Visual SLAM

Temporally Scalable Visual SLAM

Unlike previous visual SLAM approaches that maintain static keyframes, our approach uses new measurements to continually improve the map, yet achieves efficiency by avoiding adding redundant frames and not using marginalization to reduce the graph. We use an online binocular visual SLAM system that uses place recognition for both robustness and multi-session operation. Additionally, our system automatically detects elevator rides based on accelerometer data and illustrates the capability to map a large area over extended period of time.