Scribd Logo
Upload

Welcome to Scribd!

  • 24 views

    Presented By: Dewan Imdadul Islam

    Uploaded by

    dfmfhf fff
    This document presents a study on human activity recognition using deep learning. It discusses using deep neural networks like LSTM-RNNs to learn important features from unlabeled sensor data collected in smart home environments to perform activity recognition. The study aims to develop a novel technique to analyze, train and learn from unlabeled data so it can be classified using standard classifiers. It will use CASAS smart home datasets and apply LSTM to learn features and annotate activities without labels.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd

    Presented By: Dewan Imdadul Islam

    Uploaded by

    dfmfhf fff
    24 views13 pages
    This document presents a study on human activity recognition using deep learning. It discusses using deep neural networks like LSTM-RNNs to learn important features from unlabeled sensor data collected in smart home environments to perform activity recognition. The study aims to develop a novel technique to analyze, train and learn from unlabeled data so it can be classified using standard classifiers. It will use CASAS smart home datasets and apply LSTM to learn features and annotate activities without labels.

    Original Description:

    dgdjdgds

    Original Title

    gdhhdgdh

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document presents a study on human activity recognition using deep learning. It discusses using deep neural networks like LSTM-RNNs to learn important features from unlabeled sensor data collected in smart home environments to perform activity recognition. The study aims to develop a novel technique to analyze, train and learn from unlabeled data so it can be classified using standard classifiers. It will use CASAS smart home datasets and apply LSTM to learn features and annotate activities without labels.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    24 views13 pages

    Presented By: Dewan Imdadul Islam

    Uploaded by

    dfmfhf fff
    This document presents a study on human activity recognition using deep learning. It discusses using deep neural networks like LSTM-RNNs to learn important features from unlabeled sensor data collected in smart home environments to perform activity recognition. The study aims to develop a novel technique to analyze, train and learn from unlabeled data so it can be classified using standard classifiers. It will use CASAS smart home datasets and apply LSTM to learn features and annotate activities without labels.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    of 13

    STUDY ON HUMAN ACTIVITY

    RECOGNITION
    USING DEEP LEARNING

    Presented by :
    RABI GAYAN(180302015)
    Mtech IT 3rd Sem
    Guide: Dewan Imdadul Islam
    CONTENTS
     Objective
     Introduction
     Problems
     Approach to this problems
     Datasets
     Background of HAR
     Literature Survey
     Conclusion
     References
    OBJECTIVE
     To develop a novel technique for
    analyzing, training and subsequent
    learning of un-annotated data that can be
    accepted for classifications using standard
    low compute classifiers.
    INTRODUCTION
     Human Activity Recognition (HAR) has
    gained immense research attention over
    the past years due to its vast applications
    range including robotics, home
    automation, security surveillance, remote
    health-care and assisted and independent
    living for elderly.
     The problem of HAR can broadly be
    divided into vision-based and sensor-
    based recognition .
    Introduction contd.
     In vision-based HAR, video-cameras are
    mostly used to generate digitized visual data,
    which are then analyzed using computer
    vision techniques to detect the activities.
     In sensor-based HAR, small sensory
    devices, either in form of wearable
    devices/smart phones carried by the human
    or embedded into the environment, generate
    time-series data defining state change events
    or various physical parameters.
    Problems
     We observe that standard classifiers are
    often dependent on feature rich datasets
    and require extensive annotations to
    perform well.
     Massive amount of data gathered in smart
    homes through tape-on sensors often
    lack such rich features.
    Action Plan
     We will work with such dataset where a large number
    of sensor responses are recorded in a smart home
    scenario.
     Our primary motivation is to learn this data so as to
    extract the critical features and annotate each activity.
     We will employ a deep neural network based
    technique to learn the inherent features in such data
    and see if traditional classifiers would do well in
    recognizing the different activities.
     We will apply a variant of RNN called Long Short
    Term Memory (LSTM) to learn the important features
    from the data and subsequently perform annotations.
    Datasets
     In order to validate our work, we focus on
    data collected in a smart home environment.
    Towards that we focused on datasets
    generated by the Centre for Advanced
    Studies in Adaptive Systems(CASAS) [4] at
    the Washington State University’s School of
    Electrical Engineering and Computer
    Science Department.
     CASAS repository contains over 60 datasets
    from which we have chosen two of the
    datasets, namely the Aruba and the Tulum
    datasets.
    Dataset Used
    Background
    • This stage involves the extraction in short or long-time intervals,of sensor data.
    • From the data is forward, to a single device that has more powerful hardware.
    Data
    Acquisition

