Video analytics for intelligent surveillance system applications

Abstract

Video analytics has significantly received a consideration, especially in surveillance applications such as helmet wear analysis, stabbing action detection, person re-identification and face detection. Intelligent surveillance systems are important for these applications, and this is therefore considered to be the primary motivation for this work. This thesis focuses on surveying cast shadow detection methods for moving object differentiation in surveillance environment which are very essential for higher level analysis such as helmet wear analysis, stabbing action detection, person re-identification and face detection. Further, helmet wear analysis using deep learning approaches in surveillance scenario for life-saving and security is considered. The need for an hour to mitigate the effects of irregular and inappropriate behaviors in public places such as banking, airport and public gatherings is a secure and intelligent abnormal activity surveillance system. Hence, stabbing action detection for ATM surveillance applications is proposed. Consequently, for the identification of the culprits, involved in crimes, person re-identification by means of traditional approach and improved face detection, using proposed de-blurring technique is performed. A complete overview of the literature on cast shadow detection for segmentation of moving object against varying illumination conditions has been addressed in detail. The algorithm flexibility to challenging situations such as real time, illumination changes, shadow and ghost, has been elaborated. The detailed discussion of various methods handled by the existing algorithms includes chromaticity, physical, geometry, and texture, and MFF with respect to cast shadow detection for foreground segmentation in gradual illumination conditions. However, this summary is significant, when it categorizes different methods and it also offers a suitable algorithm for each specific dataset. This will support the researchers in selecting the suitable method for various applications lik