Behavior Recognition Method For Intelligent Multi-mode Research

Intellingent video tracking is a new arising research task in recent years,which integrates technologies of computer vision,pattern recongnition,artificial intelligence and so on Combining image processing, automatic control and information technology, target tracking has developed to be an advanced technology, which can automatically distinguish objects in real-time from the video image, acquire and predict the target position information, and track the target movement.It has a lot of potential applications in security surveillance,intelliance traffic system,video compression and index,etc.In this dissertation,basing on studying algorithms of moving objects detection and moving objects tracking,a set of intelligence video tracking has been designed and realized.On the research of moving object detection ,an improved background subtraction algorithm based on Gaussian model has been proposed,and a hybrid detection framework by combining Gaussian model and frame differencing algorithm has been presented. And test results for the Gaussian model the impact of the vulnerability of the shadow of the problem, through light and color characteristics of model analysis, to improve the HSV color space based on shadow detection algorithms.On the research of moving object tracking ,the Kalman filter calculate fast and properly,but it asks the object has linear-Gaussian moving characteristic(eg,in the short time,the object moving is uniform motion).After Comaniciu thought out the Mean Shift algorithm, the method can depend on target area and getting gray distribution and compute the displacement from target center,which can be fit to fast target movements and the shape or lighteness change frequently.For the characteristic of video itself, the Mean Shift algorithm used in the thesis had been made some improvement. In this thesis,we choose the Epanechnikov kernel in the experiment.To compare with model's density estimation and candidate's, calculating the new location of the target.Experimental results show that the proposed algorithm can track the moving target in video efficiently and precisely, and also can meet high real-time situation for its small calculation.Finally,a motion detection and tracking experimental system with complicated background conditions is designed,which provides an experimental flat to validate relative algorithms.It specifies the main frame of system,the functions of every model,estimating methods and limitations.The experimental results indicates that the system is certainly valuable in stated range.