Widespread adoption of new wireless technologies and diffusion of self-organizing networks ruled by ad-hoc protocols, have fostered, in recent years, the interest in the research of new methods for the positioning process, mainly focused on passive localization approaches. Differently from the active case, where an electronic device is attached to the tracked entity, in passive localization systems the target is not required to partecipate actively in the positioning process. The localization systems rely on the impact that the activity of people within the vicinity of a wireless network induce on the wireless links. When a target crosses a network link, it causes perturbation on the received signal strenght (RSS). By modelling and measuring the perturbations over all the network links monitoring a limited area, the system is able to reconstruct an image of that area and locate the moving targets. Device-free localization (DFL) is promising for a wide range of applications, such as intruder detection and tracking, emergency rescue, border protection, context awareness in mobile computing, ambient assisted living in smart spaces, access control and counting, machinaries remote control in industrial environment, crowd density and pedestrian traffic monitoring in public spaces. Thanks to the recent progress in this technology, the network can be composed by any kind of wireless device that is able to use commercial radios (e.g., operating at 2.4GHz over a small bandwidth) and sensing the environment changes. The devices can be pre-deployed during the off-line installation phase or randomly placed during the on-line working phase, inside or in the surroundings of the detection area. Each device can estimate the coordinates of the other network nodes by a preliminary phase of cooperative localization [21, 22]. The RSS field is affected by the presence of the target in such a way that it is possible to estimate its position with a reasonable accuracy. Usually, positioning systems adopt fingerprinting approach to model the RSS perturbations and estimate the target position by matching the RSS real-time measurements collected by all the network links to the stored ones. These maps, built in the calibration phase, provide the statistical model of the power attenuation and fluctuations of RSS for each possible target position. The drawback is that the calibration process can be very complex, expensive and time consuming since, in fingerprinting, the target has to visit all the positions of the area for digital map construction. In order to simplify the construction of the perturbation maps, previous works studied analytical approaches such as diffraction model [5] or ray tracing [12]. In this case, the calibration phase is faster and easier, since it requires only the tuning of few model parameters. Since only the single-target case has been presented in literature, here we extend the analysis proposing new models and algorithms for multi-target device-free localization, validating the analysis by extensive measurements campaigns. Concerning the multiple-target case, previous works carried out some experiments in order to model the relation between the impact of one single target and the new effect measured on the RSS when additional targets enter in the area. In [1, 4] the additive model has been proposed, while in [10] another linear relationship between the total measured variation in RSS and the number of targets in the area, has been derived from an experimental campaign. In this thesis work, we aim to validate both additive and diffraction-based model, by an intensive measurements campaign. We evaluate the models with four different localization algorithms: Joint Maximum Likelihood (JML), Successive Cancellation (SC) and Radio Tomographic Imaging (RTI). While the former two are based on the ML estimation, the latter employs different estimation criteria. JML is an extension of the basic ML estimation, adopting the joint perturbation maps for the estimation of the combination of the targets position. SC approach, instead, was introduced in [15], where an iterative algorithm has been applied for the position estimation of up to four mobile targets. RTI was originally presented for the single-target case in [3, 6, 7, 8]. The same approach has been applied also to scenario with a small group of users in [9], assuming the total effect in RSS variations as the sum of the single entity effects. The advantage of this technique is the simplification of the localization process to the estimation of the motion image of the area by processing the power attenuations or variance fluctuations measured in the radio signal over the network links. In this thesis, we implement RTI in a novel manner, exploiting jointly the attenuation in power and the variance of the RSS measurements, for the multiple target scenarios, both with ML and WLS criterion. Then, we implement the particle filtering (PF) technique for target tracking in dynamic scenarios, extending [14] to the multi-target scenario. Dynamic Bayesian filter increases the localization accuracy and allows to reduce ambiguity problem that arise in multi-target scenarios. Finally, we consider the case of up to seven targets located along the LOS path in order to find the relationship between the total perturbations measured and the number of adjacent targets located along the LOS path. Models and localization algorithms here presented, have been validated by numerous experimental campaigns.
