Non-linear Video Target Tracking Algorithm

Video target tracking is a very active research area in video and image processing. It has combined Pattern recognition, image processing, automatic control, artificial intelligence and computer applications technology, and so on. Its research results are widely used in aerospace, traffic surveillance and military guidance, and other fields. In recent years, many experts and scholars have been making a deep research and put forward many tracking algorithm. The aims of Video target tracking is to gain position information, motion parameters of the moving object or interested regional from each frame image. We have deeply studied the nonlinear video target tracking algorithm, and given the main research work and innovation points in this paper as below:1) This paper presents a new target tracking algorithm based on the particle filter and the edge orientation distribution to track video moving targets whose trajectory is nonlinear and in a complicated backgrounds. The edge orientation image in target area is obtained by edge operator and the edge orientation distribution is established by Gaussian kernel firstly. And then the edge orientation distribution is used as the characteristic to fuse with the particle filter to complete tracking for non-linear video moving target. The computer simulations show that the presented method can track the non-linear and non-Gaussian moving target in the complicated backgrounds effectively, and had a better performance compared with the method based on gray distribution.2) The IMMPF algorithm is a combination of the interacting multiple model (IMM) and the particle filter (PF); It can be adapted to the maneuvering targets'state estimation with high accuracy. The different sequence between the IMM and the PF leads to different IMMPFs frameworks. In the paper, we proposed three kinds of IMMPFs and compared their performance in the target tracking scenarios. The simulation results show that the IMMPF3 generates better performance in the targets' state estimation. Especially, when the number of samples of particle filter is large enough, the IMMPF3 behaves stable.3) In the last part of the paper, we presented a new tracking algorithm to track low-SNR weak dim targets at a reduced particle cost, Multi-Rate Unscented particle filter TBD (MRUPF-TBD). Current Unscented particle filter track-before-detect (UPF-TBD) algorithms update the entire particles set at every time to estimate the state. When the target has small maneuvers or motionless, it is unnecessary to update the entire particles set. The MRUPF-TBD algorithm extends the multi-rate into the UPF-TBD algorithm. The number of the particle of full rate, or 1/3 rate is give according to the maneuvers of the target. The computer simulations show that the presented method has a better performance then current UPF-TBD.