54 Y. Reuland et al. 0 400 800 −4 −2 0 2 Falsification Thresholds Measured Value Combined Uncertainty Falsified Model Instances Candidate Models Model instances Displacement at Sensor 1 (x10−6m) Fig. 6.4 Falsification plot for sensor 1 based on the maximum displacement induced by a footstep signal 13.9 m 0 19.5 m 6.5 15 10 5 13.9 m 0 19.5 m 6.5 15 10 5 13.9 m 0 19.5 m 6.5 15 10 5 13.9 m 0 19.5 m 6.5 15 10 5 Sensor 1 Sensor 2 Sensor 3 Sensor 4 Accepted model instances (by sensor indicated above subplot) Falsified Model Instances Hall boundaries Sensor Locations Fig. 6.5 Instances of human locations that are falsified by each sensor Similarly, locations of the occupant are falsified using measurement data from all four sensors as shown in Fig. 6.5. From the figure, it can be seen that sensor 1, which is closest to the true location of the person, falsifies most locations of the occupant and is most informative of all sensors. In EDMF, the final candidate location set is comprised of locations not falsified by any of the sensors. Therefore, combining the information from all four sensors, the candidate location set is obtained and shown in Fig. 6.6. This set of candidate locations includes the true location of the step taken by the occupant. From Fig. 6.6, it can be seen that there are locations not falsified at the periphery of the hall and between sensor locations indicating a more comprehensive study of the sensor layout configuration might be useful in improving the accuracy of identifying candidate locations. Entropy-based justification will help in determining the usefulness of additional sensors for falsifying locations at the periphery. A detailed study of the model class required for localization might be useful in improving the robustness of candidate occupant locations obtained. Structural parameters such as support stiffness and material properties such as density and Young’s modulus were treated as deterministic and known values. 6.5 Discussion In this paper, a methodology has been presented for human detection and localization that can explicitly take into consideration ambiguities of the inverse problem. The methodology utilizes error-domain model falsification for updating knowledge regarding the presence and location of an occupant in a hall using vibration data. The methodology is
RkJQdWJsaXNoZXIy MTMzNzEzMQ==