Dense 3D Mapping for Indoor Environment Based on Feature-point SLAM Method

Building accurate and dense 3D maps of indoor environments is a significant task for mobile robotics, with applications in navigation and semantic mapping. Although the current featurepoint SLAM algorithm is relatively mature, the existing SLAM method often can't work because it cannot extract enough feature information in the some scenes with changeable lighting, especially in indoor environment. In this paper, to deal with the problem of image illumination changing, we propose an adaptive threshold feature point extraction algorithm. With such a new solution, our system can work in environments with the varied illumination. A dense 3D mapping system exploits improved feature point method to estimate the pose of the camera, then the three-dimensional dense map is builded by the optimized camera pose. The surfel model and the deformation map are utilized to further fuse and optimize the point-cloud maps. Finally, the ideal 3D maps are obtained. We fully evaluate our method on the public data sets. Experiments show that the system has a good effect to build dense 3D map in the indoor environment.