Archives

Recognition of Violent Activity Response Using Machine Learning Methods with Wearable Sensors


Princy Randhawa, Vijay Shanthagiri and Ajay Kumar
Abstract

In India, even though it has superpower and development in the economy, but still there are many cities where crimes occur against the woman. The barbarity against a woman can be brought to an end if any system will be built which will act as a deterrence for the culprit. Violent activity recognition is important for woman safety especially with activities such as chain-snatching, kidnapping, molestation, rape, assault, etc. being very common. In this context, a smart- jacket using stretch sensors, pressure sensors and accelerometer data (9 DOF) are used for generating body-movements data and to record different kinds of movement patterns during normal activity as well as during an assault. Two active sequence patterns (normal action and violent attack) recorded for the classification using machine learning models: Multivariate regression analysis and Decision Trees. It has been proved and concluded in the paper that multivariate Regression models cannot be applied to classify the data when the dependent variable does not change with time continuously. An alternative model, Decision Tree is suggested by changing the feature of the data only and not changing the values of the data. A novel method is proposed for the classification by a depiction of the data in terms of the length of ASCII characters. Finally the results demonstrate that the Decision tree is better than Multivariate regression analysis for the application.

Volume 11 | 11-Special Issue

Pages: 592-601

DOI: 10.5373/JARDCS/V11SP11/20193071