Human Position Estimation Based on Filtered Sonar Scan Matching: A Novel Localization Approach Using DENCLUE

Abstract

In this paper, a novel human localization approach called Human Position Estimation based on Filtered Sonar Scan Matching (HPE-FSSM) is proposed for the purposes of estimating the trajectory of a walking person in human-robot coexisting environments, by matching consecutive pairs of sonar scans obtained with a mobile robot. The work proposes a novel concept of removing large amounts of noise and spurious readings present in two consecutive raw scans by using a spatial clustering method called DENCLUE (DENsity-based CLUstEring). Moreover, an Edge Feature Based Leg Recognition (EFBLR) algorithm is developed, which, in both scans, extracts the data points characterizing the human legs. Finally, a Likelihood field (LF) based scan matching technique is implemented to estimate the roto-translation of leg pair between the refined scans. This procedure is repeatedly implemented to estimate human position, required for following. Extensive experiments carried out in different environments establish the supremacy of the proposed algorithm, compared to other state-of-the-art algorithms.