Sensor fusion-based robust navigation in the GPS-denied environment using monocular camera and in-vehicle sensors

Abstract

Autonomous driving technology has become an interesting issue for the automobile industry with advances in localization, obstacle detection, and the Internet of Things (IoT) technology. In order to make a safe self-driving, the vehicle must be able to estimate the precise pose of the vehicle body on the road. GPS is a sensor used generally to measure the absolute pose, but the bunch of buildings, tunnel or any indoor place can block the satellite microwave signals. Recently, many research validated the pose estimation method using other sensors such as wheel encoders, camera, and IMU for GPS-denied environments. In this paper, we proposed a compensation method of a monocular camera and in-vehicle sensors for robust localization. The method overcomes a scale ambiguity issue of monocular visual odometry and angular velocity measurement error of wheel odometry by using in-vehicle sensors. Furthermore, we proposed the fusion method between a monocular camera and in-vehicle sensors, which based on Extended Kalman Filter. We validate the proposed algorithm on the SKT real road dataset, which includes loop road, highway, and sloped road.