Background Learning with Support Vectors: Efficient Foreground Detection and Tracking for Automated Visual Surveillance

• Published in: Background Modeling and Foreground Detection for Video Surveillance pp. 293 - 316
• Format: Book Chapter
• Language: English
• Published: Chapman and Hall/CRC 2015
• Subjects: SVDD; Adaptive Threshold; Data Set; Target Detection; Background Pixel; Lagrange Multipliers; Statistical Background Modeling; Label Pixels; Kernel Density Estimation; Visual Surveillance; Training Stage; Training Samples; Non-parametric Density Estimation; Estimated Probability Density Function; Foreground Detection; Kernel Bandwidth; Background Modeling; Non-parametric Kernel Density Estimation; Pixel Models; Training Frames; AKDE; Foreground Regions; Default Coordinate System; SVR Approach; RM
• DOI: 10.1201/b17223-19

With the increase in the availability and computational power of digital imaging devices, it is natural to think about integrating artificial intelligence models with processing of digital images and videos used for visual surveillance applications to improve their performance. However, for such applications to be efficient, there is a need for addressing a number of significant challenges. Accounting for the presence of regions that do not belong to objects of interest, global ambient illumination variations in the environment over long periods of time, and the real-time constraints inherent to visual surveillance applications are among such obstacles. This chapter demonstrates two main categories of mathematical and computational techniques developed to help improve the accuracy and efficiency of automated visual surveillance systems.