Comparative Analysis of Action Recognition Models: Investigating CNN with LSTM for Optimal Performance

The study investigates the various action detection techniques in computer vision with an emphasis on identifying human actions including jogging, sleeping, drinking, and falling. The usage of two neural networks based on LSTM (Long Short-Term Memory) units and a combined CNN and LSTM approach are among the study’s key findings. In areas like augmented reality, animation, games, and robots, the significance of action recognition is emphasized. LSTM networks and convolutional neural networks are frequently used in the approach for action recognition. The study also illuminates the difficulties in recognizing human activity in video sequences or still photos, such as background noise, partial occlusion, and variations in lighting and appearance.