Localization/Tracking Algorithm based on Sensor Fusion in Vehicular Networks

Abstract

With growing autonomous vehicle industry each year, the usage of the autonomous vehicle continues to expand creating demands in many areas. However, safety has been an issue yet to be solved due to the difficulties regarding various environments that have to be considered. Hence, vehicles have been equipped with several sensors such as radar, lidar, camera, and ultrasonic sensors to predict random events around the vehicle. Whereas the filter helps to reduce the difference in the performance of each sensor resulting in decision errors that are crucial to the system invoking the possibility of collision.

Localization is one of the most important issues in the autonomous vehicle industry. Until now the camera has not been implemented for localization, although it is capable of providing important features in real-time tracking. This thesis aims to propose a sensor fusion algorithm using distance information measured from a monocular camera. Other sensors include inertial measurement unit (IMU) and ultra-wideband (UWB) communication modules. Both Extended Kalman filter (EKF) and unscented information filter (UIF) are implemented and compared for the fusing IMU, UWB communication module and camera sensors. The stability of the algorithm has been evaluated in a 2D vehicle driven scenario. Simulation results demonstrate the potential of the autonomous vehicle navigation system with the proposed algorithm, and experimental results show validation of the vehicle position estimation. Performance improvement has been shown as both EKF and UIF systems are provided with multiple sensor data including camera, UWB, and IMU.