Project 44: Performance Evaluation of Visual Tracking on Encoded Videos
Contact Information:
Prof. Hongkai Xiong
Email: xionghongkai@sjtu.edu.cn
Project Description and Objectives:
In computer vision, visual tracking deals with non-stationary image streams that change over time, intending to detect and locate moving objects. It can be used in various multimedia applications, such as visual surveillance, human-computer interaction, augmented reality, and visual servoing control (a control task based on computer vision data). In some applications, the video has to be encoded and transmitted to another end for implementing the tracking task, the performance of which might be affected by the distortion introduced by video encoding. Therefore, it would be interesting to see the visual tracking algorithms’ performance under differently encoded video versions, which will further affect the control accuracy of the visual servoing control tasks.
Eligibility Requirements:
Basic knowledge of video codec (e.g., H.264 and HEVC), control theory and computer vision.
Mastering more than one programming language, C/C++ and MATLAB preferred.
Main Tasks:
Create a library of distorted video versions by choosing different combinations of critical encoding parameters.
Use the created library to test the performance of some of the selected tracking algorithms under differently encoded video versions.
Try to understand and explain the performance curves that are obtained from the previous step based on the implementation detail of each tracking algorithm.
Website: