Scene Estimation for Making Active Decisions

Abstract

Scene estimation problem has been examined along with the structural description of differents objects in the scene. The data structure of scenes is established through a tree structure. This uses depths of different objects in the scene. Depth map is computed using Horn’s reflectance map and Shah’s linearization technique of the reflectance map. The center of mass is considered for the localized rectangle of each object to define the hierarchy of objects present in the scene. The localization of objects is based on least square estimation technique on YOLO output images. The localization method is fully free from any kind of thresholds and is a new concept. Comparison with other methods shows our method is more effective. Finding the structural description is also a new scheme. The nearness of objects at different levels of the hierarchical tree for the scene is based on some weights that help for making active decisions on the positions of objects relative to the camera observer. Finally, some applications shows the merits of the proposed scheme in some details.