Spatio-temporal interest points matching in video

Despite the wide applications of matching descriptors, much work hasn´t been done on it. In the current paper, we extended the use of the existing techniques in object detection in 2D images to video sequences by extracting the Spatio-temporal Interest points using spatio-temporal Harris detector. Features descriptors are computed on the pixels around these interest points. To match between descriptors, we build a KDTree using nearest neighbor search.