Complex Human Action Recognition Using a Hierarchical Feature Reduction and Deep Learning-Based Method.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Human action recognition is crucial in computer vision but often neglects temporal information and incurs high computational costs.
• Existing methods struggle to efficiently utilize temporal dynamics for accurate action prediction.
• Preprocessing phases in action recognition can be computationally intensive.

Purpose of the Study:

• To develop an efficient and accurate automated human action recognition system.
• To address the computational challenges in action recognition through optimized frame selection.
• To improve the utilization of temporal information for enhanced prediction accuracy.

Main Methods:

• Proposed a Hierarchical Feature Reduction & Deep Learning (HFR-DL) method.
• Automated selection of representative frames and extraction of key features.
• Employed background subtraction, HOG, CNN, LSTM, and Softmax-KNN classifier.

Main Results:

• Achieved significant improvements in accuracy and speed compared to eight state-of-the-art methods.
• Demonstrated the effectiveness of the HFR-DL model on the challenging UCF101 dataset.
• Validated the method's ability to handle complex, real-world human activities.

Conclusions:

• The HFR-DL method offers a computationally efficient and highly accurate solution for human action recognition.
• Automated representative frame selection and deep learning integration enhance performance.
• This approach advances the field of automated video analysis and action understanding.