Image-retrieval Method Using Gradient Dilation Images for Cloud-based Positioning System with 3D Wireframe Map

Abstract

We propose a novel cloud-based precise positioning system that uses visual sensing data. Any mobile module with a vision sensor and wireless communication can be a client and can receive benefit from this system. When the client module takes a picture of an environment and uploads it to the server, it receives the shooting position with 6 degrees of freedom (DoFs) with an accuracy on the order of centimeters within a couple of seconds. The server maintains a map of the environment and localizes the uploaded picture in the map. The contributions of this paper are threefold. First, we develop a new visual localization method using a 3D wireframe map. The method proceeds in three steps: (i) the generation of an arbitrary perspective 2D image composed of line segments from a 3D wireframe map, (ii) the gradient dilation of a line segment image for effective image retrieval, (iii) pixelwise-AND-based image-similarity evaluation by parallel computing. Second, we build 3D CAD models of an actual building from a 2D design drawing and with manual measurements. Third, we experimentally evaluate our method using virtual sensing data.

Experimental Investigation on Map-Environment Disparity to the Performance of Image-Retrieval-Based Localization

“…The workflow of our previously proposed image retrievalbased localization method [8] is drawn in Fig. 2.…”

“…Here, the precisive map is regarded Fig. 5 The simulational experiment localization workflow (modified from [8]) Then the 250 number of images with localization error from the smallest are selected and set as the query images (q 1 ∼ q 250 ) for experiment. By this process, some inadequate images such as a no-line image or a lessline image by taking a wall from point-blank, can be removed.…”

“…We believe that the 3D CAD models can be the most appropriate map for device and robot localization problem. Using 3D CAD models, we have proposed a novel map-based localization system [8]. In this system, maps derived from CAD models are located on a server so that connected devices can get their own positions through inquiry process with the server.…”

“…To tackel this problem, we propose a simulational experiment using architectual 3D CAD models those Fig. 2 The workflow of image-retrieval-based localization method (modified from [8]) take both roles as a virtual environment for generating virtual sensing images and as a map for localization calculation. We prepare mainly two types of CAD models; the one, named "precisive model", consists of precisive structures, and the others, named "broken models", are made from the precisive model with shifting placement of several parts purposefully.…”

Inundation Analysis Method for Urban Mountainous Areas Based on Soil Conservation Service Curve Number (SCS-CN) Model Using Remote Sensing Data

“…The SCS-CN model requires appropriate parameter selection according to land-use type, terrain, and parameter localization. (21,22) The flood inundation analysis method proposed here is divided into three steps: data pre-processing, rainfall-runoff calculation based on the SCS-CN model, and flood inundation analysis. The specific technical route is shown in Fig.…”