Environment Independent Key Frame Extraction and Moving Cast Shadow Suppression in Video Surveillance

Abstract

Video Key-frame extraction is one of the dynamic research issues in video arrangements and data recovery. Video Key-frame approximates the video frames in the arrangement which can be used to represents the lengthy video sequence, which reduces the time to process and summarizes the video sequence. The reason for Key-frame extraction from a video grouping is multifold: the main design is dimensionality reduction, second is a reduction in the memory required to store the information and third is the decrease in the many-sided applications to process the diminished data for various video examination applications. Key-frame extraction brings about proficient video summarization, video content analysis, data recovery and data handling. The present work has great scope for many computer vision applications such as Traffic Surveillance, Video Compression, Video Summarization, Security Surveillance, Automatic Object Identification and Tracking, Human Tracking, etc., newlineThis research work presents algorithm for background subtraction with a capability to adapt to scene changes over a time besides being invariant to sudden illumination variations. The proposed algorithm makes use of LBP as a feature set and Artificial Neural Network (ANN) to model the background frame which results in its being invariant to gradual illumination variations. To controls sudden illumination variation gradient background model with temporal information has been used. The presented background subtraction algorithm outperforms existing algorithms for standard video sequences and is demonstrated in the result analysis section. The presented algorithm achieves 85% and 0.009% of precision and false alarm respectively.In this work, hybrid shadow detection and suppression algorithm have been presented using both pixel and region based approaches. First, candidate shadow pixel using chromaticity is extracted, and then texture based approach is initiated to eliminate non-shadow pixels from candidate shadow pixels, thereby increasing the accuracy