Object Recognition: Human Detection, Tracking and Counting

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Title: "Object Recognition: Human Detection, Tracking and Counting"

by

ZHENG, Zichen

Abstract

Developing an efficient and effective human detecting, tracking, and 
counting system is a challenging and hot topic in computer vision area, 
since it utilizes various computer vision techniques, such as feature 
extraction and feature detection; as well as machine learning techniques, 
like dimension reduction, clustering, and classification.

In this report, we describe methods to achieve human detection, tracking, 
and counting respectively. These methods are combination, adaption, and 
enhancement of existing algorithms. Specifically, we trained our own human 
head detector by the histogram of gradients (HOG) descriptors. And we 
implemented the mean-shift procedures and the adapted kernel based 
compressive tracking method to track people. Further, we devised a method, 
which applied human detection and tracking, to count people appearing in a 
video sequence.

OpenCV with C++ interfaces were used to develop the system, as OpenCV is a 
powerful cross-platform library for real-time computer vision.


Date            :  7 May 2013 (Tuesday)

Time            :  10:30am to 11:30am

Venue           :  4475 (lift 25-26)

Advisor         :  Prof. Quan Long

2nd Reader      :  Dr. Huamin  QU