Sensors and algorithms development for body sensor networks in healthcare environment

With the wide diffusion and popularity of laptops, cell-phones, Personal Digital Assistants (PDAs), GPS devices and other intelligent electronic in the post-PC era, computing devices have become more portable, mobile and cheap. Nowadays the electronic in uences the daily life of each man and many tasks hard to do in the past now have become reality and easy to perform thanks to the signi cant advances in technology. From this viewpoint the emergence of wireless sensor networks (WSNs) is essentially the latest trend of Moore's Law toward the miniaturization and ubiquity of computing devices. Wireless sensor networks are used in order to perform activity recognition in heath care eld, the results of this application show how that it is effective in patient's actions monitoring. Moreover an application regarding Heart Rate Variability (HRV)will be presented. This work is based on the analysis of the Rpeak to R-peak intervals (RR-intervals) of the ECG signal in the time and/or frequency domains. Doctors and psychologists are increasingly recognizing the importance of HRV; in fact, a number of studies have demonstrated that patients with anxiety, phobias and posttraumatic stress disorder consistently show lower HRV,even when not exposed to a trauma related prompt. Importantly,this relationship existed independently of age, gender, trait anxiety, cardio-respiratory tness, heart rate, blood pressure and respiration rate. The SPINE-HRV is composed of a wearable heart activity monitoring system to continuously acquire the RR-intervals, and a processing application developed using the SPINE framework. The RR-intervals are processed using the SPINE framework at the base station side through a time-domain analysis of HRV. The analysis provides seven common parameters known in medical literature to help cardiologists in the diagnosis related to several heart diseases. In particular, SPINE-HRV is applied for stress detection of people during activities in their everyday life. Experimentations carried out by monitoring subjects in speci c activities have shown the effectiveness of SPINE-HRV in detecting stress. Currently few research prototypes based on BSNs exist that allow for HRV analysis. However SPINE-HRV represents the fi rst prototype using a wireless chest belt so making the system more comfortable than systems using wired electrodes or handheld devices. Furthermore, because the chest belt is a commercial product for sport and tness activities, it has been designed to be robust against body movements. SPINE-HRV is currently applied to stress detection that is computed through an effective threshold based algorithm. The experimentation of such an application has been carried out on different subjects performing different activities of the everyday life: walking, working at the PC, watching TV, sleeping, and driving. The obtained result are interesting as they show that SPINE-HRV is able to detect stress by performing only a time-domain analysis of HRV with respect to more complex computational methods based on the frequency-domain analysis. Thus, SPINE-HRV can be actually used to detect stress of human beings in real-time. Currently, we are focusing our research efforts in improving the stress analysis algorithm by introducing frequency domain features as well as comparing the obtained results to the clinical blood test for the stress hormone, which has been identi ed by the medical community as the quantitative measurement of the emotional stress level. In the second part of this thesis will be described two smart video transcoder processes in order to develop a media gateway. The aim of this network device is to bring about a conversion of the input bitstream into another one characterized by a different video codec. The codecs involved in the transcoding algorithm are the H.263+ (Annex I) and the H.264 baseline pro le. The scope of this study focuses on the possibility of reusing the Intra modes extracted from the input bitstream. Regarding H.263+ to H.264 transcoding, two different thresholds are evaluated for 4x4 blocks and 16x16 macroblocks: all the incoming modes that lead to costs over threshold are rejected and a re-estimation is performed. Otherwise, the incoming Intra mode is directly passed to the H.264 encoder. On the other hand, all the H.264 Intra modes are mapped into the H.263+ Intra modes and passed to the H.263+ encoder skipping the Intra prediction stage. Performance in terms of PSNR and elaboration time of our algorithms are compared to that of the full transcoding approach. A high correlation with PSNR scores is obtained and a significant reduction of computational burden for both transcoding processes is also achieved. The two video transcoder architectures are proposed in order to perform the H.263+ to H.264 conversion and vice versa. Referring to the rst transcoder, two adaptive thresholds are implemented. Both thresholds, used for 4x4 Intra block mode decision and 16x16 Intra block mode decision, vary according to the overall macroblock cost in order to consider the level of detail of the under-study macroblock. This solution is an innovation relating to the approaches proposed in literature based on the usage of a single xed threshold. We can assert that these algorithms represent a basis for the implementation of a low complexity fast transcoder for real-time applications thanks to the low complexity of the modi cation introduced, and also for the reduced computational burden of the entire trancoding process. In fact, we demonstrate a decrease of about 32% in the overall elaboration procedure using an arbitrary QP. The proposed platform also shows high reliability in terms of perceived quality. This is confi rmed by PSNR evaluations for fast transcoding output. PSNR differences are limited to 0.1 dB for all sequences used in the tests. So, the quality of the full transcoding output is very close to the one obtained by the fast transcoding technique. In addition, the overall increase in the bitrate is less than 12%. The H.264 to H.263+ transcoder uses a mapping between the incoming H.264 modes that is rather different to the one proposed in literature. The obtained results, using several standard sequences and QP, show that the overall quality is the same for the output bitstream obtained by the full transcoder and the proposed smart transcoder algorithm, and the bitrate increase is limited to 9% in the worst case. With this mapping it is possible to cancel the computational burden of the Intra mode prediction process. All these considerations allow us to assert that the proposed algorithm can be used in real-time transcoding architectures. Similar analysis concerning Inter frame pictures are actually under study by the authors in order to reduce the complexity of motion estimation procedure in transcoding architectures.

