LAUSR

ABSTRACT As a key step of video surveillance, motion detection attracts the attention of the scientific community. Detection and labelling dynamic objects are addressed in this paper. We cope with sudden illumination changes, the presence of shadow and partial occlusions. The main contribution of this work is the use of data: RGB image coming from an HD camera and depth coming from a Kinect, contour, movement resulting from the temporal inter frame difference and optical flow. The proposed 2D–3D combination achieves complete detection of moving objects. We carried out an experimental evaluation of the proposed approach on challenging dynamic scenes; we notice an increase in detection rate.