A Review on Classifications of Tracking Systems in Augmented Reality

Augmented Reality (AR) is now becoming an exceptional technology that offers a new world. Users to enable their senses to feel, listen and see the surroundings in different and better ways witness a unique experience. AR technology is used to superimpose the real view of the user with the virtual scenes. This research work illustrates the core details of AR, its definition, history, and development process. In addition, discusses ideas having various approaches during utilization of AR frameworks along with glyph tracking system. Current applications of AR technology completely depend on the proper utilization of AR frameworks.


Glyph Recognition Technique: AN Overview
Glyph recognition is a well acclaimed technique for character recognition by addressing the problem present in augmentation. Specifically, started with a simple approach where the focus is on integrating and observing the randomness in glyph into data augmentation of character images. Existing techniques enhance the size of data from distorting images. In addition, some methods are based on injection of noise into glyphs [1][2][3]. It results the radical variation in data present in glyph. Some techniques are used to exploit the glyphs as public database for augments glyphs and kanji by injecting noise into glyphs.
For the expansion of datasets, data augmentation method is widely accepted for better outcomes. Current methods are majorly based on image processing techniques to obtain data augmentation. In image processing [6], they take original images and generate new images with the help of source images by carrying out flipping, cropping, noise injection and horizontal elastic distortion. One of the popular techniques is also available for deep neural networks, to generate models for data augmentation system. Still, some of the demerits had been noticed in the use of such techniques such as these methods [7] only considered the variation of images. They were not giving the importance to the variation of glyph like width of strokes in character, angles and shapes of glyph [8].
The procedure starts with glyph segmentation, which is a more challenging task as compared with segmenting or labeling the natural images[9 due to certain factors such as: Unfamiliarity: It is generated when the participated crowd might have never seen or familiar with the ancient writing system before. It becomes hard when the interaction has been made between humans with their surroundings. The whole procedure depends on the intuition based for object categories that are much unseen. Then on the hit and trial method, some similarities have been made with the already known objects for achieving the outcomes. Visual Complexity: Each language consist uniqueness in it that is hard to read and visually complexed e.g. Maya language which is very famous as ancient writings. Another thing is that each community used different glyph for representing their language like Egyptian hieroglyphs are of well-separated glyphs as compared with other language. It makes the procedure quite complex as some inner details and many deformations are not always visible. Uncertainty: Some signs and their characterization consists several uncertainties. The main reasons behind such uncertainties are incomplete understanding of changing angles and shapes of signs or severe damage present in the cryptic coming from different places. These are also belonging from different eras, so unclear relationships are noticed due to the presence of frequent signs and new symbols.
Glyph dataset are generated so that user can enable the learning process for robust shape representation specially designed to perform automatic recognition tasks. In place of manually performing the operations from catalogues, new glyph samples are faster for experts. They can obtain data with reasonable accuracy by utilizing retrieval tools and an automatic classification. In addition with such tools, obtained shape representations can be used for the measurement of quantitative similarities. After that, glyph samples are placed into lower dimensional spaces (2-D) depending on their shape representations. These tools are utilized for mapping in augmented reality where it could be beneficial for experts during catalog construction. Such tools are quite popular in between scholars and a trended topic to categorize the non-frequent glyph samples.

