11 Remote Space- and Time-Resolved Skin Perfusion Detection …
201
case, this encompasses tracking of local maxima of the brightness; at the same time,
any other correlations between image plane and a probability distribution are also
taken into consideration. In contrast to the optical flow method for example, the
tracking is not done by gradient operations. Instead inside a radial search window,
the mean central point of features belonging to the tracked objects is estimated. The
search window is repeatedly shifted to this position, and is estimated around a new
center until their positions concur. This algorithm features a simple approach for rapid
and robust tracking of single or low number ROIs. Compared to other algorithms,
this methodology is less susceptible to changes in illumination or slight alterations
in morphology of the tracked object. Within limits, it is also a more robust technique.
Nevertheless further information, to derive a probability distribution from the image
data is required.
11.5.3 Estimation of Respiration and Heart Rate by Motion
Tracking
As mentioned above, the tracking of observed objects in the PPGI sequence considerably improves the quality of extracted PPG signals. In doing so, there is a suppression
of motion and motion artifacts which would otherwise be a source of distraction from
the main objective of PPGI, i.r., the analysis of skin perfusion. Although movement
patterns are hampering in their own way, they also contain useful information about
vital processes. In particular, movements caused by breathing are visible in most
PPGI videos, especially if they are recorded from skin regions of the thorax. But also
recordings of other region, for example, on the extremities contain breathing motion
since these propagate mechanically also in other body parts.
The breathing correlated movement is compensated by applying a motion tracking
algorithm, and the calculated displacement vector of a fixed region contains this
breathing correlated motion. Other artifacts such as movements due to postural
changes or bodily gestures, are contained in the movement time signal as well. To
separate the latter from breathing movements, band-pass filtering exemplifies a good
first approach. As long breathing rhythms and other motion artifacts differ in their
frequency band, they can be easily segregated. This method holds good for short
single disturbances. In the event of more complex disturbances, additional information either of the noise or the breathing movement or better still both, must be
necessarily available.
The required information, concerning the breathing rhythms as well as disturbing
signals, can be elicited by choosing the appropriate mathematical model specific
to the application being used. Alternatively, redundant sensor signals (e.g., of a
standard patient monitoring) can be used if available. Whatever may be the case,
assessment of breathing correlated motion can obviously be best assessed if there
are no other disturbing movements present at the same time. However, this is just a
generalization; for a large number of PPGI applications, approximate assumptions
201
case, this encompasses tracking of local maxima of the brightness; at the same time,
any other correlations between image plane and a probability distribution are also
taken into consideration. In contrast to the optical flow method for example, the
tracking is not done by gradient operations. Instead inside a radial search window,
the mean central point of features belonging to the tracked objects is estimated. The
search window is repeatedly shifted to this position, and is estimated around a new
center until their positions concur. This algorithm features a simple approach for rapid
and robust tracking of single or low number ROIs. Compared to other algorithms,
this methodology is less susceptible to changes in illumination or slight alterations
in morphology of the tracked object. Within limits, it is also a more robust technique.
Nevertheless further information, to derive a probability distribution from the image
data is required.
11.5.3 Estimation of Respiration and Heart Rate by Motion
Tracking
As mentioned above, the tracking of observed objects in the PPGI sequence considerably improves the quality of extracted PPG signals. In doing so, there is a suppression
of motion and motion artifacts which would otherwise be a source of distraction from
the main objective of PPGI, i.r., the analysis of skin perfusion. Although movement
patterns are hampering in their own way, they also contain useful information about
vital processes. In particular, movements caused by breathing are visible in most
PPGI videos, especially if they are recorded from skin regions of the thorax. But also
recordings of other region, for example, on the extremities contain breathing motion
since these propagate mechanically also in other body parts.
The breathing correlated movement is compensated by applying a motion tracking
algorithm, and the calculated displacement vector of a fixed region contains this
breathing correlated motion. Other artifacts such as movements due to postural
changes or bodily gestures, are contained in the movement time signal as well. To
separate the latter from breathing movements, band-pass filtering exemplifies a good
first approach. As long breathing rhythms and other motion artifacts differ in their
frequency band, they can be easily segregated. This method holds good for short
single disturbances. In the event of more complex disturbances, additional information either of the noise or the breathing movement or better still both, must be
necessarily available.
The required information, concerning the breathing rhythms as well as disturbing
signals, can be elicited by choosing the appropriate mathematical model specific
to the application being used. Alternatively, redundant sensor signals (e.g., of a
standard patient monitoring) can be used if available. Whatever may be the case,
assessment of breathing correlated motion can obviously be best assessed if there
are no other disturbing movements present at the same time. However, this is just a
generalization; for a large number of PPGI applications, approximate assumptions