Kalman Filtering for Sensor Fusion in a Human Tracking System

Abstract

Robotic systems need to be context-aware in order to adapt their tasks to the different states of their environment. This context-awareness does not only imply the detection of the objects which are near the robot but it also includes the tracking of people who collaborate with it. Thus, human-robot interaction tasks become more natural and unobtrusive because robots are able to change their behaviour depending on this context information. In industrial environments, these context-aware systems should also guarantee the safety of human operators who interact with industrial robots. Therefore, a precise localization of all the limbs of the body of the operator has to be determined. In this chapter, the use of an inertial motion capture system for tracking full-body movements of the operator is described. It is composed of 18 IMUs (Inertial Measurement Units) attached to the body of the operator which determine the rotation angle of each joint. It has several advantages over other motion capture technologies: easy installation, self-containment, occlusions-fre e and precise rotational measurements. However, it accumulates a small error (drift) in the estimation of the global translation of th e human operator in the environment which becomes considerable after several movements of the operator. Therefore, an additional location system based on UWB (Ultra-Wide Band) signals has been added to correct this drift accumulation. The features of both tracking systems are complementary. The inertial motion capture system registers accurate joint rotation angles at a high rate while the UWB location system estimates global translation in the environment at a low rate. The combination of these systems will reduce the drawbacks of each one with the advantages of the other one. On one hand, the global translation measurements of the UWB system will correct the accumulated drift of the motion capture system. On the other hand, the high rate measurements of the motion capture system will complete the periods of time when there are not any measurements from the UWB system. Firstly, a simple fusion algorithm of both tracking systems is presented. This first fusion algorithm transforms measurements from the two systems in the same coordinate system by recalculating the transformation matrix each time a new measurement from the UWB system is received. This approach relies heavily on the accuracy of the measurements from the UWB system because the transformation matrix recalculation assumes that the last UWB measurement is completely correct. Thus, errors in UWB measurements are not considered and only the translational errors of the motion capture system are corrected. Furthermore,