The Systemic Research Of Moving Object Detction Algorithm In Image Sequence

Moving object detection in complex environment is a basic problem in Computer Vision, Because of its good prospect and high application value. At the present time, the theory of moving object detection has been used widely in many field, such as military spy, television surveillance, traffic detection, medical research, experimentation measure and so on.This thesis is focused on moving object detection algorithms in complex environment, and it mainly includes 4 sections:1. The basic algorithms of moving objects detection, such as adjacent frames difference algorithm, Optical Flow algorithm and background subtraction algorithm, is analyzed, and the implement result of these algorithms with some common video sequence is given. Some improved algorithm, which can improve the deficiency of basic algorithms, is also researched.2. A new method for moving target detection based on positive and negative difference maps is presented. Compared with the basic frames difference approach, it gets two difference maps-positive difference map and negative difference map at the same time. Making use of the relativity of the two maps, the moving objects can be detected easily and exactly. In this thesis, we disused the basic principle, implement steps and the experiment result of the algorithm.3. The new object detection algorithm based on positive and negative difference maps is integrated with object tracking algorithm. This new object-tracking algorithm can detect and track objects automatically.4. An object detection method with EM (Expectation Maximum) algorithm of dynamic layer representations is researched and improved. Previous algorithm contains Optical Flow computation, affined transformation, and clustering algorithm, and it is not convenient for detecting object quickly. So we use Blob-Matching algorithm to get motion parameters and make use of EM algorithm to classify these parameters. These algorithms do not use Optical Flow computation and affine transformation, so it smoothes the difficulty of calculation.5. The prospect of moving object detection algorithm is analyzed simply.