Simultaneous tracking of features acquired by multiple video cameras mounted on 
a rig opens new possibilities for ego-motion estimation and 3D scene modeling. 
In this paper we propose a novel approach of tracking three video streams at 
once. The color image features are detected using interest operators and 
described with SIFT. Since standard tracking techniques perform outlier 
detection only according to relative orientation between temporal image pairs 
and hence suffer from outliers which cannot be identified by the epipolar 
constraints, we improve the outlier detection using temporal and spatial 
trifocal constraints. Furthermore, these spatio-temporal constraints allow the 
system to perform a guided matching, which increases the number of 
tracked features.