Camera localization and mapping using delayed feature initialization and inverse depth parametrization

On-line motion estimation of a single camera moving freely through an environment is an interesting and challenging problem with a wide rage of applications. This estimate position from measurements of self-mapped natural visual features represents perhaps the more difficult class of the (SLAM) simultaneous localization and mapping problem. In this paper we show an approach for delayed features initialization for monocular SLAM based in inverse depth parameterization. The original undelayed aproach has shown good results for the monocular SLAM problem in a scheme of EKF for the estimation of the stochastic map and the pose of the camera. In our work we show that using a delayed feature initialization can improve the performance of the original method in some aspects.