Inference of Non-Overlapping Camera Network Topology using Statistical Approaches

This work proposes an unsupervised learning model to infer the topological information of a camera network automatically. This algorithm works on non-overlapped and overlapped cameras field of views (FOVs). The constructed model detects the entry/exit zones of the moving objects across the cameras FOVs using the Data-Spectroscopic method. The probabilistic relationships between each pair of entry/exit zones are learnt to localize the camera network nodes. Increase the certainty of the probabilistic relationships using Computer-Generating to create more Monte Carlo observations of entry/exit points. Our method requires no assumptions, no processors for each camera and no communication among the cameras. The purpose is to figure out the relationship between each pair of linked cameras using the statistical approaches which help to track the moving objects depending on their present location. The Output is shown as a Markov chain model that represents the weighted-unit links between each pair of cameras FOV