Human Activity Recognition Using Neural Network
0 likes66 views
The document discusses human activity recognition using neural networks. It presents a technique for identifying six common human activities (walking, sitting, standing, squat, punching, and head movement) using logistic regression, logistic regression with cross-validation, and convolutional neural networks. The technique uses a publicly available activity recognition database. It aims to build a precise computer application that can sense human body movements and extract context information in a user-friendly manner.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1565
Human Activity Recognition Using Neural Network
Swapnaja Jadhav1, Tejas Dalal2, Karan Pawar3, Atul Kulkarni4, Arif Manyar5
Department of Computer Engineering, DYPIT Pimpri, Pune
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Human activity recognition can be foundin a
variety of study domains, including medical organizations,
survey systems, security monitoring, and human computer
interface. This paper provides a viable technique to
identifying six common human- centered actions (walking,
sitting, standing, squat, punch and moving head) using
Logistic Regression, Logistic Regression CV, and the CNN
algorithm. A precise and pleasurable computer application
that sense human body movements to acquire context
information. As a repository, an activity recognitiondatabase
is regarded publicly available in this case.
Key Words: CNN Algorithm
1. INTRODUCTION
With the increasing rise in the need for security and
surveillance, particularly in crowded areas like airports,
shopping malls and social gatherings, the problemof human
detection and activity recognition has attainedimportance
in the vision community. Human activity recognition is an
important area ofcomputer visionresearchandapplications.
The goaloftheactivityrecognitionisanautomatedanalysisor
interpretation of ongoing events and their context from
video data. Its applications include surveillance systems,
patient monitoring systems, and a variety of systems that
involve interactions between persons andelectronic devices
such as human-computer interfaces. Most of these
applications require recognitionofhighlevelactivities,often
composedofmultiple simple actions of person’s lifestyle
2. MOTIVATION
The aims of Human-centered computing are to appreciate
individual activitieswiththeirsocial perspective.Importantly
the classification performance of the learned model using
new data set as compared tothe previous one, withreduced
set of features and improved results
3. LITURETURE SURVEY
Syed K. Bashar, Md Abdullah Al Fahim and Ki H. Chon”
Smartphone Based Human Activity RecognitionwithFeature
Selection and Dense Neural Network”[1]Human activity
recognition (HAR) has grown in prominence in recent years
due to the embedded sensors in smartphones, with
applications in healthcare, surveillance, human-device
interactions, and pattern identification. An activity- driven
hand-crafted neural network model forrecognizing human
activities is presented in this study. Selecting meaningful
features from the provided time and frequency domain
characteristics ismade easier with the help of an algorithm
developedusingneighborhoodcomponentanalysis.Afterward,
afour-layer deep neural network is utilized to classify the
input data into several groups. The fact that we were able to
outperform most previous models despite utilizing fewer
features shows just how importantfeatureselectionis.When
compared to existingstate-of-the-artmethods,ourproposed
model outperformed the majority of other methods while
using less features, demonstrating the critical nature of
feature selection. The model was evaluated using apublicly
available dataset of six daily activities fromthe UCI Health
Risk Assessment (HAR).
Asmita Nandy, JayitaSaha, Chandreyee Chowdhury,Kundan
P.D. Singh” Detailed Human Activity Recognition using
Wearable Sensor and Smartphones”[2] Human activity
detection is increasingly being employed in smart homes,
eldercare, and remotehealth monitoringandsurveillance.To
betterservethesegoals,actionssuchas sitting in a chair oron
the floor, taking a slow or brisk stroll, jogging witha weight,
and so on must berecognized comprehensively. Few studies
haveattempted to differentiate between hard activities(such
as walking while carrying a heavy burden) andtheir inverse
(walking), which is crucial for effectivehealth monitoring of
the elderly andpatientsrecoveringfromsurgery.Theusageof
wearable and smartphone-embedded sensors has been
presentedasa solution for this goal in this work. As a result,
the contribution of this work is to create an ensemble of
classifiers to providea framework forpreciseidentificationof
static and dynamic activities, as well as their intensive
equivalents. The ensembleisconfiguredsothattestinstances
are classified using weighted majority voting. The basis
classifiers'outputperformance forthe training datasetissent
into a neural network to determine their weights. We
determined that our work has a recognition accuracy of
greater than 94%.
