Outlier Detection in Video Sequences under Affine Projection

A novel robust method for outlier detection in structure and motion recovery for afJine cameras is presented. It is an extension of the well-known Tomasi-Kanade factoriza-tion technique designed to handle outliers. It can also be seen as an importation of the LMedS technique or W S A C into the factorization framework. Based on the computa-tion of distances between subspaces, it relates closely with the subspace-based factorization methods for the perspec-tive case presented by Sparr and others and the subspace-based factorization for afine cameras with missing data by Jacobs. Key features of the method presented here are its ability to compare different subspaces and the complete au-tomation of the detection and elimination of outliers. Its pe flormance and effectiveness are demonstrated by experi-ments involving simulated and real video sequences. 1