HUMAN ACTIVITY RECOGNITION SYSTEM
1Aniket Paulzagde, 2Anurag Nimje,
3Prof. Swati Shamkuwar
Department of Information Technology.
G.H. Raisoni College of Engineering Nagpur, Maharashtra
Abstract
In this modifying and developing world there are many vision based intelligent system which attracts the attention of many researchers to work on it, modify and develop it to the next level.
- Human Activity Recognition System (HAR) has multifaceted application all over the world due to usage of acquisition devices such as smartphones, video cameras, which are used for human activity data recordings
- The human Activity which are fetched or stored in the form of videos are based on the analyzation of video frames by the usage of computers to impulsively find the human activity without manual computations.
- Most of all networks for recognition of task use convolution neural network with the help of convolution and pooling layers followed by a few numbers of fully connected layers which identifies similar patterns in the frames in the intervals of video to recognize the activity by providing 79-80% accuracy based on the activity is done in the video from the human being.
- The main aim of Human Activity recognition system is to identify and understand the actions of people in video and export corresponding tags which can be achieved through the machine learning algorithm which identifies the human activity in video or an image .
The aim of Human Activity Recognition System (HARS) is to identify activities from the set of datasets on the movement of human being to the environment conditions. There are many applications like video surveillance, healthcare, and interaction of computer and human, are based on the HAR research. This study highlights improvements in advanced activity detection techniques, especially in activity representation and classification techniques.
We discuss various widely used approaches for classification. We categorize existing literatures with a thorough taxonomy that includes representation and classification methods, as well as the datasets they worked, with the goal of giving the brief to these algorithm and the ways to compare them.[1]
Keywords: Classification Model, Frame Rate Algorithm, Pooling Frames Kinetic-400 Dataset
