Robust object tracking
Miranda Elaine Glynis Armstrong · Asian Conference on Computer Vision · 1995
We describe an object tracker robust to a number of ambient conditions which often severely degrade performance, for example partial occlusion. The robustness is achieved by describing the object as a set of related geometric primitives (lines, conics, etc.), and using redundant measurements to facilitate the detection of outliers. This improves the overall tracking performance. Results are given for frame rate tracking on image sequences.