Document-Level Relation Extraction using Deep Learning and Natural Language Processing

Human Action Recognition (HAR) is crucial for monitoring elderly individuals living alone, ensuring timely assistance during distress. This paper presents a HAR framework leveraging image processing and artificial intelligence to enhance performance across key benchmarks, including accuracy, computation speed, memory efficiency, and practical usability. By utilizing skeletal data, recurrent neural networks, and gait classification techniques, the approach achieves improved results on RGB video inputs. The proposed method reduces computation time significantly using Divide and Conquer, Sliding Window, and spatial aspects of Long Short-Term Memory (LSTM) architecture while maintaining low resource requirements for native device compatibility. Testing on the Fall Detection and NTU-RGB datasets demonstrates its effectiveness in handling real-time detection with reduced processing times compared to existing methods.