Action Recognition Using Nonnegative Action Component Representation and Sparse Basis Selection.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Human action recognition in videos is crucial for applications like surveillance and human-computer interaction.
• Traditional methods often struggle with the semantic gap between low-level features and high-level actions.

Purpose of the Study:

• To propose a novel approach for human action recognition using high-level action units.
• To develop a sparse model that effectively represents and recognizes human actions in videos.

Main Methods:

• A context-aware spatial-temporal descriptor (locally weighted word context) was developed to enhance feature discriminability.
• Action units were learned using graph regularized nonnegative matrix factorization, creating a part-based representation.
• A sparse model utilizing a joint l2,1-norm was employed to refine action unit representation and reduce noise.

Main Results:

• The proposed method demonstrated improved discriminability of spatial-temporal descriptors.
• The learned action units effectively bridged the semantic gap in action recognition.
• The sparse model successfully preserved representative units and suppressed noise, leading to accurate action recognition.

Conclusions:

• The developed approach using high-level action units and a sparse model significantly enhances human action recognition.
• The method provides a robust and effective way to represent complex human actions in videos.
• Experimental results on public datasets validate the effectiveness and superiority of the proposed approach.