High accurate affordable car navigation using built-in sensory data and images acquired from a front view camera

Abstract

Nowadays cars are equipped with various built-in sensors such as speedometers, odometers, accelerometers, and gyros for safety and maintenance. Also, front view images can be economically acquired by a low-cost camera available in smartphones or black boxes. The combination of the built-in sensory data and the images can be an effective complement to a GPS based navigation. Therefore, we propose a car navigation framework to determine car position and attitude using the built-in sensory data such as a speed, angular rate and the images from a front view camera. The method consists of three steps, 1) dead reckoning using the velocity and yaw rate provided in real-time, 2) image georeferencing based on a sequential bundle adjustment using the dead reckoning results and 3) final estimation using a Kalman filter with the georeferencing results. The experimental results show that the proposed method can provide the positions with a reasonable accuracy level, which can be meaningful to complement a traditional GPS based navigation with a low cost.