Spatio-temporal coverage optimization of sensor networks
|Advisor:||Gagné, Christian; Parizeau, Marc|
|Abstract:||Sensor networks consist in a set of devices able to individually capture information on a given environment and to exchange information in order to obtain a higher level representation on the activities going on in the area of interest. Such a distributed sensing with many devices close to the phenomena of interest is of great interest in domains such as surveillance, agriculture, environmental monitoring, industrial monitoring, etc. We are proposing in this thesis several approaches to achieve spatiotemporal optimization of the operations of these devices, by determining where to place them in the environment and how to control them over time in order to sense the moving targets of interest. The first novelty consists in a realistic sensing model representing the coverage of a sensor network in its environment. We are proposing for that a probabilistic 3D model of sensing capacity of a sensor over its surrounding area. This model also includes information on the environment through the evaluation of line-of-sight visibility. From this sensing model, spatial optimization is conducted by searching for the best location and direction of each sensor making a network. For that purpose, we are proposing a new algorithm based on gradient descent, which has been favourably compared to other generic black box optimization methods in term of performance, while being more effective when considering processing requirements. Once the sensors are placed in the environment, the temporal optimization consists in covering well a group of moving targets in the environment. That starts by predicting the future location of the mobile targets detected by the sensors. The prediction is done either by using the history of other targets who traversed the same environment (long term prediction), or only by using the previous displacements of the same target (short term prediction). We are proposing new algorithms under each category which outperformed or produced comparable results when compared to existing methods. Once future locations of targets are predicted, the parameters of the sensors are optimized so that targets are properly covered in some future time according to the predictions. For that purpose, we are proposing a heuristics for making such sensor control, which deals with both the probabilistic targets trajectory predictions and probabilistic coverage of sensors over the targets. In the final stage, both spatial and temporal optimization method have been successfully integrated and applied, demonstrating a complete and effective pipeline for spatiotemporal optimization of sensor networks.|
|Document Type:||Thèse de doctorat|
|Open Access Date:||24 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.