Using a Selective Ensemble Support Vector Machine to Fuse Multimodal Features for Human Action Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Traditional human action recognition (HAR) relies on RGB video, which has limitations in accuracy and robustness.
• RGB-Depth (RGB-D) data, captured by sensors like Microsoft Kinect, offers improved performance for HAR.
• There is a growing need for robust and accurate HAR systems in various industries.

Purpose of the Study:

• To propose a novel multimodal feature fusion method for human action recognition.
• To develop a selective ensemble support vector machine (SESVM) classification model for enhanced HAR.
• To demonstrate the efficiency and effectiveness of the proposed method compared to existing algorithms.

Main Methods:

• Fusion of improved HOG features from RGB data, depth motion map-based local binary pattern (DMM-LBP) features, and hybrid joint features (HJF).
• Implementation of a frame-based selective ensemble support vector machine (SESVM) classification model.
• Integration of selective ensemble strategy with SVM base classifier selection to maximize classifier diversity.

Main Results:

• The proposed method achieves high accuracy and robustness in human action recognition tasks.
• Experimental results on public datasets show the method is simple, fast, and efficient.
• The SESVM model effectively integrates multimodal features, outperforming other action recognition algorithms.

Conclusions:

• The proposed multimodal feature fusion approach combined with SESVM significantly improves human action recognition.
• The method offers a practical and efficient solution for real-world HAR applications.
• Further research can explore advanced feature extraction and ensemble strategies for HAR.