Lidar and Vision Based People Detection and Tracking

Abstract

This paper presents a multi-sensor architecture to detect moving persons based on the information aquired from a lidar and vision systems. The detection of the objects are performed relative to the estimated robot position. For the lidar the Gaussian Mixture Model (GMM) classifier and for the vision the AdaBoost classifier is used from which the outputs are combined with the Bayesian rule. The estimated person positions are tracked via the Extended Kalman filter. The main aim of the paper was to reduce the false positives in the detection process with the use of a sequentially combined classifiers.