Robust Indoor Localization Using Histogram of Oriented Depth Model Feature Map for Intelligent Service Robotics

Abstract

Indoor service robots accumulate errors when its own reference map differs from the true environment such as when furniture has been rearranged in an area. This results in poor localization accuracy when pose estimation relies on outdated maps. Traditional methods address this issue by rebuilding the reference map from scratch. In this article, we propose a histogram of oriented depth model (HODM) and its extraction approach using laser rangefinders and RGBD cameras. HODM aims to provide a light and robust localization module so that mobile robots work in environments required by rearrangement from time to time. The key concept of HODM is based on using histogram-based model matching for estimating indoor primary structures and floor layouts. HODM localization will use an HODM map as a reference in scan matching, and the experimental results show that the localization error is lower than traditional non-HODM-based localization methods. In the same indoor location, the mapping process is only required once. The HODM approach presented in this article shows two major contributions. First, the localization error is lower even with an outdated reference map when compared with traditional localization methods, and, second, the computational time of HODM is fast. We validate this proposed method through numerical simulations and actual experiments in our laboratory using our experimental mobile robot developed in our NTU-iCeiA robotics laboratory.