Research On Tracking Registration Algorithm Based On Natural Features

Augmented reality technology is mainly used to superimpose and integrate fictional images,video,audio and other data information with concrete things,so as to achieve the purpose of enhancing the effect of the real environment.Nowadays,science and technology are constantly advancing and people’s ability to study augmented reality technology is gradually improving,and it has been more maturely applied in all aspects of people’s lives.As an important technology that affects the effect of augmented reality,tracking and registration techn ology is the basic condition for realising the integration of reality and reality,and is also the key to determining the performance of augmented reality systems.As one of the hotspots of augmented reality research,this paper focuses on the study of nat ural feature-based tracking and registration technology.The main research contents are as follows:(1)Study of feature point extraction and matching problem: A combined approah based on morphological retinal key point descriptor(MREAK)and ORB algorithm is proposed to address the problems of ORB algorithm such as unsatisfactory matching accuracy and inability to cope with scale size changes.Firstly,the FAST algorithm with increased orientation calculation is used to evaluate the feature points of the target image,and secondly,the MREAK descriptor is used to describe the extracted feature points,which further improves the precision of feature matching while ensuring the speed of description.Finally,a combination of Hamming distance and PROSAC algorithm is used to reject the incorrectly matched feature point pairs,thus guaranteeing the accuracy of the algorithm.Experimental results show that the improved algorithm has better robustness in the presence of changes in the external environment.(2)Research on target tracking registration problem: In response to the problems of the TLD target tracking algorithm such as tracking punctuality,unsatisfactory accuracy and sensitivity to lighting changes and easy target loss,this paper proposes to improve the first two modules(tracking and detection)of this tracking algorithm.Firstly,the histogram of the detected video frame sequence is enhanced with a balanced histogram after a hue mapping change and combined with median filtering to reduce the noise of the image.The KCF algorithm is then improved by using the fused features of the Histogram of Orientation Gradients(HOG)and colour features(CN)after dimensionality reduction and replacing the original tracking algorithm in the tracking module with this algorithm;secondly,in the detection module,the Kalman filter is introduced to predict and correct the target position,and the variance classifier in the three-stage cascade classifier is improved by adapting its fixed The threshold value is adaptively adjusted and improved.Experimentally verified,the improved TLD algorithm not only effectively improves the speed and accuracy of target tracking,but also solves the problem of tracking failure after target disappearance.