Research On Multi-Marker Based Visual Tracking Algorithm

In recent years,vision-based target tracking techniques have been rapidly developed and widely used.Although a large number of research results have been achieved in this field,there are still many challenges: for example,targets do not have texture information,targets Poor tracking,target is featureless,target is partially or heavily obscured,target is out of tracking field of view,scene lighting conditions The changes of the target,real-time requirements,etc.,all increase the difficulty of target tracking.For this reason,this paper mainly focuses on targets that are heavily obscured,targets that are not in the field of view,targets that are too large to track,targets that are not characteristics,etc.,a method based on multiple marker tracking is proposed,where markers are affixed to the target and around the target,and the tracking of the target is put into Switch to a method of tracking the surrounding markings.The problem of bad target tracking is solved by the method of sweeping the camera around the markings of the target.In this paper,based on the summary and analysis of previous work,we propose a multiidentification target tracking algorithm based on the following main studies It reads as follows:1)The latest domestic and international research results based on multi-marker target tracking algorithms are collated and summarized,and a comprehensive overview of the technological development in this field is provided.Based on the introduction of the principle of this class of algorithm,the focus is on the marker-based positioning tracking algorithm,and the different kinds of marker such as ARToolkit,ARToolkit Plus,ARTag,April Tag,etc.are introduced in depth,and the different kinds of marker are discussed,and the tracking effect in the presence and absence of the marker in the case of masking is compared.In a combined comparison,April Tag outperforms several other types of markings in terms of detection rate and accuracy.2)In-depth introduction of the main key techniques of the identification selected in this paper,the principle of encoding and decoding of the identification,and the improvement of the identification detection algorithm,in the original identification detection algorithm,the Sauvola algorithm in the local binary method is used,because the Sauvola algorithm is strict on the selection of the template window size,the effect on noise suppression is not ideal,the identification details information will be lost.In this paper,the Canny operator was selected to improve the image detection results by a non-extreme value suppression method to achieve the lowest possible detection of noise.3)In this paper,the identification method is based on the camera coordinate system for positioning.Due to the limitations of the Pn P(Perspective-n-Point)algorithm used for identification,the positioning error of the camera becomes larger as the distance between the camera and the target coordinate increases.This paper addresses this issue by adding a camera a priori posture.4)Based on the in-depth study of the logo-based tracking algorithm,a visual tracking and localization approach based on neighborhood multiple logo confidence assessment is proposed.The main point of innovation is to convert the tracking of the target into a method of tracking the markings around the target.During the tracking process,the final confidence level of each identity is obtained by detecting the extracted neighboring identity features,and then combining the normalization results of the Hamming distance of each identity with the normalization results relative to other identities,as well as the probability of occurrence of each identity,and then selecting the neighboring identity with the highest confidence level to determine the location of the target to be tracked.In order to avoid the failure of tracking when neighboring identifiers are out of sight or for other reasons,multiple identifiers are used to track together.For this purpose,multiple marker groupings were used to determine the final target location by tracking multiple neighboring makers and assessing the confidence level.