GLYPH CHARACTERISTICS
Glyphs are well suited to visualize multivariate data by including the data of geometric objects with properties like orientation color and size. By using glyph, the information can be visualized containing the certain variables. The variable associated with glyph can influence the appearance of the actual glyph. Different variables are getting mapped with different glyph properties. By using this technique, multiple parameters can be visualized simultaneously. Glyph prototypes have multiple sets of properties based on their shape. With few glyphs, some properties are easy to perceive that are well distinguishable but can illustrate only some of the variables. If the properties are hard to perceive but able to carry a larger number of parameters, therefore, a tradeoff between the number of visualized variables and complexity of a glyph prototype has to be created. To solve this problem, a supertorus prototype with the subset of its properties has been used. These are roundness, opacity, size and color of the glyph. The supertorus principle is used to derive the letter by appropriately transforming the equation [99]. To depict data types, some properties are used such as for a variable containing information the property of size should be used. Another one is hue property that is universal by nature. Some of the other properties of glyph are as follows: Glyph mapping: By using mapping scheme, there is an establishment in between a variable and the desired corresponding glyph property. It illustrates the relationship between property values and the data values. Piecewise linear functions are used for glyph mapping consist of the discontinuities. At certain ranges, it permits the accentuation. Glyph placing: It defines the position of hmglyphs by depicting the data and their layout, which are going to be perceived pre-attentively. In addition, glyphs are added with the rendered surface or volume to visualize the contextual data. Therefore, proper location needed for this property, which is critical for several reasons. Enough glyph should be present that are evenly distributed. This feature is used to characterize the depicted data.  [12] Recognition method using decision tree and BP neutral network.
To recognize Mongolian Historical Document 2011 Mesh glyph [13] Glyph recognition method using normalization method To recognition high quality glyph and convert into chinese handwriting.
2014 Character glyph [14] character recognition system using a gradient based features and run length count (GBF-RLC) Recognize the handwritten character.
2016 glyph of the character '' [15] Glyph recognize using curve matching method Measure of correctness and accuracy of font reconstruction 2016 Shape glyph [16] Hamming distance code Glyph recognition 2017 Handwritten Tamil glyph [17] Glyph recognization using hybrid feature extraction & hierarchical classification Recognize the Offline handwritten Telugu character.
2018 Stroke and quoted glyph [18] Character recognition using convolutional neural network (CNN) method Correction of Kanji character recognition 2020 Maya glyph [19] Glyph classification using traditional shape descriptor and neural network Recognize the complexity of maya writing The remaining manuscript is structured as follows: Sect. 2 represents the comparative existing techniques that are used in Tracking system. discussion and the last section presents the concluding part.

Literature Review
AR systems support the same environment applications which are computer generated (or synthetic) and user's world (the coexistence of real elements) [20]. In present time, such applications are very much in demand because it permits the user to carry out the operations in much efficient, intuitive and effective way. AR interfaces are operable for both 2D and 3D and generated with the superimposition of virtual. AR systems are full of unique features that provide real time applications for creating textual or pictorial real world. This interface is registered under 3D and permits their user to make interaction with virtual and real elements at the same time. The visual appeal is the most critical problem for AR systems because it is the final result that a user views. Therefore, according to the requirement of designing of AR systems, substantial efforts are placed for seamlessly fitting the information into the scene [21]. Ideally, the main purpose of this technique is that a user cannot be able to discriminate in between virtual and real scenes that have the information. To make such kind of interfaces needed virtual elements for showing both the consistencies such as photometric (mutual reflections, shadowing, chromatic adaptation to scene illumination) and geometric (occlusions identification, correct size/placement). Such types of problems are not even trivially solved under simplified conditions.
In real environment, these problems related with the correct positioning of virtual is known as registration. With the help of tracking, the synthetic elements can be adequately registered with the real scene. Tracking becomes the solution of the problems related with correct positioning.
Numerous techniques based on tracking technologies are now developed and available, such as thermal imaging, GPSs, ultrasound, magnetic sensors, optical sensors, movement sensors etc. [22]. These tracking devices are well equipped to capture visuals and features from the real world. Depending on such information, AR system automatically regulate show, where and when the virtual scene should be displayed. Mostly, optical tracking is utilized for regulation and controlling purpose because of the requirements of accuracy, low cost and robustness. Optical tracking is further characterized into two types which are marker (glyph) based and markerless (without glyph).
For registration, marker based tracking is a better acclaimed and useful. It is used to perform camera pose estimation to achieve artificial patterns present in the real time environment. While, the markerless tracking is used to situate the virtual objects in the real scene which makes it different as compared to marker technique.
For optical tracking, on AR markers under built environment, Lamberti et al. [33] have proposed a review on its applications. They highlighted the significances of AR marker as well as concluded that to build an effective AR system, the biggest single problem is the necessity of precise trackers with long-range and accurate sensors. Optical tracking is considered as the most accepted technology for such requirement. For AR applications, such tracking is required to find out the viewing direction, angle and position of camera. Depending on the characteristics and outcomes of the tracking algorithms, markers can be classified as: ID Markers (glyphs): These are used for general applications of AR systems and are rectangular 2D markers. They can very robustly tracked and detected the location due to distinctive black border and have a fixed structure [34]. The configuration of different encoded information including with a few hundred markers can easily make an efficient system by using inside pattern. Picture Markers: These are the combination of markerless and ID markers including their features. They have a distinctive and strong rectangular border just like ID markers. In addition, they can consists enough visual content of any arbitrary image inside the boundary. Such feature makes it faster as compared with borderless markers. It is also known as distinctive border [35]. Quick Response codes and Barcodes: These are readable 2D-optical machine illustrations of data items. Barcodes are used very often and a well-known technology utilized in day to day life. Quick Response (QR) code markers are analogous to ID markers. They work for white background and contains of black square modules arranged in a square grid. Depending on this, they can be interpreted and separate the information from patterns as well as read by imaging devices [36].

