Study on Extraction of Walking People and Standing People Using Mobile Laser Scanner

Abstract

The study proposes a methodology of extraction and tracking of walking and standing people using mobile laser scanners. Measurement experiment was carried out in a populated real environment, i.e. exhibition hall, to assess the accuracy of extraction and tracking. Steps of the extraction and tracking are ; 1) Laser scanner movement is tracked with SLAM (Simultaneous Localization and Mapping) using horizontal laser data, 2) Moving objects i.e. walking people are extracted by elimination of static objects from horizontal laser data based on the reconstructed trajectory of the horizontal laser scanner. The experiment revealed that accuracy of pedestrian extraction and tracking was 79.2%. 3) Three-dimensional data are generated by combining the laser scanner trajectory and slant laser data. After candidates of standing people are extracted through the clustering of three dimensional data using mean-shift algorithm, standing people are further extracted from the candidates with template matching using twelve templates. Out of 287 standing people, 190 was extracted through this process (extraction rate is 71.2%), while 287 clusters are extracted as standing people though 190 clusters actually corresponds to standing people (percentage of correct classification is 66.2%).