Calibration of a Terrestrial Laser Scanner - Institute of Geodesy and ...
Calibration of a Terrestrial Laser Scanner - Institute of Geodesy and ...
Calibration of a Terrestrial Laser Scanner - Institute of Geodesy and ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
-HkKkHk)T5.3 Position-Fixing Using Total Station 107measurements into the a priori estimation to obtain an improved a posteriori estimation. It can be said thatthe time update acts as predictor equations <strong>and</strong> the measurement update acts as corrector equations.time update("predict")measurement update("correct")Figure 5.12: The Kaiman Filter cycle consisting <strong>of</strong> a time update <strong>and</strong> a measurement update according to [Welch <strong>and</strong>Bishop, 2004].In the following notations, the superscript minus (~) states for a priori estimations at step k given knowl¬edge <strong>of</strong> the process prior to step k.Otherwise, they are a posteriori state estimations at step k given mea¬surement zk. Since there is a discrepancy between a predicted measurement <strong>and</strong> the actual measurement, adifference is present, which is called filter innovation. The gain matrix K weightsthe update state estimation includingthe measurements:the innovation to obtain*fcxk==xk + Kk ik (5.26)zk xk(5.25)A summary <strong>of</strong> the Kaiman filter cycle including the equations are given in Table 5.10, adapted from [Welch<strong>and</strong> Bishop, 2004]. The filter has to be initialized by initial values for xq <strong>and</strong> Pq-Table 5.10:Bishop, 2004].Kaiman filter equations for the time update <strong>and</strong> the measurement update, adapted from [Welch <strong>and</strong>Time UpdateMeasurement Update1) Project the state ahead(1) Compute the Kaiman gainxk= ^k-ixk-i + rfc_ixfc_iKk=Pk-Hl{HkPk-Hl + Rk)^(2) Projectthe error covariance ahead(2) Update estimate with measurements zkPk=*fc-iPfc-i*ri + rfc_iQfc_ir£_1xk = xk + Kk(zk -Hkxk)(3) Updatethe error covariancePk = (I-KkHk)Pk-{I-+ KkRkKlThe Kaiman filter described uses all collected data <strong>and</strong> information prior to the present epoch. If the systemstate is not <strong>of</strong> interest in real-time <strong>and</strong> the Kaiman filter is applied in post-processing, the system statecan be smoothed by using information <strong>of</strong> epochs after the present epoch. Therefore, the filter runs not