In questo lavoro di tesi, dai contenuti sia teorici sia sperimentali, si è affrontato il problema della localizzazione passiva con reti wireless, nota anche come device-free-localization (DFL). In particolare, alcune tecniche sperimentate per il singolo target posizionato all'interno di un'area monitorata da una rete wireless, sono state estese al caso multitarget. Negli ultimi anni, i sistemi di localizzazione sono stati oggetto di interessanti studi e la base per lo sviluppo di nuove applicazioni che spaziano dal controllo assistito in ambienti familiari ai sistemi di sicurezza per la rilevazione di accessi non autorizzati, dal monitoraggio del movimento di masse pedonali alla regolazione del traffico stradale secondo i movimenti della folla pedonale. La grande diffusione di questi sistemi è dovuta alla flessibilità a dimensioni e costi ridotti dei dispositivi adoperati per la rete e al loro funzionamento senza che sia necessario il coinvolgimento attivo del target da localizzare. Questa tecnica si basa sulla stima della posizione del target attraverso l'analisi dell'attenuazione e fluttuazione indotta dai target mobili sui segnali radio (Radio Signal Strenght, RSS) scambiati tra i nodi della rete. Il processo di localizzazione richiede la costruzione di mappe (analitiche o sperimentali) che forniscano una descrizione statistica delle variazioni RSS al variare della posizione del target, qui indicate come 'mappe di perturbazione'. A questo proprosito, in [5] è stato proposto e testato un modello analitico capace di approssimare con buoni risultati le mappe, riducendo così i costi di calibrazione di perturbazione, sulla base di una modellazione analitica della diffrazione generata da target. In questa tesi, viene proposta un'estensione di tale modello al caso di target multipli posizionati sullo stesso link, con l'obiettivo di definire la correlazione che regola gli effetti dei singoli. Lo stesso modello è stato confrontato con il modello additivo, la cui validità è stata oggetto di discussione in molti articoli recentemente pubblicati. Quest'ultimo assume che l'effetto dovuto alla presenza di più target equivale alla somma dei singoli effetti, quando ciascun target è considerato come l'unico all'interno dell'area di analisi. Nella tesi sono stati inoltre proposti diversi algoritmi per la stima della posizione dei target a partire dalle mappe di perturbazione, valutate sperimentalmente con campagne di calibrazione (fingerprinting) o analiticamente con i modelli sopra descritti. In particolare, per la localizzazione, entrambi i modelli sono stati implementati in quattro diversi algoritmi di localizzazione, e le performances discusse. Per la localizzazione con due e più target, sono stati implementati quattro metodi: stima a massima verosimiglianza congiunta, cancellazione successiva e metodo tomografico con approccio a massima verosimiglianza o ai minimi quadrati. In particolare, la tecnica di cancellazione successiva prende spunto dall'analisi fatta in [15], ma applicata al modello diffrattivo singolo target e al modello additivo. Per quanto riguarda l'approccio tomografico, rispetto alle tecniche in letteratura, è stata proposta una nuova metodologia che permette di sfruttare congiuntamente le informazioni di attenuazione e varianza generate dai diversi target e misurate sul segnale RSS, includendo le mappe di calibrazione nella stima dell'immagine rappresentante l'area di analisi. Tutti i metodi sono stati testati attraverso numerose campagne di misura condotte in quattro diversi scenari sperimentali. Inoltre, l'introduzione del tracking Bayesiano tramite particle filtering (PF) esteso a più target ha permesso di migliorare l'accuratezza della localizzazione. Infine, è stata effettuata una breve analisi sulle variazioni RSS misurate al crescere del numero di target posizionati lungo il link, che ha confermato la non validità del modello additivo, soprattutto se questi sono molto vicini tra loro. Il modello analitico diventa quindi cruciale per l'estensione a più target.
