The Algorithm Research Of Moving Object Detection Based On Multiwavelet Transformation

The thesis focuses on moving object detection for sequence images. It means to discover foreground object that exist relative motion with background from sequence images, and further to extract the independent moving object according to two-dimensional character of images. The moving object detection not only has important theory value in computer vision domain, but also has huge application potential in aspect of intellectualized monitoring etc.The thesis accomplishes moving object detection by frame difference combine with multiwavelet transformation. The general process to detect moving object by sequence images' frame difference can be summarized as: sequence images' first order difference , object pre-alarm, difference image denoising and binaryzation, binary difference image symmetrical difference and mathematics morphology processing, evaluation of object area's related parameters and original image's mark.The thesis gives a method of moving object detection according to above process, the main innovation of the article lies in: In aspect of moving object pre-alarm, the moving object pre-alarm method based on the combination of frame difference and the background model is proposed. The method first carries on pre-alarm processing to obtainable sequence images, and on this basis once detects object appears, carries on moving object pre-alarm processing based on the background model object. If they all detect the object, then we can confirm the object appears. In aspect of difference image noise elimination, the method combine with median filter and multiwavelet sub-image revise is used. We first use the median filter method eliminate partial noises and carries on necessary preparation for following processing, and then eliminate surplus noise use the adaptive threshold character of multiwavelet sub-image revise, so we can make condition for object area extraction.Through simulation by using MATLAB, the result indicates that the method of this thesis can well detect and extract moving object from sequence images.