Detecting people carrying objects based on an optical flow motion model

摘要: Detecting people carrying objects is a commonly formulated problem as first step to monitor interactions between and objects. Recent work relies on precise foreground object segmentation, which often difficult achieve in video surveillance sequences due bad contrast of the with scene background, abrupt changing light conditions small camera vibrations. In order cope these difficulties we propose an approach based motion statistics. Therefore use Gaussian mixture model (GMMM) and, that model, define novel speed direction independent descriptor detect carried baggage those regions not fitting description average walking person. The system was tested public dataset PETS2006 more challenging including lighting changes color compared existing systems, showing very promissing results.