GPS World, January 2011
WIRELESS LBS Personal Navigation VT Thus a speed sample S t i s i p i p i 1 ground N i 2 is obtained In addition this sample is statistically through robust and non robust estimators yielding E S and thus deriving the persons ground speed profi le On the other hand accelerometers are the key sensors to enable qualitative observation computation to later derive a qualitative attitude that is the detection of a collapse Accelerations are involved in the computation of four types of qualitative observations and its use is based on the following three statements Independence of any initial attachment or placement of the device on the body is fundamental to ensure a loose and easy start up of the device Independence of any sensor errorcalibration should not be an issue Balance is the key observable to perform collapse detection First balance changes are extracted from accelerometers as they sense the gravity vector projection on each axis and any change on these projections is interpreted as balancing the device Indeed balance is not exactly attitude the gravity vector defi nes a normal plane called equilibrium plane which is a 2 degree of freedom object Nevertheless the left degree of freedom not sensed in this approach corresponds to the heading changes which do not contribute to collapse situations Therefore given a 3 axis acceleration sample A N t i ai x ai y ai z N i 1 an analysis is performed using robust and non robust statistical estimators as monitoring the fi rst and second statistical moments of this sample enables detection of variations on the gravity distribution among the axes Finally thresholding is performed on the propositional calculus to obtain balance change extraction Second given an acceleration sample A N high accelerations are extracted using the distance operator d i a i E A N and a threshold based propositional calculation Third accelerations are also used for low motion detection Given an acceleration sample A N fi rst and second moments of the acceleration norm sample E A N and V ar A N E A N E A N 2 are computed and evaluated through threshold based propositional calculations to detect norm wise low acceleration profi les Finally accelerations are the key observations to perform periodicity detection Given a set of accelerations A N two deterministic analyzers are used to extract periodicity patterns EMS and FFT The fi rst technique enables computing j local maximum values one for each sub sample A Nj j 1 m where A N Um j 1 A Nj Evaluating the j local maximum values interdistance along time against some thresholds enables periodicity detection The FFT analysis complements the periodicity detection achieved by the EMS technique In addition to the extraction itself a fi gure of merit FOM is computed for each qualitative observation Consisting of a rational number between 0 and 1 it is an empirical magnitude describing how many extractions have been done for a certain observation in relation to the maximum possible amount of extractions This fi gure enables a reliability computation similar to a discrete probability function Nevertheless at this stage of development we do not claim completeness and therefore do not state that FOM computation is a discrete probability function High Level Processing from qualitative observations to qualitative states So far one may think that the navigation requirements are already fulfi lled a person can be localized in a seamless indoor and outdoor way and thus can be feasibly reached if needed But IEGLO seeks to enhance this navigation concept to provide contextual information about the person and eventually activate automatic warning messages in case of undesired motion behavior To do this the qualitative navigation concept has been developed by analogy of the quantitative navigation qualitative or quantitative observations are used to estimate qualitative or quantitative states The qualitative states in IEGLO are t R V motionless walking Ω collapse This particular choice of the navigation state is fully driven by the user requirements With the estimation of the collapse and motionless states IEGLO can provide the user with an automatic distress detection system These two states specially represent the type of undesired behaviors that IEGLO seeks to detect and respond to In addition to the distress states walking is useful to support the pedestrian navigation concept which is based on single point navigation As can be seen in Figure 2 collapse estimation is performed by means of the balance change and high acceleration qualitative observations motionless estimation is performed by means of the low motion qualitative observation walking estimation is performed by means of the ground speed segment and periodicity qualitative observations In all cases the weighted combination of the qualitative observation FOMs is performed to determine the qualitative state FOM as a degree of truth The role of the FOMs is crucial when generating automatic alarms in case of eventual distress situations The more accurate the FOM the fewer false alarms are generated Note that in this high level processing approach every model q f M must be fed by values that are external to the process These values help to fi ne tune the adjustment of the model to the user carrying the device In pedestrian navigation values like step strength and time to step play a role in the walking model and fully depend on the individuals way of walking In IEGLO the knowledge of the individual user is GPS World January 2011 www gpsworld com 68
You must have JavaScript enabled to view digital editions.