Mohanad Babiker, Othman O. khalifa, Kyaw Kyaw Htike ,
Aisha Hassan, Muhamed Zaharadeen,” Automated Daily
Human Activity Recognition for Video Surveillance Using
Neural Network”[3] Due to consumer needs for security,
surveillance video systemsaregarneringgrowingattentionin
thefieldofcomputer vision. Observinghumanmovementand
predicting such senses of movement is promising. The need
arises to design a surveillance system capable of overcoming
the limitation of relying on humanresourcestocontinuously
watch, observe, and record normal and suspicious events](https://image.slidesharecdn.com/irjet-v9i5417-221005061943-06640845/85/Human-Activity-Recognition-Using-Neural-Network-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1566
without being distracted,aswellastofacilitatethecontrolofa
largesurveillance system network.Intelligenthumanactivity
system recognition is built in this work. Background
subtraction, linearization,andmorphological operationwere
among the digital image processing techniques used at each
stage of thesuggested system. The human activity features
database, which was taken from the frame sequences, was
used to build a robust neural network.
The activities model in the dataset was classified using a
multi-layer feed forward perceptron network. The
classification results show that all three stages of training,
testing, and validation were completed successfully. Finally,
these findings lead to a positiveperformance in the rate of
activity recognition.
Neslihan Kose, Mohammadreza Babaee, Gerhard Rigoll,”
multi-view human activity recognition using motion
frequency”[4]Spatiotemporaldifferencesinsubsequentvideo
framescan be used to address the problem of human activity
recognition. The useofmulti-viewmoviesisadvocatedinthis
research as anew method for recognizing humanactivities.A
naive background subtraction is conducted first, employing
frame differencing between neighboring frames of a movie.
Following that, each pixel'smotioninformationiscapturedin
binary, indicatingwhether or not motion exists in the frame.
The frequency of motion in each pixel throughout the clipis
calculated by a pixel wise sum of all the different images in a
view. These motion frequency features are used to evaluate
categorization performance. Increasing the number ofviews
used for feature extraction enhances performance, as
different views of an activity provide complimentary
information, according to our findings. Experiments on the
multi-view human action datasets i3DPost and INRIAXmas
Motion Acquisition Sequences (IXMAS) show significant
accuracy.
Soo Min Kwon, Song Yang, Jian Liu, Xin Yang, Wesam Saleh,
Shreya Patel, Christine Mathews, YingyingChen“Hands-Free
Human Activity Recognition Using Millimeter-Wave
Sensors”[5] We demonstrate a hands-free human activity
identification framework using millimeter-wave(mm Wave)
sensors in this demo. Our network, in comparison to other
systems, respects user privacy and can modify a human
skeleton conducting the activity.
Furthermore, we demonstrate that our network can be built
in a single architectureandfurthertunedforgreateraccuracy
than networks thatcan only produce solitary outcomes (i.e.
onlygetposeestimation or activity recognition).Toprovethe
practicality and durability of our model, we will present it in
various circumstances (i.e., in front of various backgrounds)
and effectively showcase the correctness of our network.
4. SYSTEM ARCHITECTURE
5. Objective
Design a simple, light weight, and accurate system
that can learn human activity with minimum user
interaction.
Compare and find a modelthatbestfitoursystemin
terms of accuracy and efficiency.
Reduce the labeling time and labor works using
active learning.
The aim of the system we propose is to
continuously track human activities.
Algorithm
⚫ Input – Well Annotated CSV Video Dataset
⚫ Output – Detecting and tracking human
Activity using video processing with better
efficiency
isolate gradient edges
10. Cropping out the Region of interest from step9
extraction
4. For i=1 to Pn do // Extracting Feature
5. For Layers(K): 1 -> K-1 do // here Kis 3
6. Obtain the edge feature map Fm
7. End For
8. End For
9. Apply equation on the feature map datasets to
1. Begin:
2. Pre-Process the CSV Video Dataset
3. Fed the video sample to CNN for feature](https://image.slidesharecdn.com/irjet-v9i5417-221005061943-06640845/85/Human-Activity-Recognition-Using-Neural-Network-2-320.jpg)
Human Activity Recognition Using Neural Network
