A fusion algorithm for building three-dimensional maps

Recently various algorithms for building of three-dimensional maps of indoor environments have been proposed. In this work we use a Kinect camera that captures RGB images along with depth information for building three-dimensional dense maps of indoor environments. Commonly mapping systems consist of three components; that is, first, spatial alignment of consecutive data frames; second, detection of loop-closures, and finally, globally consistent alignment of the data sequence. It is known that three-dimensional point clouds are well suited for frame-to-frame alignment and for three-dimensional dense reconstruction without the use of valuable visual RGB information. A new fusion algorithm combining visual features and depth information for loop-closure detection followed by pose optimization to build global consistent maps is proposed. The performance of the proposed system in real indoor environments is presented and discussed.