    • This stage includes an analysis and manipulation of data to guarantee there no invalid values
    that could compromise the learning stage.
    Data • This stage usually begins by removing noise values,proceeded by a segmantation of the
    dataset and then apply that into feature selection and extraction to get more valuable data.
    Manipulation

    • This stage comprehends a training mechanism to help build a classifier model with labelled
    or unlabelled data.
    Machine • Several different classification model can be used for determining the most accurate
    solution performance evaluation mechanisms are used.
    learning

    Figure 1: Basic steps in HAR system


    LITERATURE SURVEY
    Author Paper details JIST

    Paulo Barbosa Human Activity Recognition From this paper I understand


    Based on Transfer the data preparation design
    Learning,2018
    Jindong Wang, Yiqiang Chen, Deep Learning for Sensor-based From this I got the
    Shuji Hao, Xiaohui Peng, Lisha Activity Recognition: A understanding of deep learning
    Hu Survey,2019 .
    Sumalatha Ramachandran, An intelligent system to detect Here, I got the idea of
    Lakshmi Harika Palivela and C. human suspicious activity using Proposed framework
    Giridhara deep neural networks,2019 architecture.
    Schalk Wilhelm Pienaar, Reza Human Activity Recognition From this, I got the idea of
    Malekian, Senior Member Using LSTM-RNN Deep Neural Model Training
    Network Architecture,2019

    N. Yala, B. Fergani, and A. Towards improving feature For comparison with the other
    Fleury extraction and classification for state-of-the-art results, we will
    activity recognition on streaming compare the study done , over
    data,2017 the same dataset for feature
    extraction and classification
    CONCLUSION
     To the best of our knowledge, no existing
    work has performed annotations on un-
    annotated smart home data using
    RNN/LSTM, and our work is a first
    towards that direction.
     In our work, we will propose a novel
    technique for analyzing, training and
    subsequent learning of un-annotated data
    that can be used for classifications using
    standard low compute classifiers.
    References
    [1].
    PauloBarbosa,”HumanActivitiesRecognition:ATransferLearningApproach,
    ” FACULDADE DE ENGENHARIA DA UNIVERSIDADE DO PORTO
    2018.
    [2]. Jindong Wang, Yiqiang Chen, Shuji Hao, Xiaohui Peng, Lisha Hu,” Deep
    Learning for Sensor-based Activity Recognition: A Survey,”Elsevier Pattern
    Recognition Letters ,Volume 119, 1 March 2019.
    [3]. Sumalatha Ramachandran, Lakshmi Harika Palivela and C. Giridhara,” An
    intelligent system to detect human suspicious activity using deep neural
    network,” Journal of Intelligent & Fuzzy Systems, vol. 36, no. 5, pp. 4507-
    4518, 2019.
    [4]. Schalk Wilhelm Pienaar, Reza Malekian, Senior Member,” Human
    Activity Recognition Using LSTM-RNN Deep Neural Network
    Architecture,”IEEE,2019
    [5]. N. Yala, B. Fergani, and A. Fleury,” Towards improving feature extraction
    and classification for activity recognition on streaming data,” Journal of
    Ambient Intelligence and Humanized Computing,2016

    You might also like