Sensor Deployment for Network-like Environments

TL;DR

This paper introduces a two-step gradient ascent method for optimally deploying omnidirectional sensors in network-like environments, effectively handling complex terrains and reducing local optima issues.

Contribution

It presents a novel two-phase optimization approach combining coarse and fine adjustments for sensor placement in network environments.

Findings

Effective sensor deployment in simulated airport environment

Reduction of local optima impact through initial coarse optimization

Distributed online fine-tuning of sensor positions

Abstract

This paper considers the problem of optimally deploying omnidirectional sensors, with potentially limited sensing radius, in a network-like environment. This model provides a compact and effective description of complex environments as well as a proper representation of road or river networks. We present a two-step procedure based on a discrete-time gradient ascent algorithm to find a local optimum for this problem. The first step performs a coarse optimization where sensors are allowed to move in the plane, to vary their sensing radius and to make use of a reduced model of the environment called collapsed network. It is made up of a finite discrete set of points, barycenters, produced by collapsing network edges. Sensors can be also clustered to reduce the complexity of this phase. The sensors' positions found in the first step are then projected on the network and used in the second finer optimization, where sensors are constrained to move only on the network. The second step can be performed on-line, in a distributed fashion, by sensors moving in the real environment, and can make use of the full network as well as of the collapsed one. The adoption of a less constrained initial optimization has the merit of reducing the negative impact of the presence of a large number of local optima. The effectiveness of the presented procedure is illustrated by a simulated deployment problem in an airport environment.