People tracking in video

• Main Author: Goh, Angela Shi Yun
• Other Authors: Cham Tat Jen
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/59126
• Institution: Nanyang Technological University

This report is a documentation of the development of a robust real-time visual tracking system using pixel-wise posteriors. Detailed explanation on this method will be done in the subsequent chapter. This system tracks an object or a person through a video camera by using a computer program written using matrix laboratory (MatLab). This tracking system was primarily designed for robust, real-time, visual tracking of previously unseen objects from a moving camera. The tracking system has four main sections mainly representing each frame with a simple generative appearance model and includes a rigid registration of frames, segmentation and online appearance learning. The key to the success of this method is the use of pixel-wise posteriors, as opposed to likelihoods in many other methods of tracking. Furthermore, various methods such as zero level set embedding function, signed distance function, logarithmic opinion pool for fusing and introducing a wrap were looked into and implemented as mentioned in the proposed paper [1].