Three-dimensional Reconstruction of the Semantic Terrain of Rescue Equipment Walking

Abstract

Most traditional multi-legged rescue equipment uses semi-autonomous control, which is not flexible and maneuverable. In addition, conditions such as long-distance communication and environmental particularity make it difficult to achieve real-time control of rescue equipment. Therefore, more and more attention has been paid to the autonomous behavior ability of rescue equipment. Environmental awareness is a prerequisite for rescue equipment to walk independently. Only with extremely strong environmental awareness ability can rescue equipment correctly plan the movement route and avoid unnecessary danger. Aiming at the situation of complex terrain in the wild, a method for constructing semantic terrain is proposed. The triangulated network is used to connect the scattered point clouds into regular surfaces, and then the basic features of the surface are used to classify the terrain to forma semantic terrain recognizable by rescue equipment. We performed 3D reconstruction on the point cloud data set that we built. Experiments show that this method is fast, and the construction of million points only takes 2.5s, which can basically achieve real-time performance and can meet the effect of real-time autonomous movement of rescue equipment.