Real-time segmentation and tracking module of target of interest from video sequence in object recognition systems

Abstract

This paper proposes a real-time objects segmentation and tracking module from video sequences, which can be effectively used in real-time object recognition systems. The module is based on background subtraction method in combination with CAMshift (Continuously Adaptive Mean shift) algorithm. In the first step, background subtraction method is applied to determine pixels of moving objects in video stream. Then, foreground pixels are used as starting point for CAMshift algorithm. CAMshift finds optimal size, position and orientation of moving objects. After that, key frame extraction method is applied in order to choose only relevant frame in later objects classification.