Interest Point Detection in Videos Using Long Point Trajectories
Rahul Nallamothu, T. Vineeth, Gaurav Kumar Yadav, Amit Sethi, Tony Jacob · 2014
Interest points in videos are spatio-temporal points within a video that can be used to build complex video processing systems such as action recognition and video retrieval systems. In this paper, we propose a method to localize spatial interest points in a video in the temporal domain, and discuss the results obtained with our experiments in comparison to the popular traditional handcrafted interest point detectors under various transformations. Our scheme is primarily based on the extraction of dense trajectories of objects in videos obtained through optical flow computations. The Ramer|Douglas|Peucker(RDP) algorithm is used on the dense trajectories in order to extract key points that are localised in the temporal dimension.