DEEP LEARNING FOR HUMAN ACTION RECOGNITION

Abstract

The abstract describes work being done to create a model that can identify and categorize human motions
including running, jogging, walking, clapping, hand-waving, and boxing from a collection of movies.The project's main goal
is to develop a system that can analyze video footage and classify human behavior. Activities like running, jogging, and
others fall under this category.The project makes use of a dataset including footage of people carrying out certain tasks. Each
video has an attached label that describes the task performed in the clip. The primary goal of the model is to understand how
the behaviors seen in the movies map to their respective labels. Without depending on written explanations, it must be able
to interpret and identify the activities seen in the video. Once the model has learnt this association, it should be able to
predict the action (label) for an unexplored input video. This indicates that the model can learn to detect new activities
outside of its training data.In spite of having descriptions of the activities, the abstract admits that the model still has to learn
to discern between different human behaviors based on the visual input alone. This implies that the model should be able to
represent the complexity and subtlety of various activities based purely on picture analysis. The abstract also hints that the
project's approaches might be used for more than simply human action recognition. Possible applications include learning
patterns in human movement to direct different tasks and active object tracking (identifying things or individuals in CCTV
video). In conclusion, this abstract provides an overview of work in progress toward developing a machine learning model
that can extract semantic information from video data to identify human activities. Possible uses for the concept include
directing human activities based on previously observed patterns of behavior and monitoring objects in the environment.