Learning Car Speed Using Inertial Sensors for Dead Reckoning Navigation

A deep neural network (DNN) is trained to estimate the speed of a car driving in an urban area using as input stream measurements from low-cost six-axis inertial measurement unit (IMU). Three hours data was collected by through city Ashdod, Israel equipped with global navigation satellite system (GNSS) real time kinematic (RTK) positioning device and synchronized IMU. Ground truth labels for were calculated position obtained at high rate 50 Hz. DNN architecture long short-term memory layers proposed enable high-frequency estimation that accounts previous inputs history nonlinear relation between speed, acceleration angular velocity. simplified aided dead reckoning localization scheme formulated assess model which provides pseudo-measurement. The shown substantially improve accuracy during 4 minutes drive without use GNSS updates.