detection.

“…The neighborhood radii we used for this experiment was x radius = y radius = 1 and t radius = 2. We compute the uniform patterns for LBP [121], and construct feature vectors of 177 dimensions. Finally, we train a second linear SVM classifier (which we will refer to as SVM-L) with the new set of features.…”