This paper presents an accurate and robust motion stereo algorithm employin
g multiple images, taken under a general motion. The object functions for i
ndividual stereo pairs are represented, in terms of the depth; then these o
bject functions are integrated, considering the positions of cameras and th
e shape of the object functions. By integrating the general motion stereo i
mages, we not only reduce the ambiguities in correspondence, but also impro
ve the accuracy of reconstruction. Also, by introducing a weight for each o
bject function and selecting only reliable object functions, we can improve
the accuracy greatly and eliminate false matches, caused by occlusion or n
oise. Experimental results on the synthetic and real data sets are presente
d to demonstrate the performance of the proposed algorithm. (C) 2001 Patter
n Recognition Society. Published by Elsevier Science Ltd. All rights reserv
ed.