Vision Based Detection and Analysis of Human Activities
Abhiram Ravipati, Rakesh Krishna Kondamuri, A. Mary Posonia, J. Albert Mayan
Abstract
These days, it's not uncommon to see video cameras installed to keep an eye on pedestrians and motorists alike in a variety of public spaces. The proliferation of camera footage necessitates the creation of some means of deducing the nature of the activity captured on film. Due to the widespread availability of acquisition devices like handsets & camcorders, HAR may be used in a wide variety of contexts. The proliferation of electronic gadgets and software has been matched by a revolution in data extraction made possible by breakthroughs in AI. Difficulties such backdrop congestion, partly blockage, variations in size, perspective, illumination, & look, make it difficult to recognize human actions in video frame or still photographs. Multimodal recognition is necessary in several fields, such as CCTV, human computer interaction for characterizing human behavior. This study surveys the most current and cutting-edge findings from the study of how to categorize human actions. This research proposes a taxonomy for studying human activity and explore the benefits and drawbacks of this diverse set of approaches. Disabled people's daily routines, including resting, moving, traveling down & up the stairs, speaking, & laying, have been frequently tracked using cellphones. Human motion detection often makes use of well-established ML & DL techniques include CNN, & LSTM Network.