Soft-biometrics based people description in surveillance systems

• Main Author: Ong, Glen Hong Ting
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/60197
• Institution: Nanyang Technological University

In the modern days, the increasing reliance on surveillance infrastructure has created uproar in many Computer Vision applications. With the current technology, most users employ traditional surveillance systems which could only provide record and playback of any saved footages. As such, it would be extremely time-consuming for a person to view intensively and identify subjects of interest from the lengthy footages. In this project, the use of Support Vector Machines (SVM), Soft Biometric traits and other computer processing applications would extend the level of surveillance searching which provides an accelerated search or identification of the subjects of interest from the saved surveillance footages. This report describes the review of literature material related to the methods adopted in this project as well as the implementation of the combined ideas such as the SVM classifier and soft biometric traits to obtain a ‘search and identify’ function based on the descriptions in the area of surveillance security systems. Three soft biometric traits are selected as the target traits of an individual to form a unique signature for subjects. The selected traits are the presence of spectacles, cap and the pattern of top worn by subjects. As this project is required to detect these traits, three different classifiers are needed. Images are trained and tested based on the different types of body part extractions (specifically the head and torso areas) and various normalization techniques and procedures. The best techniques used for each trait group which yields the highest classification rate will be employed for the detection of the respective trait of subjects. The best techniques used for detecting each trait are used to construct the classifiers for the detection of the three mentioned traits. With these classifiers constructed, a subsystem of a proposed search system is built. This subsystem, namely the human description system, is responsible for the detection of subjects based on the descriptions given by the user. This subsystem is then integrated with another subsystem (human detection system) designed and built to extract all Abstract 2013/2014 Soft Biometrics Based Human Description in Surveillance Systems ii subjects from their background by my project mate, Qi Xiang. On top of that, Graphical User Interface (GUI) is constructed. This GUI is built to ease the process of searching which bridge the link between the programming interface and the user. An evaluation of the project is conducted at the end of this project and will be presented at the end of this report.