A SOM and feature based solution for correspondence problem in binocular vision

摘要

Aim of this work is to propose a robust solution to the correspondence problem in multi-camera systems applied to video surveillance. The proposed system merges two different approaches: Self Organizing Map (SOM) and feature based corresponding analysis. The novelty of this work consists of the used approach and the ability to work without the assumption of epipolar geometry. The proposed approach does not require a calibration stage and it does not introduce any constraint about the camera positions. The correspondence problem is solved only for few points (the barycenters of the detected moving objects) to obtain a 3D motion analysis of the moving objects. The first obtained results using two cameras seem to be encouraging.