Indoor localisation systems have two principal subsystems, stationary units -- anchors, for which positions must be known and mobile units -- tags, which positions are determined with the localisation system. In the doctoral thesis, both subsystems are addressed to overall improve the indoor localisation system.
In the first part of the thesis, a localisation system with Ultra-Wideband radio is described. The system measures the distance between anchor and tag with the time-of-flight method. Two scenarios used in simulations and experimental validations are presented. The first scenario was placed in a gym, presenting a big open space. The second was in the Laboratory of Robotics, presenting
realistic conditions with obstacles and NLOS conditions, the environment in which localisation systems typically operate. Then an analysis of geometric dilution of precision for anchors and tag for both scenarios is performed. Reference systems and procedures for anchor position and tag pose are described.
For the anchor system, a quick calibration is desirable for the new anchors that are positionally undetermined in the working coordinate system. In the second part of the thesis, anchor calibration with an additional calibration unit for improving anchor localisation is presented, and its effect is analysed. Three localisation methods were tested for the anchor calibration: multidimensional scaling, semidefinite programming and iterative trilateration.
First anchor localisation accuracy was studied by simulating the change in the number of additional calibration modules and their positions. Analysis of the calibration unit's optimal position is presented. All analyses are conducted for both scenarios. In the second part of the simulations, an analysis of the effect of the height difference between anchors and the calibration unit is presented.
Experimental validation of anchor calibration was performed in two scenarios with reference measurements made by an electronic tachymeter. Additional analysis of the calibration unit's position effect on anchor localisation is presented. Finally, complete anchor calibration in a working coordinate system with four calibration modules on the calibration unit is conducted. Errors of less than 0,32 m are achieved in 3D.
In the third part of the doctoral thesis, orientation-induced distance error between anchor and tag is addressed. Model based on neural network is presented. A mechanism for rotation of the tag around the azimuth and elevation plane that is used in measurements for learning datasets is described. Analysis of the learning data set with additional neural networks and measurements of received power is presented, which confirmed the correctness of the measurements. For defining the number of neurons in the hidden layer of the neural network, many neural networks with different configurations were made. Validation of selected neural network configuration on a subset of training data not used in the learning process was performed.
For experimental validation of the neural network model, measurements were made with the localisation system in the Laboratory of Robotics. Reference measurements were made in combination with an electronic tachymeter and Optotrak reference system. Measurement results of tag in six poses with the use of the neural network model are presented. The use of the model improved the measured distances for 0,02 m. In the end, the effect of the model on tag localisation is presented.
|