Markerless (without glyph):
This technology is mostly referred for 2D borderless markers and maybe misleading for 3D. It needs to consist moderately textured content because it does not have a clear rectangular boundary. Due Markerless (without glyph): This technology is mostly referred for 2D borderless markers and maybe misleading for 3D. It needs to consist moderately textured content because it does not have a clear rectangular boundary. Dueto some exceptional features of advanced algorithm and oint descriptors of 2D distinctive visuals, they can be robustly tracked and detected [37]. Markerless 3D: This tracking is the most advanced and highly integrated optical tracking technology. It provides some unique features of any real world such as the tracking and detecting the object with the mapping of 3D technique. Though, the 3D object requires to be scanned from different angles/perspectives and have sufficient visual features for determining the distinctive visual characteristics [38]. CAD edge model tracking: This technique is a very innovative and consist the feature of3D CAD model. To enable a precise scaled and accurately localized augmentation, edge-based pose initialization has been considered. Based on pose initialization, the tracking method automatically moved to Markerless 3D tracking. User is required to initialize the very first pose for the preparation of CAD edge model [39].

Marker (glyph) Based Trackers
To estimate the orientation and position of camera, there is a need for artificial markers based on indoor positioning methods in AR systems. Seo et al. [40] proposed an inspection scenario for ARbased field. They suggested the framework of a Building Information Modelling (BIM) based construction defect management system based on artificial 2Dmarkers. Koch et al. [41] have suggested the approach based on an outside in tracking. At the head of potential operator, an infrared invisible marker was located in this marker. This infrared marker estimated the viewing direction and the position from the outside to track and detect the object. However, it requires an assumption that the proper maintenance is present during re-installed infrared tracking devices. Calculations have been made for obtaining appropriate routes when the operator's position and the location of the target (maintenance component) are determined. Topological routing graphs are used for this purpose and produced depending on either 3D building geometry or derived 2D floor plans. Along with that, for achieving minimal path costs, calculation of path have been made by routing algorithms (e.g. Dijkstra, A). User is already familiar with the route, navigation instructions as well as current orientation and position of the object. However, following instructions are usually indicated for arrows on 2D floor plans and point markers [42].
In mechanical engineering based industries, AR is used for supporting repair and maintenance operations such as in engines present as vehicle equipment. By the use of a see-through Head Worn/Mounted Display (HWD/HMD), the user is getting familiar with the natural view which consists of the animated sequences of the procedure, text, arrows and labels for the execution of tasks, their facilitation, localization and comprehension. However, some of the major disadvantages of HMDs are there, such as the distinct uncomfortable head mounted device and the low display resolution.

Natural Indoor Markers
In general, AR based applications majorly need the artificial markers. Following markers consist of visual patterns and dimensions with very distinctive images. In these markers, computer vision algorithms are used for real-time applications and utilized for superposition of virtual 3D content with reference objects onto the live view of camera. Apparently, it is unaesthetic and practically inefficient for installing the natural indoor markers all over the interior facility. Researchers solved this problem by suggesting that to use natural markers that are easily and already accessible on-site. They proposed Position marks of fire extinguishers are another effective natural markers. These are the examples of sign marker but with device ID tags or textual information hints. Some works are available on picture markers which is currently the most appropriate one. Picture markers such as fire extinguisher signs or exit signs have a clear indication in a rectangular shape. They consists a strong visual content with distinctive boundaries with other signs. Furthermore, more mature and technically sophisticated trackers are available for the tracking and detection of the object. A demerit is witnesses by the user by the fact that exit signs are ubiquitous and used very often. It can be used more than once within a building. Due to this, the absolute indoor position cannot be determined precisely. Therefore, navigation system can advantageously be used. It can consecutively re-position the markers if the initial positions are known. Many work preferred navigation based AR system due to this reason rather than AR-based absolute positioning.

