Temporal calibration algorithms estimate time offset between data streams acquired by different devices.
Time difference between a hardware device acquires data and Plus receives the data is slightly different in each device. This time difference can be up to a few hundred milliseconds (depending on the hardware devices, data acquisition settings, computer hardware, connection type, network or CPU load, ...).
Hardware devices often do not provide an accurate timestamp for the acquisition of each frame, and even if they do, the time reference (clock) of the various hardware devices are typically not synchronized. To compensate for these inaccuracies and temporally synchronize data streams, Plus allows estimation and compensation of fixed time delays and can provide timestamp jitter filtering.
Time offset estimation
Temporal calibration method in Plus assumes that timestamps on the received data items have a constant time offset compared to other clocks.
The calibration algorithm requires data acquired while performing continuous linear quasi-periodic motion for at least about 5 full periods (typically about 10-15 seconds). Although the algorithm may work for fewer periods it has not been tested under these conditions. When an image stream is used the data assumed to be collected by an ultrasound probe imaging a planar object. Furthermore, it is assumed that the probe is undergoing periodic motion in the direction perpendicular to the plane's face (e.g., moving the probe in a repeating up-and-down fashion while imaging the bottom of a water bath).
The first step of the algorithm is the computation of a normalized position signal from each data stream.
Time offset can be computed between two devices at a time (using either tracking of imaging data). If there are more devices then each device has to be included in at least one temporal calibration.
Notes:
- The time offset is determined by computing the time offset that leads to the highest correlation between the two signals. For the correlation computation the moving signal is linearly interpolated at the time positions where the fixed signal is known. As the acquisition rate of imaging data is typically lower than the acquisition rate of tracking data, it is preferable to use video as fixed signal and tracker as moving signal.
- The fixed and moving signal is cropped to the common time range. The moving signal is further cropped to the common range with
SetMaximumMovingLagSecmargin. - The visual inspection of the aligned signals provides useful diagnostic information about the root cause of temporal misalignments, such as large occasional delays or varying acquisition rate.
Timestamp jitter filtering
Timestamping on a personal computer with a non-real-time operating system is generally feasible with about 1ms precision when using multimedia timers. However, due to the non-real-time nature of general-purpose operating systems, slight variations in the timing may occur due to hardware interrupts and varying processor load. Also, some hardware devices transfer the data through Ethernet network, in which case there may be an additional unpredictable delay before the data collector receives the data.
If the acquisition hardware device acquires data items at regular time intervals and reports missing data items (e.g., by providing a monotonously increasing index for each frame), then it is possible to apply a timestamp jitter filtering method that ensures regular time intervals between the generated timestamps. The filtering method can also detect items with unreliable timestamps (items that are received with unusually long delay). In order to prevent data corruption, a data item is discarded if the difference between the measured and filtered timestamp is larger than a predefined threshold.
A line is fitted to the last N pairs of item indexes (independent variable) and unfiltered timestamps (measured variable) using simple linear regression. The filtered timestamp is the time corresponding to the frame index according to the fitted line.
- Configuration settings
- vtkPlusTemporalCalibrationAlgo
ClipRectangleOriginandClipRectangleSizedefine a region to take into account when line detection is performed on images. Optional attribute, if not specified then the whole image is used.-
SaveIntermediateImagescontrols if diagnostic information of line detection in images is written to files.
Default ="FALSE".FALSENo debug information will be written.TRUEImage files are written to the output directory that show the lines along image intensity is sampled and the detected line.
- SetMaximumMovingLagSec defines the maximum time lag that will be considered by the algorithm, in seconds.
Default =
"0.5 sec"
- Example configuration file
