Application of Machine Learning Methods to Determine the Movement of a Car Using a MEMS Accelerometer Signal

Currently, automobile security systems use remotely controlled electronic relays that break the vehicle's electrical circuits, the operating algorithms of which require autonomous determination of the vehicle's motion/rest states. Due to price restrictions on the implementation of relays, the role of motion sensors is played by MEMS accelerometers. When installed in the engine compartment, the accelerometer measures the acceleration of movement against the background of vibrations from a running engine and from uneven road surfaces, and the spectral composition of the signal depends on the make of a particular car. It is proposed to individually configure the parameters of the machine learning algorithm for each car, which classifies the states of motion/rest. When setting up the algorithm, statistical processing of acceleration in the motion/rest sections is first carried out, and a set of informative parameters are selected that are used in the formation of the training sample. The results of the classification algorithm based on the support vector machine method are presented at various sampling rates of the acceleration signal and sliding window sizes for extracting informative parameters.