Compilation of Digital Work Order
BIM can offer the information on natural markers like their geometry, location, type, spaces and facility maintenance elements. It also provides the additional data such as facility maintenance data like their manuals and maintenance schedules for the compilation of a digital work order (DWO). An assumption is made on BIM that it has all the information of locations in their routing algorithms for producing the navigation paths. Depending on routing algorithms, suitable navigation schemes can be generated for the ensuing activity. Mostly, a DWO features with the following information: • Maintenance information: It consists the augmented repair instructions and device instructions for animation sequences, charts, texts etc. • Location information: It states the initial point of operator, natural markers with navigation route and the maintained components. • Navigation information: It features the information of about augmented guidance instructions such as maps, indications, virtual models, distances, arrows etc.

Indoor Navigation Augmented Reality Systems
Such system reports the issue of how to navigate indoors towards the maintenance component of interest. This can be done by estimating the initial position of operator, the positions of natural markers with target location and their navigation route. These systems are preferred for real-time navigation and supported on-site applications of AR systems. Pre-defined natural markers are mostly suitable for this purpose. After recognition, image frame of camera and its size with marker's location is used to predict the field of view of camera (orientation) and the position of operator with distance to the marker. Virtual navigation performs tasks in organized steps. Started with the recognition of marker, first step decides the navigation instructions, which then synchronized and superimposed with the live view of as long as the visibility of marker. Further, dynamic instructions are depending on the current orientation and position of the operator on a 2D map, superimposed direction arrows (2D and 3D) and targeted directions [43]. During the visibility, a superimposed simplified 3D model highlighted the location of target. After the obtained values, without the large deviations, operator strictly follows the suggested navigation path. Once the next marker is recognized, then the loading of next navigation step has been automatically done. In case of no marker recognition due to small in size or the other factor, then the last instruction of the previous step is automatically presented to the operator.Furthermore, the magnetic compass is used with the camera for the improvement in the estimated values of orientation and position of the object. During the absence of markers, the compass gives the information of orientation. Additionally, if the navigation route consists the direction turn than the compass is utilized for automatically switching to the next navigation step. For instance, after a right or left or right turn, loading of the next navigation step has been done and recognition of next potential marker completed. Table 2.2 illustrates the development in marker system with their type, technique and applications.

