Discounted Least-Squares Gearshift Detection Using Accelerometer Data

A gearshift detector based on accelerometer readings is derived, and its performance is investigated with the aid of experimental data. The employed methodology contains a low-order parametric model of the acceleration performed by the vehicle in combination with the minimization of a fading-memory least-squares criterion for parameter estimation and tracking. It is shown that the resulting filtering equations converge so that filtered accelerometer outputs (i.e., instantaneous acceleration and its rate of change) are obtained by linear time-invariant filtering of the raw accelerometer data, where the tradeoff between noise suppression and tracking ability is entirely determined by a single user-chosen variable. Two detection variables are formed, and they are combined to produce the detector output. Evaluation on experimental data illustrates the potential of the proposed method.