DIGITAL

Evaluation of Matching Quality for MER Data

Publication from Digital

Eva Rott

, 1/2014

Abstract:

The objective of this work was to create a software component to evaluate matching quality using stereo image data from planetary rovers, the US twin MER rovers being the most representative at the time of writing. For verification, the program used two matching algorithms, HFVM and SGM, in particular and offered nine ways for evaluation. It was designed as operational tool to find the best suited algorithm for space rovers. The nine criteria were chosen based on the requirements posed by the application of stereo matching in space. The evaluation tool uses reference data like ground truth for evaluating the matching results. The methods could be separated in the following classes: 1.-class focused on the evaluation of the disparity maps by comparing them to ground truth. 2.class warped the left to the right image using the disparity maps.  3.class classified pixels based on the idea of the receiver operator characteristics. Two different criteria belong to that class. 4. class
 calculated histograms based on the error in the disparity maps. 5.class used information about the backmatching error to classify the matching results. In a similar method the coverage of matches is assessed. 6.class evaluated by assessment of the quality of the input images using information from the computed disparity maps. 7.class compared the execution time of different matching algorithms. The software was tested and it was proven that executing all of the criteria with the same settings (input and matching algorithm) lead to the same evaluation result. This should prove that the implementation of the different evaluation methods is correct. The tested set of images was laboratory data used in the Mars 3D challenge of the EU FP7 Project PRoVisG. The matching result was evaluated using all methods.