Camera motion compensation from T-junctions in distance map skeleton

In the field of aerial surveillance, tracking targets in images is complicated by the possible motion of the camera, especially if frame differencing is used to detect moving objects. We propose in this paper to exploit the high similarity in sequences acquired from a nearly static camera. In this case distance maps grown from image edge points share many similarities and T-junctions of distance map skeletons appear to offer precisely located reference points. Each T-junction is attributed seven features: the value in the distance map, three orientations of the junction branches and R, G, B image intensities. Registering images is carried out on a division of the images into tiles, looking for the dominant translation per tile of matching T-junction points. The obtained displacement field allow for the compensation of small camera motion. This was tested on image sequences captured by a smartphone held in hand while targeting a given static scene with a few moving vehicles and pedestrians.