Multi-View Human Action Recognition Using Skeleton Based-FineKNN with Extraneous Frame Scrapping Technique.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Human Action Recognition (HAR) is a key research area in computer vision.
• Existing HAR algorithms like 3D CNNs and CNN-LSTM are computationally intensive due to complex models and extensive weight adjustments.
• These complex models necessitate high-end hardware for real-time performance.

Purpose of the Study:

• To propose a novel, computationally efficient HAR system.
• To overcome the dimensionality challenges associated with complex HAR models.
• To develop a real-time HAR solution using simplified features and classification.

Main Methods:

• An extraneous frame scrapping technique was developed.
• 2D skeleton features were extracted using the OpenPose technique.
• A Fine-KNN classifier was employed for action recognition.
• The proposed method was evaluated on the Multi-Camera Action Dataset (MCAD) and INRIA Xmas Motion Acquisition Sequences (IXMAS) dataset.

Main Results:

• The proposed OpenPose-FineKNN with Extraneous Frame Scrapping Technique achieved 89.75% accuracy on the MCAD dataset.
• The system attained 90.97% accuracy on the IXMAS dataset.
• The proposed method demonstrated superior performance compared to existing state-of-the-art techniques like CNN-LSTM.

Conclusions:

• The proposed extraneous frame scrapping technique combined with 2D skeleton features and Fine-KNN offers an effective solution for HAR.
• This approach successfully addresses the complexity and computational demands of traditional HAR algorithms.
• The method shows significant potential for efficient and accurate real-time human action recognition.