Self-adaptive fire evacuation by using self-designed 3D indoor positioning system

The efficiency of indoor fire evacuation plays an important role for survival improvement and the development of smart fire evacuation system can help solve this problem. An ideal intelligent indoor fire evacuation system must consider users’ physical features and provide a customized evacuation route based on their positions. Meanwhile, it should be able to track the real-time environmental conditions in internal networks in indoor environments. In other words, this type of fire evacuation system should be able to react immediately to the environmental changes in indoor area and provide real-time and valid navigation at any time during movements, avoiding panic/stampede and congestions in exits. This kind of system will be of great importance in future application for human safety. It requires the guidance service to be able to provide current user locations and provide a nearest available exit based on this information, integrated with fire expansion information. It is highly possible under the quick development with the improvement of indoor positioning technologies. The research presented in this thesis has developed a novel 3D positioning system in order to provide solutions for user localization and navigation during fire evacuation and the effect with and without support of navigation will be assessed based on the results of simulation. This study will first provide a review of popular indoor positioning technologies and select possible techniques based on the demands of flexible localizations with satisfied accuracy and low cost with few infrastructures. Pedestrian Dead Reckoning (PDR) and visual tracking are then selected as promising candidates to be combined for 2D positioning and tracking. The applications of the corresponding algorithms for the selected positioning technologies have been chosen based on the comparison of the accuracy and the easiness of operations in the review as well. It will then provide a self-designed system with the integrations of the above selected techniques for horizontal positioning of each floor, within the testing environment located inside a four-floor building of University of Nottingham Ningbo China (UNNC). This 2D passive vision-aided PDR positioning system proposed by this study can achieve an average positioning accuracy of 0.08m on a single floor with less impact of occlusion, which is higher than the systems using similar sensors while using a simpler algorithm, fewer sensors and quicker computation. It has also been tested under the situation with severe occlusions in the selected building. Its accuracy (0.16m) is still comparable to the other studies with less occlusion, which has shown the reliability and stability of the performances for the algorithms while still keeping the advantages of fewer sensor requirement (low-cost) and better sensor accessibility (user-friendly). The system is then further developed into a 3D version with the ability for floor identification by using a smartphone-based barometer. It also achieved a comparable accuracy of height estimation (0.5m) to other studies using the barometers while using fewer sensor and simpler computation. The accuracy of the floor detection is around 98%. The above achieved accuracy in both horizontal and vertical directions are better than the required accuracy targeted by several emergency services, including the Federal Communications Commission (FCC). The above designed tracking system as well as the applied algorithms for sub-systems is the major theoretical contribution of this research. The system can also be applied for speed and inter-personal distance measurement, when tracking the movement of the pedestrians. These measured parameters can be applied into the simulation of the indoor fire evacuation process with the support from the smartphone-based navigation system by using a social-force based agent-based model, with the integration of a simplified fire expansion model. Moreover, the PDR based action recognition can also provide good support for posture recognition and localization reporting of people for later rescue. With the establishment of the simulation model, this study is able to discover the bottlenecks inside the selected building under normal conditions. Moreover, it is able to compare the efficiency of two evacuation strategies, i.e. nearest exits and random walking. These two strategies can represent the indoor evacuation with and without the support of the navigation system. According to the results, the evacuation with the support of navigation system (nearest exits) is more efficient with higher survival rate, shorter average evacuation time, and shorter average evacuation distance. With the above experiments and simulations, this study has achieved an initial success of developing an indoor evacuation navigation system, with promising results