Real-time EKF SLAM System Using Confidence Map Of Depth Information

Simultaneous localization and mapping (SLAM) is a technique to computationally construct or update a map of an unknown environment while simultaneously tracking a system’s location within the environment. In particular, vision-based SLAM employs a visual camera as a primary sensor. This system attempts to perform simultaneous tracking and feature mapping without additional sensing units, such as a laser sensor, gyroscope, and accelerometers. Stereo-based SLAM employs a stereo rig as the sensing unit, in which a pair of cameras is equipped so that it provides depth information acquired from binocular disparity. In this paper, we introduce a visual SLAM system using a confidence map of the depth estimates of feature points. The confidence map is used as a reliability measure of depth estimates by stereo vision. The experimental results show that the proposed system can obtain stable performance in a dynamic environment.