Jamet - 7 
In fact, the actual operationally of the process may raise more complex questions, and 
need a rather large amount of experiments. The efficiency relies on the ability of the 
operator to locate the errors. This could be tested independently, but is also linked to the 
overall accuracy of the results. A complete study should evaluate the reached accuracy 
as a function of the amount of work spent in editing. 
This study is currently running at IGN. Four test sites have been chosen, from a very 
flat area (Corbeil-France), to a mountainous one (Grenoble-France), the other two sites 
corresponding to intermediate relief (Reims and Agen, France). For each site, a couple 
of 1:30 000 scale aerial photographs was digitised at a step of 17 microns. The DEM 
and the smoothing processes were computed at lower ground resolutions (correlation at 
1.5 m, smoothing at 5m to 25m). The quality criterion chosen rely on the comparison 
between the elevation of manually plotted topographic features (elevation extrema, 
rivers...). The study will include a qualitative estimation of the difficulty for controlling 
the algorithms (determination of the parameters), an overall estimation of the resulting 
accuracy, and an estimation of the needed editing time. 
(Raw DEM) 
(BDTopo in black 
Automatic result in gray) 
process) 
Figure 1 
5m interval contour lines on the Reims test site (7x3 km^ area) 
Results on this latter point aren't available yet. However, a qualitative look at the 
automatically computed contours and at the accuracy results gives some clues. In