Colour profile identification for networked surveillance tracking systems

• Main Author: Yeo, Kenneth Kai Xiang.
• Other Authors: Chan Chee Keong
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision; DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/39419
• Institution: Nanyang Technological University

In the advancement of digital technology today, current CCTV visual security surveillance systems with eventually be replaced by IP networked based visual security surveillance systems thereby paving the way for automated people surveillance tracking and recognition systems, which allow easy integration into the network. Current human recognition systems rely on face recognition algorithms such as Viola-Jones face detection and Principal Component Analysis (PCA) but do not consider the combination of coloured outfits that people wear. The proposed solution is to design the Intelligent Tracking Unit (ITU) to be installed in the Intelligent Network Location-Tracking system, which can be easily connected to any network in order to automatically track and recognise people. This report introduces the colour matching ability of the ITU to simulate tracking of a coloured test picture which resembles people wearing various coloured outfits. The study conducted prior to the development of this algorithm on the BGR and HSI components of the colour model revealed that the Hue values are closely correlated to the colour of the test paper, provided that the colours were not greyscale. Another finding revealed that the Hue values had the least standard deviation as compared to the BGR and HSI colour components. This made it favourable for the development of the Hue Clustering algorithm, which takes a background masked image and transforms it into a Hue Profile consisting of a series of Hue Cluster Centroids. It was also found that although the BGR values had reached their upper or lower limits, the inaccuracy of hue values were insignificant. Experiments conducted using the Hue Clustering algorithm yielded expected results which conformed to the intention of the design. Optimal values for the Hue Clustering algorithm were also determined to ensure the correct number of colours detected. Finally, the Hue Profile Comparator algorithm was designed and implemented to return a percentage match point for two Hue Profiles. The experiment was conducted for the cameras which had their field of view directed at the same image, and the results yielded expected values of 95.91% match point, confirming the successful implementation of this algorithm.