The quality of the block parameters strongly depends on the size and 
number of gross errors present in the data. The On-Line approach, 
estabilished on the basis of an analytical plotter system has 
contributed significantly in cleaning the data as it ds. collected. 
The Step-by-Step method has established a progressive refining in 
terms of size of remaining gross errors. Both approaches seem to be 
of great value, as used together with Data Snooping and Robust 
Estimation. The combination of these techniques has showed 
efficiency in improving reliability and in saving processing time 
[06]. 
The role of the On-Line, in avoiding the cumulative amount of gross 
errors in the data is relevant. Its demand of an Analytical Plotter 
is still a difficulty for many companies that have only conventional 
equipments. 
This research investigated a set of algorithms and programs to 
overcome the demand of an Analytical Plotter, to assess the data 
progressively, as it is observed in a conventional enviroment. The 
Step-by-Step approach is combined in the system, to detect larger 
errors first, but is used also in a progressive way in time. On the 
refined data, available at a certain phase, least squares adjustment 
was performed and Robust Estimation was applied in some phases to 
improve detection and location of remaining errors. 
The choices of the functional mathematical models at the detection 
phasis were based not on an optimal fitting, but on the principle of 
minimum parameterization, so that the rigid model did not fit any 
existing blunder, improving the evidence of any gross error present 
at the observations involved at the adjustment. 
The system was called a "quasi on line" approach because it may 
start as soon as two photogrammetric unities are measured, and 
finish the cleaning of data soon after the last unity of the clock 
is measured. Then one, two or at most three executions of a block 
aerotriangulation are enough to conclude the adjustment of the 
block. The completely off line approach frequently requires from 
four to seven executions. It is well know that the difficulties of 
detection and identification affect also the reliability. 
2. PROPOSED ALGORITHMS 
Two main treatments are considered: ay the first when the 
photogrammetric unity for measurements is the photo; and b) when the 
unity for observations is the stereomodel. In the first case, the 
Observed data enter the progressive filtering and go through the 6 
steps: Si, windowing of  photocoordinates; S2, active lenght 
invariant transformation of corresponding images; S3, analytical 
aerial levelling; S54, strip connection, S5 revision of coding; and 
S6, control treatment. In the second case, the steps S1, S2 and $3 
are substitued by step Sc, strip formation from independent models. 
Each one of the above steps involves little processing effort and 
the volume of data is also small. The algorithms of the first step 
are simpler than the last ones. They detect only very large errors, 
- hj -