Dynamic Scene Recognition Using Spatiotemporal Based DLTP on Spark

Abstract

Scene recognition plays a significant role in the fields of pictorial information retrieval and scene understanding. However, the existing methods for scene recognition mostly consider only the images. Moreover, due to the rapid growth of multimedia data, there is a growing demand for distributed processing of large-scale video data. In this paper, we present a novel method for dynamic scene recognition from videos that considers spatiotemporal information in a distributed environment. Firstly, to obtain the dynamic information from the videos, we propose the directional local ternary pattern from three orthogonal planes, which provides valuable information about the nature of dynamic textures. Then, we utilize Apache Spark to conduct distributed computing for large-scale video data. Finally, we employ a convolutional neural network to classify the dynamic scenes. The experimental results show that our proposed approach outperforms most of the state-of-the-art methods in terms of accuracy. Moreover, we also experimentally demonstrate the scalability of the proposed method.