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1565 Human Activity Recognition Using Neural Network Swapnaja Jadhav1, Tejas Dalal2, Karan Pawar3, Atul Kulkarni4, Arif Manyar5 Department of Computer Engineering, DYPIT Pimpri, Pune ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - Human activity recognition can be foundin a variety of study domains, including medical organizations, survey systems, security monitoring, and human computer interface. This paper provides a viable technique to identifying six common human- centered actions (walking, sitting, standing, squat, punch and moving head) using Logistic Regression, Logistic Regression CV, and the CNN algorithm. A precise and pleasurable computer application that sense human body movements to acquire context information. As a repository, an activity recognitiondatabase is regarded publicly available in this case. Key Words: CNN Algorithm 1. INTRODUCTION With the increasing rise in the need for security and surveillance, particularly in crowded areas like airports, shopping malls and social gatherings, the problemof human detection and activity recognition has attainedimportance in the vision community. Human activity recognition is an important area ofcomputer visionresearchandapplications. The goaloftheactivityrecognitionisanautomatedanalysisor interpretation of ongoing events and their context from video data. Its applications include surveillance systems, patient monitoring systems, and a variety of systems that involve interactions between persons andelectronic devices such as human-computer interfaces. Most of these applications require recognitionofhighlevelactivities,often composedofmultiple simple actions of person’s lifestyle 2. MOTIVATION The aims of Human-centered computing are to appreciate individual activitieswiththeirsocial perspective.Importantly the classification performance of the learned model using new data set as compared tothe previous one, withreduced set of features and improved results 3. LITURETURE SURVEY Syed K. Bashar, Md Abdullah Al Fahim and Ki H. Chon” Smartphone Based Human Activity RecognitionwithFeature Selection and Dense Neural Network”[1]Human activity recognition (HAR) has grown in prominence in recent years due to the embedded sensors in smartphones, with applications in healthcare, surveillance, human-device interactions, and pattern identification. An activity- driven hand-crafted neural network model forrecognizing human activities is presented in this study. Selecting meaningful features from the provided time and frequency domain characteristics ismade easier with the help of an algorithm developedusingneighborhoodcomponentanalysis.Afterward, afour-layer deep neural network is utilized to classify the input data into several groups. The fact that we were able to outperform most previous models despite utilizing fewer features shows just how importantfeatureselectionis.When compared to existingstate-of-the-artmethods,ourproposed model outperformed the majority of other methods while using less features, demonstrating the critical nature of feature selection. The model was evaluated using apublicly available dataset of six daily activities fromthe UCI Health Risk Assessment (HAR). Asmita Nandy, JayitaSaha, Chandreyee Chowdhury,Kundan P.D. Singh” Detailed Human Activity Recognition using Wearable Sensor and Smartphones”[2] Human activity detection is increasingly being employed in smart homes, eldercare, and remotehealth monitoringandsurveillance.To betterservethesegoals,actionssuchas sitting in a chair oron the floor, taking a slow or brisk stroll, jogging witha weight, and so on must berecognized comprehensively. Few studies haveattempted to differentiate between hard activities(such as walking while carrying a heavy burden) andtheir inverse (walking), which is crucial for effectivehealth monitoring of the elderly andpatientsrecoveringfromsurgery.Theusageof wearable and smartphone-embedded sensors has been presentedasa solution for this goal in this work. As a result, the contribution of this work is to create an ensemble of classifiers to providea framework forpreciseidentificationof static and dynamic activities, as well as their intensive equivalents. The ensembleisconfiguredsothattestinstances are classified using weighted majority voting. The basis classifiers'outputperformance forthe training datasetissent into a neural network to determine their weights. We determined that our work has a recognition accuracy of greater than 94%. Mohanad Babiker, Othman O. khalifa, Kyaw Kyaw Htike , Aisha Hassan, Muhamed Zaharadeen,” Automated Daily Human Activity Recognition for Video Surveillance Using Neural Network”[3] Due to consumer needs for security, surveillance video systemsaregarneringgrowingattentionin thefieldofcomputer vision. Observinghumanmovementand predicting such senses of movement is promising. The need arises to design a surveillance system capable of overcoming the limitation of relying on humanresourcestocontinuously watch, observe, and record normal and suspicious events
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1566 without being distracted,aswellastofacilitatethecontrolofa largesurveillance system network.Intelligenthumanactivity system recognition is built in this work. Background subtraction, linearization,andmorphological operationwere among the digital image processing techniques used at each stage of thesuggested system. The human activity features database, which was taken from the frame sequences, was used to build a robust neural network. The activities model in the dataset was classified using a multi-layer feed forward perceptron network. The classification results show that all three stages of training, testing, and validation were completed successfully. Finally, these findings lead to a positiveperformance in the rate of activity recognition. Neslihan Kose, Mohammadreza Babaee, Gerhard Rigoll,” multi-view human activity recognition using motion