TY - GEN

T1 - Using motion from orthographic projections to prune 3-D point matches.

AU - Chen, Homer H.

AU - Huang, Thomas S.

PY - 1989/12/1

Y1 - 1989/12/1

N2 - The goodness of a match between points, given their 3-D position estimates, can be evaluated by checking how close corresponding points can be brought to coincide by a rigid motion computed from the given data. The performance of this well-known technique, however, is affected by the data accuracy. In stereo imagery, the depth values (z coordinates) estimated are subject to a higher range of uncertainty than the other two coordinates. The authors present a novel pruning method that discards the z coordinates and uses only x and y coordinates of points to compute the motion. It is noted that this method is more effective in detecting false pairings than the traditional method, which uses full 3-D coordinates. Since only x and y coordinates are used, the method is virtually equivalent to computing motion from orthographic projections. The authors dervive a least-squares solution to this motion problem and show that the determination of rotation and translation can be decoupled. The locus of rotation is proved to be a great circle on a unit quaternion sphere. Results of testing this method on real and random data are shown.

AB - The goodness of a match between points, given their 3-D position estimates, can be evaluated by checking how close corresponding points can be brought to coincide by a rigid motion computed from the given data. The performance of this well-known technique, however, is affected by the data accuracy. In stereo imagery, the depth values (z coordinates) estimated are subject to a higher range of uncertainty than the other two coordinates. The authors present a novel pruning method that discards the z coordinates and uses only x and y coordinates of points to compute the motion. It is noted that this method is more effective in detecting false pairings than the traditional method, which uses full 3-D coordinates. Since only x and y coordinates are used, the method is virtually equivalent to computing motion from orthographic projections. The authors dervive a least-squares solution to this motion problem and show that the determination of rotation and translation can be decoupled. The locus of rotation is proved to be a great circle on a unit quaternion sphere. Results of testing this method on real and random data are shown.

UR - http://www.scopus.com/inward/record.url?scp=0024867055&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0024867055&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:0024867055

SN - 0818619031

T3 - Proc Workshop Visual Motion

SP - 290

EP - 297

BT - Proc Workshop Visual Motion

A2 - Anon, null

T2 - Proceedings: Workshop on Visual Motion

Y2 - 28 March 1989 through 31 March 1989

ER -