Three and Two Dimensions Data Fusion Based Panoramic Environment Perception for Space Modeling

This paper presents a systematic approach for 3-dimensional (3d) scene modeling based on Time-of-Flight (TOF) technology. Firstly the panoramic point cloud of scene model is built. To achieve that, point cloud data captured in different visual angles are transformed and registered into a panoramic 3d coordinate system. Based on the matching characteristics of feature points in various point clouds, these data can be synthesized into a fusion point cloud. Then the synthetic cloud is divided into several groups of point cloud fragments based on the depth scaling parameter. Secondly the 2-dimensional (2d) grayscale images which match the original point cloud data are collected and processed to generate a panoramic 2d image. Through Speeded-Up Robust Features (SURF) method, the similar feature pixels of different images are selected and disposed to fuse panoramic image. A novel 2d fusion image optimization method based on the pixel weighted theory is proposed in this paper. Finally the point cloud fragments segmented in the first step are classified according to different depth scaling values. The specific fragment is projected in 2d pixel space, and be modified as mask pattern. Therefor the objects of interest in the environment can be perceived by applying corresponding mark patterns to process panoramic image. The methodology of this environment perception approach can be the reference to TOF technology related 3d obstacles identification research.