Human Action Recognition using Salient Region Detection in Complex Scenes

Abstract

Although the methods based on spatio-temporal interest points have shown promising results for human action recognition, they are not robust in complex scenes especially background clutter, camera motion, occlusions and illumination variations. In this paper, we propose a novel method to classify human actions in complex scenes. We suppress the false detection interest points by detecting salient regions. Furthermore, we encode the features according to their spatio-temporal relationship. Our method is verified on two challenging databases (UCF sports and YouTube), and the experimental results demonstrate that our method achieves better results than previous methods in human action recognition. Keywords Human action recognition • Salient region detection • Complex scenes IntroductionAutomatically recognizing human actions is receiving increasing attention due to its wide range of applications such as video retrieval, human-computer interaction and activity monitoring. A large number of methods [1,2] for humane action recognition have been proposed, ranging from trajectory-based methods [3] and local descriptor-based methods [4] to attribute-based method [5,6].

Probability‐based method for boosting human action recognition using scene context

“…The experimental results are summarised in Table 4. The methods of Nguyen et al [31] and Zhang et al [32] achieved better accuracy. In these methods, salient region detection is imperative as a preprocessing step.…”

A Comprehensive Survey of Vision-Based Human Action Recognition Methods

“…Regardless of data type and computing method, the core aim is to extract robust human action features. Many action features have been proposed for RGB data, such as spatiotemporal volume-based features [10,43], spatiotemporal interesting point features [9,44], and joint trajectory features [12,13]. However, factors such as camera movement, occlusion, complex scenes, and the limitations of human detection and pose estimation methods limit the performance of human action representation and recognition based on handcrafted features.…”

RGB-D Data-based Action Recognition: A Review

“…Features are extracted to acquire (1) human movements, (2) spatial and temporal changes using Spatio-temporal volume-based methods [63,64], and Spatio-Temporal Interesting Points (STIP)-based methods [65,66], (3) the trajectory of skeleton joints [67,68] proposed a hybrid supervector method for action representation that achieved significant performance on common datasets. Nazir et al [76] have combined the 3D-Harris spatio-temporal features and 3D scale-invariance feature transform detection methods with visual histograms to represent actions.…”