Registration Issues in Augmented Reality

Transcription

1 Registration Issues in Augmented Reality Yan Yan Computer Science University of Birmingham Abstract Augmented reality (AR) as a future interactive technology is currently on the way from laboratory to market. Registration as the key technique highly affects how successful a virtual object is superimposed on the real world. This paper reviewed the development of three type of registration methods, which are sensor-based, marker-based and maker-less registration, and their characteristics. A brief comparison of these three registration methods was conducted after the review, which indicates that marker-less registration is likely the most advanced and adaptive technique for future AR applications. Finally, two potential research orientations to enhance the performance of marker-less registration to realise commercial AR products, including deploying advanced computational graphic algorithms and deploying correction techniques based on sensor data, were proposed Total Immersion is the company of first AR solution provider Hirokazu Kato published the ARToolKit which is the most popular AR developing tool Bruce Thomas et al. created the first AR game named AR-Quake (Figure 2) The first AR system on mobile phone published. Keyword Augmented reality, Registration, Fiducial marker, Natural feature tracking I. INTRODUCTION In the recent decades, the conceptual term Augmented Reality (AR) and its relative research have been become popular in the field of computer science, especially in the area of HCI and computer graphics. Craig proposed a formal definition of augmented reality is that [1]: A medium in which digital information is overlaid on the physical world that is in both spatial and temporal registration with the physical world and that is interactive in real time. In a simple word, this definition could be interpreted as that creating an environment where virtual entities are superimposed in the view of users in real world [2]. Although augmented reality has just used as commercial applications for a short period, its history is more than 40 years. Kipper and Rampolla briefly reviewed the history of augmented reality in their book and the remarkable events are listed as follows [3]: Ivan Sutherland created the first AR system Myron Krueger created the first AR application which allow user to interact with a virtual object (Figure 1) Tom Caudell and David Mizell invented the term Augmented Reality Ronald Azuma provided a factual definition of AR is that it has three characteristics combining with real and virtual, interacting in real time and registered in 3D [4]. Figure 1: The first AR with interaction Figure 2: AR-Quake Nowadays, after the development of more than 40 years, AR application has been used in various areas of our daily life. The main domains of AR application are shown in following list [3]: Advertising Companies such as Nisan, BMW and Lego use AR application to intuitively illustrate their products to consumer. Task support To assist people easily conduct some complicated task such as assembly and maintenance. 1

2 Navigation Creating a intuitive navigation by overlying route on the road (Figure 3). Home To project virtual furniture or appliances to aid decoration design (Figure 4). Art Some museums has been applied exhibition assisting software based on AR, such as New York s Museum. Sightseeing Allow tourists to explore some hidden or internal details of objects or place. Entertainment Providing users AR games such as Sony and Nintedo have done in their handhold game console [5]. Social networking Enriching the social networking experience such as attacking some relative information related to the people you are seeing. Education Such as AR book, which upgrades 2D book to interactive and stereo 3D graphic and audio [6]. Translation Overlying translation text on the original text (Figure 5). Figure 3: AR Navigation Figure 4: AR application shows a virtual TV Figure 5: AR translation Regardless of various type of applications, the bases of AR system should have these four key aspects [1]: a) Virtual information superimposes on real world. b) Virtual information registers on the proper position of real world. c) This registration is with respect of perspective of person in real world. d) People who are in AR experience could interact with virtual information. Among these four aspects, registration is a crucial process of realising AR application because it is the precondition of interaction [4, 7, 8]. Only the accurate and correct registration can provides a meaningful perception to users which allow them to interact with application system correctly. With the development of AR, techniques of registration have been deeply researched and realised. Many different registration techniques have been involved into AR applications and achieved a significant success. However, the evolution of registration methods are still in the progress. All of current methods have their distinctive preference and performance. The requirement of advanced registration methods is still not be fulfilled. Therefore, this paper aims to investigate the current development of AR registration techniques, to discover their advantages and disadvantages and to reveal the requirement of advanced registration methods. In this paper, the concept of AR registration will be explained in the section two. Three main registration methods which are sensor-based, marker-based and marker-less registration will be discussed with their distinctive features in the section three, four and five respectively. Finally, the current research of registration methods and the future demands will be discussed in section six. II. CONCEPTS RELATED TO AR REGISTRATION A. Notion of registration As mentioned in the previous section, registration is a critical part to realise augmented reality. Generally, registration could be interpreted to the accuracy of virtual object spatially aligning in real world [9]. Furthermore, it should reacts the changes of location and orientation when the perspective of users or environment of real world change [10]. For example, as Figure 6 shown, a 2

