UNSUPERVISED LEARNING-BASED DETECTION METHOD, AND DRIVER PROFILE-BASED VEHICLE THEFT DETECTION DEVICE AND METHOD USING SAME

摘要

An unsupervised learning-based detection method, according to one technical aspect of the present invention, is an unsupervised learning-based detection method using a supervised and learned model, comprising the steps of: generating, from driving data, a plurality of first matrix data; generating encoding information by encoding the plurality of matrix data using a convolutional neural network; deriving correlations between variables according to a time series by modeling time series characteristics of the encoding information using a long short-term memory (LSTM) network; re-implementing a plurality of second matrix data by means of a deconvolutional operation on the correlations between the variables according to the time series; and determining, on the basis of a difference between the plurality of first matrix data and the plurality of second matrix data, whether current driving corresponds to a previously supervised and learned driver profile.