Markerless (without glyph) Tracking Systems
Another difference is that, in markerless AR systems, any part of the real environment may be used as a marker. It illustrates the natural features for real scene to perform tracking. Tracking can also be divided on the basis of approaches such as outside -in tracking (optic, magnetic, ultrasonic and mechanic) and computer-vision based inside-out tracking methods. For Virtual Reality (VR) systems, outside-in tracking technique is mostly preferred and is used to setups many AR. The applicability of trackers depends on the type and amount of instrumentation required by the real world. To enable tracking, cameras, emitters, sensors have to be placed in the environment. For industrial environment applications, such instrumentation is the main concern. Although, in research, some approaches are used to tackle the problems related with instrumentation by allowing tracking without the need of instrumentation (e.g. marker-less tracking). Harris in 1993, proposed RAPiD (Real-time Attitude and Position Determination) [44] that was the earlier and the most popular tracking system for 3D visuals. Due to its efficient working, this system becomes a demanding markerless tracking and several researchers were taking interest in this marker. Numerous vision-based tracking systems are based on the work of Harris. During the extraction process from video feed, this technique made it possible to minimize the amount of information. After that, Park et al. in 1998 suggested an advanced method that permits the users to use natural features rather than artificial features during tracking. It required calculating the pose of camera due to which system dynamically attained the extra natural features [45]. By the help of such data, continuous updates had been made for the calculation of pose. This type of robust tracking provides the information even if the original fiducials are no longer in view.
In 2004, Vacchetti et al. presented a new method for real time 3D tracking by combining texture and edge texture information. For each frame, interest points are achieved in the image. Such points of interest are then coordinated and synchronized with the points of interest of reference frame utilizing for obtaining the trajectory of camera. An approach for developing line tracking model working under real-time was proposed. It consist the edge features with adaptive learning that can handle illumination changes and partial occlusion. A CAD model is used in this approach for proper tracking of object. This approach was significant in improving the efficiency and robustness of AR systems.
For outdoor AR systems working under urban environment, a model-based tracking system had been introduced. This system is appropriate for handheld devices as it enabled accurate real-time overlays. Several well-known approaches feature inside this technique such as measurements of gyroscope to deal with quick motions, edge-based tracker for accurate localization, estimation and calculation of magnetic and gravity field to evade a back store and drift of reference frames. It also included the selection of online frame to restart the process after failures or dynamic occlusions. In case of outdoor AR, markerless tracking was proposed to deliver a reliable and robust tracking with the help of mobile handheld camera. Real world could be reconstructed efficiently by edge based trackers combined with partially known 3D scenes with a sparse 3D technology.
Furthermore, a direct tracking approach was proposed by Dame and Marchand in 2010 [46]. This methodology was specially designed for proper alignment in which it included the mutual information of the object as significance metric. It can offer an accurate estimation of the displacement with a real time and robust application. 3D reconstruction and real-time camera tracking. In case of 3DOF, an approach has demonstrated for improving the orientation and accuracies for outdoor AR. It is specially designed for buildings in which corner points were taking in use to detect the vertical edges in the image. It also included the refinement of compass orientation and GPS location. In the same year (2010), research on multiple 3D objects has been placed to find out the real-time solutions for their tracking and modeling for the unknown environment. Following system can easily track 40 objects within 6 to 25 milliseconds for 3D environment.
Carrying on, in 2011, Ababsa and Mallem [47] created a framework for particle filter. To achieve the real-time estimation of camera pose, line model based trackers including framework with points were designed. This technique was appreciated for its flexibility and simplicity. They further proved that tracking of camera pose can be accurately done with their algorithm under non-smooth camera motions and severe occlusions. Also in 2011, 3D system was presented with online training and textureless object detection with the help of depth camera. This technique suggested by Park et al. [48] and enhanced the depth map by removing the need of prior object model because of the availability of any data for tracking and detection on the fly.
Moreover, markerlessvision based camera tracking, tracking -by synthesis is a promising method for the applications of AR. Simon suggested the technique of the combination of pyramidal blurring and fast corner detection that can provide high speed to the system. Later, he introduced a real-time method for tracing planar objects which were weakly textured with the estimation of their 3D pose simultaneously. Following with, Donoser et al. [49] suggested the tracking system for nontextured objects by adapting the basic idea of classic tracking by-detection approach. This approach required the object to be tracked independently in each frame. Main limitation of this approach was the performance of tracker deteriorated by observing the plane to be tracked. This was due to a significantly oblique angle in between moving object to a distant location and to the viewing direction from the camera. This limitation can be resolved by sampling technique to reconstruct the visuals. Ito et al. in 2011 suggested the methodology to use linear filter for correcting the template [50]. In real time, it tracked the plane and then corrections had been made by means of a tracked pose of the plane. Further, the technique was used for optimization. In 2011, Lieberknecht et al. demonstrated a consumer RGBD camera based on real-time method for tracking the motions of the camera within an unidentified environment. For this, it recreates a dense-textured mesh for tracking system. Uchiyama and March gave another demonstration in 2011 [51] for the tracking and detection of different types of textures comprising handwritings, fiducial markers and colorful pictures. 2012 Fiducial marker V [18] Marker recognition algorithm using Geometrical recognition process and parameter optimization method.
To check accuracy and robustness of CT scan images. 2013 Matrix marker [19] Marker recognition using image processing and rendering process Augmented reality application using Android operating system 2017 Simple, Average, complex marker [20] AR Toolkit tool and Border optimizing tool using thresholding application Design the high quality marker to enhance the performance of AR system 2019 AR Toolkit marker Design of high quality AR Toolkit markers using the optimal value for various factors like B/W ratio, edge sharpness and information complexity.
Enhance the design of marker in terms of detection accuracy and performance 2019 Pattern marker Marker recognition using image processing, normalization and template matching

Conclusion
AR shows its potential to sort out several challenges faced by users. Several professionals are engaging themselves in the advancement of AR system. AR now involves in users life with the completion of their day-to-day tasks. In addition, discusses ideas having various approaches during utilization of AR frameworks along with glyph tracking system