Classification of Vehicle Motion and Immobility States Based on MEMS Accelerometer Signal

The algorithms of immobilizer operation in automobile security systems assume the presence of an autonomous speed sensor in the immobilizer. Given strict price restrictions, the speed sensor is implemented on a MEMS accelerometer. Calculating the speed requires determining the moments of time of changing the state of immobility to motion, and vice versa. The spectral composition of vibration accelerations, which is individual for each car, requires an individual approach, so it is advisable to use machine classification methods to determine the change of states. The classification model is configured and operated based on a vector of features whose numerical value is different in the states of motion and rest. Such features are usually called informative. It is proposed to use ranking methods to select the three most informative features from the full set of features, which constitute the input vector of the supervised classification model. The results of feature ranking based on the results of test drives of two cars with two different accelerometer locations are presented. It is shown that, as applied to the problem under consideration, the Bhattacharyya and ROC methods give the best results.