Research On Anti-occlusion Algorithm For Video Target Tracking

Video tracking technology needs to deal with many problems when it is used to track the scene in real life,and the research on the tracking of the obstruction target has become one of the current hot spots.When the target is occluded,the tracking algorithm based on the apparent information of the target tends to produce a large tracking error,resulting in the missing target.In such video occlusion target tracking,it is important to perform state prediction on the occluded target,thereby improving the anti-occlusion performance of the tracking method.The establishment of maneuvering target mathematical model provides strong support for occlusion prediction.This paper integrates the target movement law and tracks the actual situation of the scene,and establishes the state space model of the maneuvering target.According to the different physical characteristics of the target motion,the target state space model is accurately expressed mathematically.Then the state space model is applied to the prediction of the motion state of the target,and the state prediction method based on Kalman filter and the trajectory prediction method based on radial basis neural network are proposed.Aiming at the problem of online tracking when the target is seriously obstructed,this paper presents a prediction-rematching target tracking strategy.This strategy solves the poor applicability of Mean Shift and its improved algorithms.Firstly,the Mean Shift and Kalman filter are dynamically combined to achieve stable tracking when the target is unoccluded.Secondly,when the occlusion occurs,the state of the occluded target is predicted from the target historical information using the Kalman filter;finally,the occlusion In the process of goal recurrence,the target optimal position is obtained through the normalized cross-correlation matching strategy,and then the target can be quickly and accurately positioned after the recurrence of occlusion to continue effective tracking.This method is fast and efficient,it is easy to operate and has error correction mechanism.It can guarantee the tracking performance after the occlusion target reappears,and validates the superiority of the algorithm through simulation.The Kalman filtering prediction process only uses the information of the previous state of the target,while the training and learning of the radial basis neural network is more fully utilized for the target historical information,and can more accurately reflect the target movement law and track the actual situation in the scene.The neural network prediction strategy came into being.Firstly,the volumetric Kalman filter is integrated into the parameter determination process of the Newton-Gauss iteration strategy to form a new iterative learning training strategy.Secondly,a new learning training strategy is introduced in the radial basis neural network to improve the recent state.Weights influence and further improve the accuracy;Mean Shift is combined with a mature and mature improved radial basis neural network to apply to occlusion video target tracking.This method can make full use of the target known information,improve the information utilization rate,and adopt local iteration to further improve the network training accuracy and analyze the stability and accuracy of this method in the actual tracking scene.