frequency”[4]Spatiotemporaldifferencesinsubsequentvideo framescan be used to address the problem of human activity recognition. The useofmulti-viewmoviesisadvocatedinthis research as anew method for recognizing humanactivities.A naive background subtraction is conducted first, employing frame differencing between neighboring frames of a movie. Following that, each pixel'smotioninformationiscapturedin binary, indicatingwhether or not motion exists in the frame. The frequency of motion in each pixel throughout the clipis calculated by a pixel wise sum of all the different images in a view. These motion frequency features are used to evaluate categorization performance. Increasing the number ofviews used for feature extraction enhances performance, as different views of an activity provide complimentary information, according to our findings. Experiments on the multi-view human action datasets i3DPost and INRIAXmas Motion Acquisition Sequences (IXMAS) show significant accuracy. Soo Min Kwon, Song Yang, Jian Liu, Xin Yang, Wesam Saleh, Shreya Patel, Christine Mathews, YingyingChen“Hands-Free Human Activity Recognition Using Millimeter-Wave Sensors”[5] We demonstrate a hands-free human activity identification framework using millimeter-wave(mm Wave) sensors in this demo. Our network, in comparison to other systems, respects user privacy and can modify a human skeleton conducting the activity. Furthermore, we demonstrate that our network can be built in a single architectureandfurthertunedforgreateraccuracy than networks thatcan only produce solitary outcomes (i.e. onlygetposeestimation or activity recognition).Toprovethe practicality and durability of our model, we will present it in various circumstances (i.e., in front of various backgrounds) and effectively showcase the correctness of our network. 4. SYSTEM ARCHITECTURE 5. Objective Design a simple, light weight, and accurate system that can learn human activity with minimum user interaction. Compare and find a modelthatbestfitoursystemin terms of accuracy and efficiency. Reduce the labeling time and labor works using active learning. The aim of the system we propose is to continuously track human activities. Algorithm ⚫ Input – Well Annotated CSV Video Dataset ⚫ Output – Detecting and tracking human Activity using video processing with better efficiency isolate gradient edges 10. Cropping out the Region of interest from step9 extraction 4. For i=1 to Pn do // Extracting Feature 5. For Layers(K): 1 -> K-1 do // here Kis 3 6. Obtain the edge feature map Fm 7. End For 8. End For 9. Apply equation on the feature map datasets to 1. Begin: 2. Pre-Process the CSV Video Dataset 3. Fed the video sample to CNN for feature
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1567 11. Apply Equation on detached ROI to representall lines 12. Applylinetracing method illustratedinGivenVideo section 13. End 6. Conclusion In this model we are abletorecognizedhumanactivitybased on their behavior. Our goal is to predictdifferent activities which human performed on daily basis. We have implemented this model using machinelearningalgorithms and CNN algorithm. Here in this model we have improved the efficiencyof model using the neurons which is best in prediction and classification. The main aim is to providethe optimization solution 8. References 1. Kwon Y, Kang K, Bae C, Chung HJ, Kim JH.Lifelog agent for human activity pattern analysis on health avatar platform. HealthcareInformatics Research. 2012 Jan; 2022(1):69- 75. DOI: 10.4258/hir.2019.20.1.69. 2. Iosifidis, A. Tefas and I. Pitas, ”Multi-view action recognition based on action volumes fuzzy distances and cluster discriminant analysis”, Signal Processing, vol. 93, no. 6, pp. 1445-1457, Jun. 2020. 3. Oikonomopoulos and M. Pantie, ”Human Activity Recognition UsingHierarchicallyMined Feature Constellations”,pp. 150-159, 2021. 4. M. Javan Roshtkhari and M. D. Levine, ”Humanactivity recognitioninvideosusingasingle example”,ImageVis. Comput., vol. 31, no. 11, pp. 864-876, Nov. 2019. 5. L. Chen, H. Wei and J. Ferryman, ”A surveyof human motion analysis usingdepthimagery”,PatternRecognit. Lett.,vol. 34, no.15, pp. 1995-2006, Nov. 2019. 6. W. Ong, L. Palafox and T. Koseki, ”Investigation of Feature Extraction for Unsupervised Learning in Human Activity Detection”, Bull. Networking Comput. Syst.Softw, vol. 2, no. 1, pp. 30-35, 2020. 7. O. D. Lara and M. A. Labrador, ”A Survey on Human Activity Recognition using Wearable Sensors”, IEEE Commun. Surv. Tutorials, vol. 15, no. 3, pp. 1192-1209, Jan. 2021. 8. A. Chaaraoui, J. R. Padilla-Lopez, P. Climent-P´erezand F. Fl ´ orez-Revuelta, ´”Evolutionary joint selection to improve human action recognition with RGBD devices”, Expert Syst. Appl., vol. 41, no. 3, pp. 786-794, Feb. 2020 9. N. Noorit and N. Suvonvorn, ”Human Activity Recognition from Basic Actions Using Finite State Machine”, Proceedings of the First International Conference on Advanced Data and Information Engineering(DaEng-2013),vol.285,pp.379-386,2021. 10. W. S. Lasecki, Y. C. Song, H. Kautz and J. P. Bigham, ”Real-time rowd labeling for deployable activity recognition”, Proceedings of the 2019 conference on Computer supported cooperative work-CSCW 13, pp. 1203, 2022