La rapida diffusione e la grande popolaritá di laptops, smartphones,PDAs, dispositivi GPS e altri apparecchi elettronici nell'era post-PC, hanno alimentato la tendenza di produrre apparecchi elettronici sempre piú portatili, versatili e a buon mercato, con capacitá di calcolo sempre piú elevate. I continui progressi tecnologici hanno condotto ad un'abbondante disponibilitá di microprocessori e microcontrollori sempre piú piccoli ed economici, equipaggiati con sensori sempre piú avanzati, storage e dotati di connessione wireless. In quest' ottica si colloca l'emergere di una nuova tipologia di reti di telecomunicazioni: le Wireless Sensor Networks (WSNs), le quali rappresentano pienamente l'ultima tendenza della famosa legge di Moore nei confronti della miniaturizzazione e dell'ubiquitá dei dispositivi elettronici. L'integrazione di capacitá di calcolo, memorizzazione e comunicazione in dispositivi di dimensioni ridotte e a basso costo ha portato alla de finizione delle WSNs. Le reti di sensori sono state pensate come possibili strumenti per l'activity recognition in campo biomedico, i risultati di tale applicazione mostrano come questa essa sia molto e fficace nel monitoraggio della azioni di pazienti. Viene inoltre presentata un applicazione realizzata attraverso una WSN. Si tratta di un un applicazione per HRV (Heart Rate Variability). L'HRV é basata sull'analisi tempo-frequenza degli intervalli R-peak raccolti da un segnale ECG. Tale studio propone un toolkit realizzato attraverso una rete di sensori wireless per l'analisi temporale dell'HRV, chiamata SPINE-HRV (Signal Processing In Node Environment SPINE). SPINE-HRV é composto da un sistema indossabile per il monitoring dell'attivitá cardiaca in grado di raccogliere continuamente gli R-peak e un applicazione in grado di processare cosí i dati raccolti. L'analisi fatta attaverso lo SPINEHRV toolkit fornisce sette parametri ben noti in letteratura medica in grado di aiutare i cardiologi nella diagnosi relativa a diverse problematiche. Inoltre tale toolkit fornisce uno strumento automatico per rilevazione di stati di stress acuti rilevabili durante tutte le attivitá svolte quotidianamente. Nella seconda parte verrá presentata una panoramica sui media gateway in particolare sui transcoder video per gli standard di codi ca video H.263+ e H.264. L'eterogeneitá sempre piú diffusa dei dispositivi presenti all'interno della rete Internet, rende necessaria lo sviluppo di dispositivi hardware o software in grado da permettere una a dabile intercomunicazione tra tali diversi dispositivi. In particolare si mostrerá come é possibile riutilizzare i modi Intra estratti durante il processo di decodi fica per aumentare l'efficienza della codifi ca in altro standard di codi ca video. Sono stati sviluppati due algoritmi in grado di selezionare attraverso una decisore a soglia, utilizzato sia per i modi 4x4 che 16x16. Verranno presentati le prestazioni in termini di PSNR e tempi di elaborazione confrontati con quelle relative l'approccio full transcoding. Tali risultati mostrano come siano stati ottenuti signi ficativi riduzioni dei tempi computazionali pur mantenendo un livello di PSNR confrontabile con quello relativo al processo di full transcoding.