- 12 = 
all the main part of methods for control densification, the preparation 
for the collection of data and the actual collection of data are by far 
the most time-consuming operations. Since in off-line methods computa- 
tional processing takes place only after a considerable time lapse, one 
is sometimes unaware even of gross errors in observation while the 
influences of more subtle errors can only be discovered after the com- 
pletion of computation. Consequently, it is often necessary to repeat 
the measurements and the computations. To minimize the negative effects 
of such occurrences, a large number of redundant measurements may be 
made, so that some erroneous observations may be rejected, without need 
for remeasurements. 
If an analytical instrument is used for collection of informa- 
tion, some helpful techniques for editing the measured data may be 
devised. The simplest of these is the on-line computation of means and 
standard deviations of replicated measurements. This computation is 
performed in real-time. Residual errors are inspected automatically 
and the averages of the measured coordinates are stored without the 
intervention of the operator. The program is designed so that only in 
cases where, for example, the standard deviation exceeds some predeter- 
mined limits, a warning message and the results of the erroneous set 
of measurements are displayed to the operator. In this way, the 
measuring sequence is interrupted only when an error occurs. Otherwise 
the operator is not required to inspect the results of computations. 
Along the same lines, more sophisticated data collection and editing 
may be programmed. The capability of the analytical instruments to 
position the measuring mark on selected points automatically, that was 
mentioned in connection with orientation procedures, will obviously 
increase the speed and improve the convenience of the measuring 
procedure especially when used in an oriented model. The collection of 
data after the determination of the parameters of relative orientation 
may be performed with different degrees of sophistication. If a rough 
relative orientation using a minimum of observations is established, 
the residual parallaxes to be measured by the operator will be quite 
small. Replicated measurements on all the required points in an 
oriented model make possible the near-real-time determination of statis- 
tical indicators of goodness of fit of measurements to the mathematical 
model. Again the inspection is automatically performed and only when 
the results are found to be erroneous by the computer is a message con- 
veyed to the operator. The next possibility is to reconstruct a 
rigorous relative orientation. This procedure is the reasonable course 
to be taken when, for example, one intends to perform a block adjust- 
ment based on independent models. One of the side benefits in that case 
is that the coordinates of the centres of projection need not be deter- 
mined empirically since in on-line computational methods they may be 
arbitrarily chosen and remain unchanged by the computation, when only 
rotations are used as parameters for relative orientation. 
Many other techniques are available to increase the speed, 
accuracy, and convenience of operation as, for example, the point trans- 
fer techniques based on the recording of photo-coordinates of transfer 
points, the point addressing and measuring techniques based on the use 
of reseau photography, and housekeeping routines (such as the inhibition 
of Fes ration of coordinates for points with the same identification 
number).