Foreground segmentation is often an important preliminary step for various video processing systems. By improving the accuracy of the foreground segmentation process, the overall performance of a video processing system has the potential for improvement. This work introduces a Fuzzy Foreground Segmentation System (FFSS) that uses Mamdani-type Fuzzy Inference Systems (FIS) to control pixel-level accumulated statistics. The error of the FFSS is quantified by comparing its output with hand-segmented ground-truth images from a set of image sequences that specifically model canonical problems of foreground segmentation. Optimization of the FFSS parameters is achieved using a Real-Coded Genetic Algorithm (RCGA). Additionally, multiple central composite designed experiments used to analyze the performance of RCGA under selected schemes and their respective parameters. The RCGA schemes and parameters are chosen as to reduce variation and execution time for a set of known multi-dimensional test functions. The selected multi-dimensional test functions represent assorted function landscapes. To demonstrate accuracy of the FFSS and implicate the importance of the foreground segmentation process, the system is applied to real-time human detection from a single-camera security system. The Human Detection System (HDS) is composed of an IP Camera networked to multiple heterogeneous computers for distributed parallel processing. The implementation of the HDS, adheres to a System of Systems (SoS) architecture which standardizes data/communication, reduces overall complexity, and maintains a high level of interoperability.
Library of Congress Subject Headings
Video recordings--Data processing; Image processing--Digital techniques; Classification--Data processing; Genetic algorithms
Department, Program, or Center
Electrical Engineering (KGCOE)
Bowen, Ryan, "Implementation, integration, and optimization of a fuzzy foreground segmentation system" (2013). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus