Tracking 3D line segments: new developments

Reports developments of research in analyzing long sequences of 3D frames obtained from stereo. The 3D primitives used are 3D line segments. Previously, the authors (1990) proposed a framework consisting of two levels. At the low level, each 3D segment is tracked from frame to frame independently and its motion is estimated by filtering over time. This is the temporal integration. At the second level, the main function is to group line segments into objects according to the similarity of their motion parameters. This is the spatial integration. The authors describe a new formulation in the same framework to track segments and estimate their motion. Using this formulation, they describe any complex motion if its angular and translational velocities can be expressed by time polynomials. They propose also a new technique based on a statistical test to decide whether a token has enough support of existence to continue being tracked or can be discarded. This is very important to deal with problems such as occlusion, disappearance and absence of tokens. Experimental results are provided. >