Correlation Kalman Filter Based On State Noise To Video Target Tracking Technology Research

Video targets tracking is one of the hotspots in current domestic and foreign research. It is a technology which extract position information of targets from the image sequences in real time and track targets automatically. It has wide applications and broad prospects in military affairs, medical science, intelligent transportation system and security monitoring etc.In the course of maneuvering targets movement, the state noises are always both correlated. However, the correlation is often ignored and the noises are considered as white noises for the sake of convenience in the present study. The mismatch between actual states of the maneuvering targets and our system may easily cause targets lost during the process of maneuvering targets tracking. In such a context, the correlation of state noises for a video sample of helicopter model is studied in this paper, and the technology of video maneuvering target detection and tracking is also researched. The main works of this paper are as follows:Firstly, the Otsu method is used to obtain threshold, segment the frame images and get the binary images which contain the targets and background. Considering the complicated background and light change, the binary images always contain false targets. We use the method of connected domain segmentation to get ride of the false targets which are too big or too small.Next, considering the false targets in the binary images, a gray association and area association method based on image processing is given in this paper. Then the gray association, area association method and the nearest neighbor association algorithm combined with the techniques of estimation and prediction are used to extract the true targets from the binary images effectively.Then, based on the CV model, we study the correlation of state noises following two different probability distributions. One is acceleration state noises follow the Singer model, zero mean and uniformly distributed; the other is following the current statistical model, rayleigh distribution. We use the Kalman estimator algorithm with correlated state noises to estimate and predict targets' moving state. Compared with standard Kalman estimator algorithm which ignore the correlation of the state noises, the Kalman estimator algorithm with correlated state noises is proved to have less estimation error and prediction error. Meanwhile, on condition that there are cross correlation between state noises and measurement noises, we make tracking test on the video target. The results show that the cross correlation between state noises and measurement noises are not strong, which means the cross correlation has a tiny influence on the precision of prediction and estimation.Finally, the Kalman estimator algorithm with correlated state noises and the data association algorithm are used in practical video tracking system, realizing continuous and stable tracking, demonstrating validity of the algorithm in video targets tracking.