A Hidden Markov Model and Internet of Things Hybrid Based Smart Women Safety
A Hidden Markov Model and Internet of Things Hybrid Based Smart Women Safety
RESEARCH METHODOLOGY
In current global scenario, the
prime question that arise in a life of a free female civilian is her safety in
public places and transports. Statistical surveys have classified multiple
devilish approach [13] on female population globally. Studies have revealed
that 30.02% of the female population experienced physical assault and another
29% got raped (of which marital rape accounts for 6.67%). Approximately 26%
were kidnapped in 2017 of which around 12% got trafficked. There is a shocking
1.03% of abusive atrocities on children as well. A.HMM Based Activity
Recognitions Hidden Markov Models (HMMs) basically represent Bayesian networks
triggered to collect contextual information based on scanty approach data in
order to recognize possible threats or conducts that may turn abusive. HMMs in other words can be described as a doubly stochastic embedded network for identifying threats based on visual appearances and verbal conversations. The model designed needs decent input of continuous 3-minutes to shoot up to an accuracy of prediction of around 97%.
In other words, one sectional unit uses observations and past information (both direct and related recent past data) to update the hidden functional processing units in order to model the expected activity. Higher the early modelling database, faster is the prediction in real world scenario while the standard maintained is 1:12.5 million learning nodes; i.e. one artificial node represents 12.5 million similar characteristics. 11% of the data served as artifacts for exponential learning paradigm but that is the only option available to model along as the exposure to real life experimentation is a bit limited. For training purposes, we have used certain neurophysiological data acquisition techniques like Electroencephalogram (EEG) and Electrocardiogram (ECG). In addition, an IOT platform triggered plethora of sensors are correlated at the input end based on the psychological and emotional indices of the victim and possible suspect.
The total time of detailed observation on including both of the above cases are 3 minutes 21 seconds with an average time of response of around 25 seconds to 125 different types of mock drills of approaches performed. The drills were spatially sparsely distributed; some were similar adjacent, some were not whereas a type of discrete autonomous face and voice recognition is also used to smooth operate in case of non-human parallel approaches, for example, robotic theft, bombing etc. HMM inputs not only include contextual data but also the current and recent past states accompanied by pre-classified positive artifacts. Positive artifacts, the contradictory term basically refers to ‘apparently unreal or unbelievable but truth’ kind of data which portrays for the less common cases like child abuse. A priority vector and a transition matrix couples along with. The priority matrix is nothing but a probability vector which deals with early stages of learning and optimize them for highest predictions by fading down the later ones.
The hidden execution units are represented by the activities. For example, a guy approached a woman with friendly gesture but had lusty eyes. In such a situation, the training units will be provided a maximal data of positive qualities of the guy but depending on observations on retina the lustiness needs to get classified and generate warning to the woman. This is sorted out from the abundant positive data from the sensors simply by the priority matrix and the transition units by identifying the rate of changes in behaviour. As in last example, the priority upon the current position and nature of eyes is to be taken care of first and the fake politeness can be processed afterwards. The matrices generate state transition matrices and control probability of switching from one state to the possible adjacent one.
A contextual variable matrix trained the HMM module, where the earlier is fed by contextual variables processed as an input of a typical k-NN or similar neuro fuzzy classifier in order to achieve regularization. An occurrence test has been performed and interestingly, the scope of ruling off low occurrence average is achieved with only 0.02% data below the lower threshold and it served the scope of faster and earlier computation and identification of motives. In addition, 79% of irrational artifacts are also ruled off which further enhanced processing speed. B.IOT Based Computations and Controls Based on functionalities, the IOT model can be classified into four sublayers; Decision layer to choose relevant actions
Comments
Post a Comment