Indoor Space Localization with a Mobile Phone by Object Tracking

Abstract

Recent researches on indoor localization have achieved a rapid progress, thanks to advances in mobile devices and networks. Related into simultaneous localization and mapping (SLAM) problems, several researchers apply different approaches, such as Wi-Fi, IMU sensors, and ultrasonic sensors. However, more intuitive and accessible system for indoor localization is required in order to achieve high-rate recognition of the current pose. In this paper, we propose the system that has the combination of visual data from a camera and inertial data from IMU sensors in indoor localization. Pre-learning of landmark images and setting up the database is the first part of our proposed localization method. Using TLD tracker and sensor data simultaneously, selected image areas are tracked and approximation of the device location can be extracted. EKF-SLAM, which uses extended Kalman filter to estimate device locations with sensor data, leads to real-scale estimation of the device approximately. From the characteristics of the camera and scale estimation from vision data and sensor data, camera poses are estimated and the landmark locations are matched. Even though abrupt changes of camera movement and angles cause errors on trajectories of the mobile device, camera pose estimation is successfully estimated, and the errors has a range from -0.1m to 0.4m, compared to the ground truth of the movement.