Monthly Technical Report October 1998 SRI Project Number: ECU-1515 Contract Number: NMA202-97-C-1004 DARPA Order Number: E645 Prepared by: Aaron J. Heller, Sr. Computer Scientist Prepared for: Michele Motsko, Physical Scientist National Imagery and Mapping Agency Reston, VA SUMMARY OF ACTIVITIES: 1. Preparation for November demos. 2. Diagnosis of problem with new Ft. Hood triangulations 3. Enhancement of FICS with "spur" detection algorithm 4. Work with new Ft. Benning imagery collection 5. VPF import 6. Experiments with dense DEMs. DETAILED DESCRIPTION: 1. Preparation for November demos. (All) In preparation for the November Classified demo, we selected a stereo pair from the NTC DPPDB collection that includes Ft. Irwin. We transfered the latest version of the APGD code to our classified system and carried out a number of experiments with the low-level extraction code to isolate and correct any problems with working over very large images. I also ingested the DPPDB, DTED, CIB, and mosaic over the OCONUS site we have been working on and selected a collection of study areas for further testing. For the IU workshop, to have unclassified imagery of NTC, I restored and reformatted USGS 1m imagery of NTC that we used on the MAGIC project. 2. Diagnosis of problem with new Ft. Hood triangulations (Quam,Heller) After receiving yet another set of incorrect TEC header file for the new Ft. Hood triangulations from GDE. Lynn Quam and I did a detailed investigation of the process GDE was using to translate from SocetSet support files to TEC headers and discovered that the parameter GRND_BIAS from the support files was being ignored by the translator, resulting in the errors. This was reported in some detail in an email I sent George Lukes, Michele Motsko, and Doug Climenson on 22 Oct 1998. In addition to this problem, I isolated another problem in the translator that was causing 1 pixel offset error in the TEC headers. These were repaired and used to generate new TEC headers from the SocetSet support files supplied by GDE, resulting in excellent camera parameters that are significantly better than those originally calculated by CMU. 3. Enhancement of FICS with "spur detection" algorithm (Connolly) Prompted by the observation that all of the detections missed by the low-level 2D algorithm are directly connected to either the complete network or the edge of the study area, we started development of a "spur" detector to be invoked by the FICS. This module scans the full road network and perimeter of the study area at full resolution and finds candidate roads that branch off the network. The FICS then uses CBACS to select an algorithm to add this road to the network. 4. Work with new Ft. Benning imagery collection (Heller) GDE delivered to SRI a copy of the tape with the SocetSet project of the scans and triangulations of the new Ft. Benning/McKenna MOUT coverage supplied by TEC. Using the corrected SocetSet to TEC header translator, I generated TEC headers for this imagery and ingested it into RCDE. The photometry of this imagery is quite different than the earlier McKenna coverage, with less detail in the highlights. The camera parameters are quite good and agree quite well with the earlier coverage. In order to make the data set as complete as possible I contacted Max Roos and requested copies of: - Camera calibration certificate and flight logs. - GPS reading from the aircraft. - Time and date of the exposures. - The SocetSet ground point file (.gpf), image point files (.ipf), and triangulation reports. These were supplied to my via FTP by Max Roos. 5. VPF import (Heller, Burns) We downloaded and installed the latest version of NIMA MUSE. I noticed that, as requested in an earlier report, the distribution now includes command line versions of the VPF and raster importing programs, allowing RCDE/BOS to directly invoke these programs without having to manually transcribe extraction parameters between the two systems. We tested vpfimp_cl to see if it had been updated to put attributes in the DXF output -- it was not. So we experimented with all of the output formats supported by vpfimp_cl and discovered that the MAPINFO MIF/MID format support attributes and is quite easy to parse. After tracking down a spec for MAPINFO MIF/MID (from http://www.geocities.com/~vmushinskiy/fformats/files/mif_mid.txt), I coded up an import routine for this format of data. This allows us to import point, line and area feature data from VPF sources such as DTOP into RCDE/BOS with full attributes. 6. Experiments with dense DEMs. (Heller, Burns) We carried out a number of experiments using the dense DEM and building footprints derived by Vexcel form the Ft. Hood panchromatic coverage. For simple buildings, (i.e., those with constant roof height, right angle corners) the following processing sequence is most successful: - Project the footprint into object space at an elevation determined from the dense DEM to yeild an approximation to the roofline of the building. - Use the most nearly nadir image to refine the shape of the roofline using the MBO system. - Extrude the roofline to the ground and refine again using the MBO system with all available imagery in which the building is visible. Complex buildings, i.e. buildings with a number of roof elevations or complex shapes, such as the "double-diamond", "lego", round buildings at Ft. Hood, cannot be handled by this technique. --------------------------------------------------------------------------- No report received from VEXEL for this period. --------------------------------------------------------------------------- GDE Systems APGD Monthly Report November 10, 1998 Technical Section Summary All of the work during this period has been on data set generation. Detailed Work Description (by SOW item): 1. Architecture Refinement No activity 2. Algorithm Survey No activity. 3. Algorithm Development No activity. 4. Multi-Sensor Registration No activity 5. Persistent Store. No activity 6. Dataset Production & Distribution: The discrepancy with SRI over the Fort Hood camera parameters has been resolved. A set of internal SOCET SET? parameters, which would normally be set to zero prior to distribution of the camera parameter, were left with non-zero values. This was not noticed since SOCET SET? automatically compensates for the non-zero values, so correct results are obtained by anyone using that system. As a result, Max Roos and NVO personnel both found the camera parameters to be excellent. New TEC headers have now been prepared and delivered to SRI. These new headers include the covariance matrices and UTM coordinates which were omitted in previous versions. There appears to be general agreement that the new camera parameters represent a significant improvement over previous parameters for the oblique images. Feature extraction work on Fort Hood has now resumed at the NVO, and the estimate as of this week is that the job is about 10% complete. We will continue that work as long as the budget permits. The feature extraction for Fort Benning has been completed and the resulting DXF feature files have been delivered to SRI. New TEC headers are being prepared, based on the triangulation by Max Roos, and including covariance matrices. 7. Evaluations: No activity 8. Interface to FRE Contractors: Discussions with USC have resulted in an informal goal to demonstrate their algorithm on classified imagery by the end of the current contract. Again, budget constraints will determine the extent of this work. 9. Demonstration Scenarios: No activity 10. Technology Transfer No activity 11. Option Years No activity 12. Program Management We are continuing to work according to the priorities agreed on with the prime. We keep in close contact with team members via weekly conference calls and e-mail communications as appropriate.