3 digital vase should be projected on a real table and shows different appearance for users who look it from different positions. The correctness of registration is remarkable important. If a virtual object is incorrectly registered in real world, the result could be located in a wrong place (shown in Figure 7), floating above/below the surface where it should be displayed or not fixed on where it should stay like swimming [9]. To realise correct registration, there are two procedures should be achieved [9]. Firstly, the tracking system of AR application which consists of different sensors should be accurate enough. Secondly, the coordinate system of real world in which virtual objects are projected should be resolved successfully at any given time with any environment changes. Only if these two procedures could be simultaneously achieved with high standard, the registration of AR would get enough correctness and robustness. B. Categories of registration Figure 8: Six degrees of freedom [11] Successful registration needs an accurate tracking system which mainly depends on various sensors integrated in AR system. These sensors track six degrees of freedom (shown in Figure 8) of entitles to determine position including both location and orientation of them [9]. Craig has demonstrated main sensors used in tracking system, including camera, GPS, Gyroscopes and others, which are briefly explained in Table 1 [9]. Table 1: Main sensors used for tracking Figure 6: Example of registration Type Camera GPS Gyroscopes Accelerometers Compass Explanation Needs the aid of environmental cues to determine location an orientation Determines three degrees of freedom (X,Y,Z) but only can be used outdoor Determines orientation values Determines the direction of something and its changes Determines direction of pointing Figure 7: Incorrect registration Among these sensors, camera is the most functional and flexible one. Combining with visionbased computational technologies, such as graphic recognition algorithms, camera-based registration is potentially competent for any AR application. In addition, it seems to be the only choice in most of indoor AR applications. However, as mentioned before, camerabased registration must need cues to aid its determination, which could be either artificial fiducial markers or real objects recognised as markers in real time computation [9]. Otherwise, all of other sensors have their own limitation to work as main registration methods individually. Minority of them (such as GPS) could be used as the main method in particular AR application (such as AR navigation). Majority of them can just be used as aided registration methods. Therefore, to appropriately classify registration methods, the categories introduced by Pucihar and Coulton is like to be the best classification and accepted by this paper [5]. The categories is: Sensor-based registration (excluding camera) 3

4 Marker-based registration (vision-based) Marker-less registration (vision-based) Details of those three registration methods will be discussed in following three sections. III. SENSOR-BASED REGISTRATION A. Notion an examples Sensor-based registration means using non-vision sensors to tracking information which is needed to realise registration [8]. These sensors mainly include mechanical sensors, magnetic sensors, GPS, ultrasonic and inertia sensors [12]. Different sensors have their own features and functions with commonly high-frequency response. There are many examples of using non-vision sensorbased registration in AR system. Feiner, et al developed an AR system attaching ultrasonic sensor as transmitters and receivers on head of users and targeted objects [13]. Tuceryan, et al using magnetic sensor to track a handheld pointer to register virtual objects on the place where the pointer pointed [14]. Additionally, Boeing has also used magnetic sensor in their heads-up display to aid manual manufacturing task [15]. B. Advantages and disadvantages Different sensors have their own advantages and disadvantages, Li et al has proposed a comparison of sensors illustrated below [12]: Table 2: Comparison of sensors Sensors Advantages Disadvantages Exactness, low time delay, no vision or Mechanical magnetic Limited use field disturbance, range suitable for exact track for small objects Magnetic sensing GPS Ultrasonic Inertia Low price, exactness, no vision occlusion, good noise immunity, suitable for large field track Suitable for outdoor large field track Low price, no magnetic field disturbance, light equipment No distance limitation, high speed, no vision or magnetic field disturbance, small size, low price Easily disturbed by magnetic field and metal in the environment Uncertain measuring precision, time delays Easily disturbed by ultrasonic in the environment, low precision in large range 3 degrees of freedom, drift, not very exactly at low speed To sum up, advantages of non-vision sensors could be summarised as follows: Easy to implement these sensors initially integrated in mainstream AR devices and accessible by defaulted interval programmes [5]. Relatively low energy and computational capability consumption. Conversely, the disadvantages of non-vision sensors are also significant: Low accuracy of sensing data [5, 16] Used in minimal environment, such as GPS can only be used in outdoor [5]. Hardly to create stable augmentation and to ensure correct context. Hardly to detect changes of user s view. Owing to these obvious disadvantages, sensor-based registration can only be used in some particular AR systems individually. The combination with vision sensor is commonly necessary to realise a complete AR experience. The trend of using sensor-based registration is using it as assisting methods and correcting techniques of version-based registration which will be discussed in the following sections. IV. MARKER-BASED REGISTRATION A. Notion of marker-based registration Vision-based registration is currently mainstream registration methods which has been commonly researched to overcome the limitations of sensor-based registration. It uses computational graphic recognition to capture special targets in physical world frame-to-frame to determine positions of registration. In the current stage, the most robust and mature way is mark these targets with fiducial markers. Fiducial markers is a type of artificial landmarks which are manually placed in physical world to allow AR application recognising them [9]. A typical fiducial marker is illustrated in Figure 9. Fiducial markers could have various patterns, either physical or electronic, static or mobile, symmetrical or asymmetrical. It could be unique patterns which are previously designed by users and recorded in programmes. More advanced, it also could be a real object, such as human face or city skylines [9]. Figure 9: A typical fiducial marker B. Various fiducial marker and relevant research As the most common techniques of AR registration in current stage, fiducial markers have been massively researched by different researchers. The following part 4

