Research On Target Tracking Based On Human Visual Attention Mechanism

The continuous development of computer technology has promoted the widely used application of target tracking in human life.Target tracking,as one of the areas of application of pattern recognition,has become a hot topic for domestic and foreign scholars.With the advent of the information age,the amount of explosive information makes the target tracking environment increasingly complex.In the past,the algorithm has the problems of weak robustness,low tracking precision and poor persistence under complex conditions such as illumination change,multi-disturbance and target self-deformation.Aiming at the problem of poor tracking stability caused by excessive interference in the surrounding environment,this paper proposes an interested target tracking based on human visual attention mechanism and a multi-local-task tracking algorithm based on human visual memory mechanism.The main contents of this paper are as follows:Firstly,this paper studies the image preprocessing method and feature extraction technology involved in this algorithm.This paper introduces the image sampling quantization technique to preprocess the target image.It mainly improves the compression-sensing feature extraction technology and provides the theoretical guarantee for the tracking algorithm of the paper.Secondly,this paper proposes a target tracking algorithm based on visual attention mechanism.The algorithm is based on the classical visual attention mechanism tracking model,and the human visual nervous system is designed to construct the motor nerve pathway,the sensory nerve passages and the central nerve path to effectively extract and deal with the target characteristics.It uses center-edge difference algorithm to calculate Gaussian pyramid multilayer feature map in different features of each pathway and generates a general feature picture.The areas of different salience are calibrated by the "Winning" rule and determine the region with the highest degree of interest as the target area to achieve target tracking.In addition,in the system,the compression sensing module is added to monitor the output characteristics in real time,which solves the problem of tracking information loss and improves the tracking robustness and stability.Experiments show that the algorithm has a good effect in dealing with complex background and illumination change tracking conditions.Thirdly,aiming at the problem of target deformation,this paper proposes a multi-task target tracking algorithm based on target template library.In our formulation,the tracking task is decomposed into several local tasks by dividing the whole picture into several fragments,and the final tracking result is obtained by combining the local tasks.Using the multi-feature matching method to filter the "rough" task grid,and then filter out the sub-tasks with "edge extension" rules integrated into the overall tracking task to achieve "fine tracking".Furthermore,drawing on the memory of human visual and brain,we build a target template library to store different target state history information,and provide a template for each step of the tracking process.Experiments show that the method can effectively solve the problem of self-deformation and interference in the target tracking.Finally,the paper summarizes and prospects the further research.This algorithm can better deal with the target tracking task under various complex conditions and show good stability and robustness.However,when dealing with multiple target tracking tasks due to excessive computational complexity of the paper algorithm caused by real-time is not good,it will be the future research direction of the subject.