The Research And Novel Improvements On Background Subtraction For Motion Detection In Surveillance

Detection of moving targets in video streams is the essential technique in intelligent video surveillance. Background subtraction algorithm is the most popular solution in motion detection, and is the most active computer vision research.In this thesis, we analysis the background subtraction in depth, and combined with the algorithm implementation, comprehensively summaries and analysis the strengths and weaknesses of a variety of background subtraction algorithms. We propose some solution for several typical difficult problems in the background subtraction algorithm:1) For the bootstrap of the algorithm, we proposed empty background extraction algorithm, which can even handle the complex scene; 2) For the foreground segmentation part, we introduce the Graph-Cut and SVM, take full advantage of the time and spatial information of the video sequence. The segmentation results was improved significantly; 3) For the shadow suppression problem, we combine the texture and probability features to improve the result of shadow suppression.In this thesis, we implement a person-counting system, which based on background subtraction algorithm. This system uses the background subtraction as a basis to detect the movement area in the scene, and use the blob segmentation and tracking algorithm to achieve real-time people counting system.