Research On TLD Tracking Algorithm And Its Implementation In Bionic Eye System

Bionic eye technology has the ability to simulate the biological eye to sense the external environment,so it has a very wide application prospect.While,wide-area monitoring can solve the problem of low efficiency and improve the protective efficiency of current video monitoring.Therefore,this paper takes the application of bionic eye technology in wide-area monitoring as background.It has important theoretical value and application significance to research on motion tracking algorithm in bionic eye system.Tracking-Learning-Detection(TLD)algorithm is an online learning tracking algorithm.It is able to achieve long-term tracking of unknown target and has the ability to re-detect the lost target.So this paper adopts TLD algorithm as the tracking algorithm of bionic eye system.This paper focuses on the research and improvement of TLD algorithm and its application in bionic eye system.The main research work is as follows.(1)Aiming at the problem that the detection area of TLD algorithm is too large to affect the real-time and accuracy of the algorithm,this paper integrates Kalman filters into TLD algorithm to solve this problem.By predicting the target position,narrowing the detection area and reducing the number of rectangular box,the improved algorithm improves the real-time performance and the identification of similar objects,which improves accuracy of the algorithm.The experimental results show that the improved algorithm integrated with Kalman filters can improve the real-time performance and accuracy of TLD algorithm effectively.(2)This paper aims at the problem that there is no limit on the samples number of online model in TLD algorithm,resulting in the increasing number of samples in the process of long-term tracking,which affects the real-time performance of the algorithm.A new online model updating strategy is designed to solve this problem.The improved algorithm can ensure the real-time performance of long-term tracking and improve the practicability of the algorithm through setting a reasonable threshold for online model to limit the number of samples.The experimental results show that the improved algorithm can improve the tracking real-time performance of long video obviously.(3)The variance classifier uses half of the initial target variance as fixed filtering threshold,which can cause the defects of misjudgment and poor filtering effect.In order to solve these defects,a TLD algorithm with adaptive updating threshold is proposed in this paper.By updating the filtering threshold dynamically,the improved algorithm can avoid misjudgment and reduce computing time for subsequent classifiers,thus improving the accuracy and real-time performance of algorithm.The experimental results show that the improvement of adaptive updating threshold can improve the tracking accuracy and real-time performance of TLD algorithm obviously.(4)The TLD algorithm synthesized by three improvements is applied to bionic eye system.Besides,the algorithm flow of binocular bionic eye system is improved by introducing epipolar constraint in this paper.By making full use of the redundant information of binocular system image to coordinate the TLD execution process of two cameras,the improved algorithm flow improves robustness and real-time performance of the system.Finally,this paper builds a multi-camera group system based on bionic eye system,and realizes the video monitoring and analysis under wide-area.