Position estimation and fall detection using visual receding horizon estimation

The purpose of this paper is to estimate the position of a human in the image frame and to use this information to diagnose falls. A nonholonomic locomotion model describes the displacement of the human due to the similarities between human and nonholonomic mobile robot displacements. To estimate the human position in the world frame, the principle of Receding Horizon Estimation (RHE) is extended in the image plane. Indeed, this estimator is able to take into account an occlusion as a visual constraint. Residuals, errors between measured and estimated visual features, are generated to feed an alert dispositive. The latter will be used for the monitoring of an elderly person in a rest home. Thus the ground is assumed to be flat and a fixed perspective camera watches the scene. The simulations highlight the efficiency of the proposed approach, both without or with occlusions.