Key-frame extraction in Spacio-Temporal Neural Networks for Human-Action Classification
Saraswati Patil, Pradyumn Patil, Yashraj Pawar, Shruti Pisal, Vishwajeet Pawar · 2023
Video analysis especially related to human-actions is an crucial part of research done in domain of computer science and mathematics. Some of the areas are image and video processing, computer vision and Video understanding systems. With widespread use of mobile phone and other comb act video recording devices people capture many events that involve human as a center object This paper presents and tests various models that can extract both spacial feature's using CNN and temporal feature using RNN. By Developing a system that train a Deep learning model and optimized it to identify the type of human action/activity and ultimately classify them into categories/classes of different action performed by humans daily. A key-frame extraction approach has been implemented to increase the model's performance. Using an innovative algorithm, this technique selectively finds and classifies key-frames. By concentrating primarily on these key-frames, overall processing time is lowered without jeopardising the model's significant performance increases.