Data association for multi-object tracking using weighted paths

摘要

In this paper, we present a novel algorithm for color based motion detection which also deals with occlusions, background clutter and multi-object scenario. This technique is applicable on both grayscale and colored images and can be applied with many object detection algorithms. The target is represented by spatio-spectral information. Then Probable targets are first identified using some detection scheme and all possible trajectories are formed between these objects. Motion detection is carried out using a probability data association filter (PDAF) like scheme. This results in detection of objects with their estimated trajectories.