A Research Of Objects Tracking Based Upon Kalman Filtering And Template Matching

Visual information plays an important role in the process of sensation world by human. We can acquire a plenty of useful information by detecting, tracking and analyzing moving objects in the video image, such as, in the military opposes, we may enhance the army long-distance attack ability by detecting and tracking moving objects of enemy; to enhance the safety coefficient of Intelligent Residential District by detecting and tracking interloper (for example a theft and so on); detection and the recognition risk behaviors of important places, airport and bank as examples, can protect people's personal safety and property. Obviously, the moving object tracking is an important research subject.This paper mainly studies on object detection, object shadow removing and object tracking methods which involved in the process of moving object tracking, and corresponding experimental results show that the research in target detection, shadow removal and Tracking achieved some purpose.The main work of this paper can be generalized as follows.(1) We have conducted a comprehensive study on motion detection with Gaussian mixture model. Against the problem of slowly update speed of background, using a method of Gaussian mixture model with internal update strategy of background. The method update background once every M frames, the detected motion pixels are not updated and the detected no-motion pixels update with the mean-value, it reduces the complexity of algorithm and speeds up the update rate of background. Experimental results show that the algorithm can detect more complete binary image of motion object with the case of reducing algorithm complexity.(2) Removing object shadow from color and texture attributes. This paper mainly study a method of shadow removing based upon rg/V values of pixel points from color and study a method of shadow removing based upon improved LBP texture with regard to the problem that the basic LBP texture shadow removing exist. Through several experiments show that the two methods of shadow removing which study in this paper can effectively remove the object shadow.(3) A method propound combines two methods to remove shadow of motion objects was proposed. Both of the shadow removal methods appear litter holes (pseudo shadow points), even spilt the object image, a method that combine two methods with two binary images which removed shadow to "OR" operator, using the feature that the same pseudo shadow point is difficult to remove as shadow point, remove holes in the object by "OR" operate. Through several experiments show that the "OR" method can extract more complete object than either of them.(4) Tracking motion objects use a method of motion prediction using Kalman filtering and template matching. The method can track motional object better after removing shadow effectively. By two groups of tracking experimental results show that the method which adopted in this paper can effectively track motional objects.