The Best Model of Convolutional Neural Networks Combined with LSTM for the Detection of Interpersonal Physical Violence in Videos

Hugo Calderon-Vilca, Kent Cuadros Ramos, Elmer Diaz Quiroz, Jorge Angeles Rojas, Rene Calderon Vilca, Alejandro Apaza Tarqui

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Citizen insecurity is directly related to interpersonal physical violence, there are algorithms that allow detecting violence in videos; therefore, it is necessary to know which is the best model for detecting violence. For this research, three convolutional neural network models were compared: Xception, InceptionV3 and VGG16 each together with a recurrent LSTM network, to find out which of the models is the best for the detection of interpersonal violence in videos. The three models were trained using the Real Life Violence Situations dataset, then violence and non-violence were classified, as a result, the InceptionV3 model is the best model, managing to classify with an accuracy of 94% compared to the VGG16 and Xception models, which obtained 88% and 93% respectively. Therefore, we recommend the InceptionV3 model for the detection of interpersonal physical violence in citizen security videos.

Original languageEnglish
Title of host publicationProceedings of the 29th Conference of Open Innovations Association FRUCT, FRUCT 2021
EditorsSergey Balandin, Yevgeni Koucheryavy, Tatiana Tyutina
PublisherIEEE Computer Society
Pages81-86
Number of pages6
ISBN (Electronic)9789526924458
DOIs
StatePublished - 12 May 2021
Event29th Conference of Open Innovations Association FRUCT, FRUCT 2021 - Virtual, Tampere, Finland
Duration: 12 May 202114 May 2021

Publication series

NameConference of Open Innovation Association, FRUCT
Volume2021-May
ISSN (Print)2305-7254

Conference

Conference29th Conference of Open Innovations Association FRUCT, FRUCT 2021
Country/TerritoryFinland
CityVirtual, Tampere
Period12/05/2114/05/21

Bibliographical note

Publisher Copyright:
© 2021 FRUCT.

Fingerprint

Dive into the research topics of 'The Best Model of Convolutional Neural Networks Combined with LSTM for the Detection of Interpersonal Physical Violence in Videos'. Together they form a unique fingerprint.

Cite this