Building 3D map for localization using human action

This paper proposes a method of building a 3D map composed of static objects for localization of mobile robots using recognition of human actions. In the monitoring systems with fixed cameras and also moving cameras, high-precision localization of moving cameras is required for measuring human actions in detail and precisely. A 3D map of the environment composed of static objects such as walls, desks and shelves is useful for high-precision localization of moving cameras in 3D space, because salient features in the environment are used as landmarks. Furthermore, the 3D map is required for generation of monitoring motion of the moving cameras and recognition of human actions in the relation of the environmental objects. In order to generate a 3D map automatically by measuring 3D points in an environment, points on static objects have to be selected from measured points, because there are dynamic objects such as chairs, books, and other mobile things in the environment. The proposed method generates a 3D map for localization by selecting points on static objects from measured 3D points based on recognition of human actions. Experimental results show feasibility of our proposed method.