Multi-Modal Calibration of Surveillance Sensor Networks

Abstract

Abstract-Target detection and localization is one of the key research challenges in sensor networks. In this paper we propose a heterogeneous wireless sensor network integrating imaging and non-imaging sensors to accomplish the detection and localization task in complex urban environments. The lowcost non-imaging sensors provide early detection and partial localization of potential targets and direct imaging sensors to focus on them. Accurate target location estimated by the imaging sensors is subsequently used for in-situ calibration of the non-imaging sensors so that localization error is minimized over time. We evaluate our approach through simulation and our preliminary results reveal that coordination across different sensing modalities increases localization accuracy and can reduce the amount of imaging data that must be carried by the network.

A Closed Form Solution for the Self-Calibration of Heterogeneous Sensors

“…A set of synthetic experiments has been performed in order to evaluate the self-calibration algorithm with varying numbers of cameras, range sensors and targets. This setup is normally representative of a sensor calibration scenario [8,26] and target localisation problems from randomly displaced sensors [19,23]. Cameras, sensors and targets were randomly displaced into a cubic region according to a uniform distribution.…”

“…In video-surveillance, dangerous targets should be localised before the situation may degenerate [5]. Military applications might also use a sensor network in order to identify ground targets [8] or hidden snipers in a combat area [24]. Workers may also be localised in order to warn them that they are trespassing a dangerous area in a factory [9].The common aspect among all these scenarios is that they all need precise measurements of the target position.…”

Tracking in Urban Environments Using Sensor Networks Based on Audio-Video Fusion

“…For hybrid Bayesian estimation with audio-video clustering, we compute the likelihood functions using Equation (9) or (10). The audio posterior density is calculated using…”

“…Tracking is one such application that can benefit from multiple sensing modalities [9]. If the moving target emits sound signal then both audio and video sensors can be utilized.…”

Multi-Modal Target Tracking Using Heterogeneous Sensor Networks

“…Target tracking is an application that can benefit from multiple sensing modalities [6]. If the moving object emits some kind of sound then both audio and video sensors can be utilized.…”