An algorithm for moving semantic objects trajectories detection in video

Abstract

Moving video semantic objects trajectories detection is an important step in the video analysis; this article analyzes the existing algorithms of semantic video object detection, which are not suitable to generic video. The proposed algorithm based on the relative location of the semantic object, according to the principle of relative motion of objects in the video, we calculate the relative distance between various semantic objects shown in the video, and then we design an object-location database, which is used to store the information of objects. Based on the database, we get the trajectories of objects. The algorithm solves the problems which are caused by the camera movement. The results of experiment show that the algorithm is suitable for general video object trajectories detection and enhance the applicability of object detection.