Multi-target device free localization
SCHIAROLI, SILVIA
2014/2015
Abstract
Widespread adoption of new wireless technologies and diffusion of self-organizing networks ruled by ad-hoc protocols, have fostered, in recent years, the interest in the research of new methods for the positioning process, mainly focused on passive localization approaches. Differently from the active case, where an electronic device is attached to the tracked entity, in passive localization systems the target is not required to partecipate actively in the positioning process. The localization systems rely on the impact that the activity of people within the vicinity of a wireless network induce on the wireless links. When a target crosses a network link, it causes perturbation on the received signal strenght (RSS). By modelling and measuring the perturbations over all the network links monitoring a limited area, the system is able to reconstruct an image of that area and locate the moving targets. Device-free localization (DFL) is promising for a wide range of applications, such as intruder detection and tracking, emergency rescue, border protection, context awareness in mobile computing, ambient assisted living in smart spaces, access control and counting, machinaries remote control in industrial environment, crowd density and pedestrian traffic monitoring in public spaces. Thanks to the recent progress in this technology, the network can be composed by any kind of wireless device that is able to use commercial radios (e.g., operating at 2.4GHz over a small bandwidth) and sensing the environment changes. The devices can be pre-deployed during the off-line installation phase or randomly placed during the on-line working phase, inside or in the surroundings of the detection area. Each device can estimate the coordinates of the other network nodes by a preliminary phase of cooperative localization [21, 22]. The RSS field is affected by the presence of the target in such a way that it is possible to estimate its position with a reasonable accuracy. Usually, positioning systems adopt fingerprinting approach to model the RSS perturbations and estimate the target position by matching the RSS real-time measurements collected by all the network links to the stored ones. These maps, built in the calibration phase, provide the statistical model of the power attenuation and fluctuations of RSS for each possible target position. The drawback is that the calibration process can be very complex, expensive and time consuming since, in fingerprinting, the target has to visit all the positions of the area for digital map construction. In order to simplify the construction of the perturbation maps, previous works studied analytical approaches such as diffraction model [5] or ray tracing [12]. In this case, the calibration phase is faster and easier, since it requires only the tuning of few model parameters. Since only the single-target case has been presented in literature, here we extend the analysis proposing new models and algorithms for multi-target device-free localization, validating the analysis by extensive measurements campaigns. Concerning the multiple-target case, previous works carried out some experiments in order to model the relation between the impact of one single target and the new effect measured on the RSS when additional targets enter in the area. In [1, 4] the additive model has been proposed, while in [10] another linear relationship between the total measured variation in RSS and the number of targets in the area, has been derived from an experimental campaign. In this thesis work, we aim to validate both additive and diffraction-based model, by an intensive measurements campaign. We evaluate the models with four different localization algorithms: Joint Maximum Likelihood (JML), Successive Cancellation (SC) and Radio Tomographic Imaging (RTI). While the former two are based on the ML estimation, the latter employs different estimation criteria. JML is an extension of the basic ML estimation, adopting the joint perturbation maps for the estimation of the combination of the targets position. SC approach, instead, was introduced in [15], where an iterative algorithm has been applied for the position estimation of up to four mobile targets. RTI was originally presented for the single-target case in [3, 6, 7, 8]. The same approach has been applied also to scenario with a small group of users in [9], assuming the total effect in RSS variations as the sum of the single entity effects. The advantage of this technique is the simplification of the localization process to the estimation of the motion image of the area by processing the power attenuations or variance fluctuations measured in the radio signal over the network links. In this thesis, we implement RTI in a novel manner, exploiting jointly the attenuation in power and the variance of the RSS measurements, for the multiple target scenarios, both with ML and WLS criterion. Then, we implement the particle filtering (PF) technique for target tracking in dynamic scenarios, extending [14] to the multi-target scenario. Dynamic Bayesian filter increases the localization accuracy and allows to reduce ambiguity problem that arise in multi-target scenarios. Finally, we consider the case of up to seven targets located along the LOS path in order to find the relationship between the total perturbations measured and the number of adjacent targets located along the LOS path. Models and localization algorithms here presented, have been validated by numerous experimental campaigns.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_schiaroli.pdf
non accessibile
Dimensione
18.78 MB
Formato
Adobe PDF
|
18.78 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/111701