5 introduces the evolution of fiducial markers and their relevant research. What should be mentioned before is that the continuous research of fiducial markers does not mean the former type of markers should be totally discarded, although the later research is aim to overcome some limitations of former markers. The use of particular fiducial markers mainly depends on the particular purpose of AR systems. For example, an AR exhibition is possibly realised with square markers which is the early type of markers due to the environment could be previously prepared and markers do not significantly intrude user experience in this case. 1) Square marker Square marker is the most typical fiducial markers, in which the unique patterns are location inside a black square frame. Different square markers are recognised by their unique central patterns. Then, camera posed for each tracked markers are determined by vertex coordinates of the square [17]. Square markers have been applied in a great number of AR systems and toolkits. The famous and common used toolkits ARToolKit also uses square markers as its standard marker system [18]. 2) Circular marker Circular marker is introduced by Naimark and Foxlin are shown in Figure 10 [19]. It could be considered as a transformation of square markers. Instead of recognising particular central pattern, circular markers should be recognised by its arrangement of white and black areas. However, circular markers meanwhile inherit the constraints of square markers. Them are hardly to be recognised when any part of markers are occluded and are too obvious to disappear from view of users [17]. 4) Alphabet combination marker Han et al. proposed this marker in 2011 which could be constructed and modified in real time with an intuitive meaning could be understood by users [17]. This type of markers use English alphabet letters as fundamental units. Users could create a marker by combine as least four letters in minimally two rows to a meaningful English word. The camera of AR system captures this marker and projects same virtual objects with the word represented. The position of virtual objects (camera poses) could be determined by analysing the midpoints of each letters. Which is the reason of arranging with minimally four letters and two rows. An example of alphabet marker is shown in Figure 12. Figure 12: Project a ship with alphabet marker Figure 10: Example of circular markers 3) Dot-based marker Dot-based marker is a relative new concept of markers. Bergamasco et al. proposed a type of dot-based markers in 2011 named RUNE-Tag which is constituted by dots arranged with different pattern in concentric rings (Figure 11) [20]. Comparing with square and circular markers, this dot-based marker has a critical advantages is that it is able to still be recognised even though party of it is occluded. However, it is still be considered to intrude view of users. Comparing with three former markers, alphabet markers have a number of advantages. It does not need a fix shape and could be created by users in real time. The pattern of marker is the word of the projected object, which provides an intuitive interaction and less-intrusive environment. Additionally, it is able to achieve a high error resilience of occlusion. 5) QR code marker QR code marker uses quick response codes as markers to achieve a high storing capacity and high reading speed, which is proposed by Kan et al. and Kong [21, 22]. A typical QR code marker is illustrated in Figure 13 below. Three big square in different corners are used to detect position. A small square is used to alignment. The other parts of the QR code contain information used to be decoded which is normally a URL to access models in remote servers. Figure 11: RUNE-Tag 5

6 in London to project a view of animated dinosaurs to users on tablets [25]. Figure 13: QR code structure Using QR code as markers several advantages. Firstly, AR systems do not need to record all relationship of markers and models in advance because QR codes contain URL to access models online, which provide a large information capacity and freedom of modification. Secondly, QR codes have similar appearance of square markers, which means the current toolkits are able to recognise them. Finally, QR codes could be decoded by various programmes. 6) Nested marker Nested marker is introduced by Tateno et al. [23], which could be considered as an improvement of square marker. In the general view, nested marker has a recursive structure, which means a larger nested markers consist of several smaller markers which possibly contain further smaller markers as Figure 14 shown. The recursive structure gives nested markers a high flexibility cameras could choose which level of markers will be used in registration. For example, if a nested marker is fat away from a camera, the whole markers will be considered as an entity to be recognised. Otherwise, if the marker is near enough to the camera, smaller markers will work for registration. In this case, one larger nested marker could be used as several markers. Besides the advantage of multi-function, nested markers also partly solve the problem between camera focus and markers. In the conventional situation, if a camera is too near a marker, it will lose the focus then cause registration failure. However, the recursive structure of nested markers a solution for this problem. If a camera is too near a marker to focus on, it can alternatively focus on smaller markers contained in larger marker. C. Advantages and disadvantages Seven type of fiducial markers have been introduced above with their particular features and advantages. In general, marker-based registration is a more reliable method than sensor-based registration. There are three advantages of using fiducial markers which are commonly approved shown as follows [5, 23]: Using fiducial markers providing high accuracy of registration. Normally, the accuracy could be controlled less than the benchmark of AR system which is 1 centimetre and 1 degree. Marker-based registration spends less computational capability and easily be implemented. Marker-based registration is robust because fiducial markers are distinctive with physical world they existing that hence easily to be detected In contrast, fiducial markers have a number of inherent limitations which negatively affect AR systems using marker-based registration. Although all researches related markers mentioned above aim to reduce or solve these limitations in some extent, such as infrared markers do not obtrude view of users, they are still obstinately existed in majority type of markers. These inherent limitations are pointed by Kipper and interpreted in the following Table 3 [26]. Limitations Occlusion Unfocused camera Motion blur Uneven lighting Table 3: Limitations of fiducial markers Explanation If markers are blocked they cannot be recognised. Unfocused camera causes low precise interpretation of markers, which may lead to total non-registration. The fast movement of either camera or mobile markers will reduce the recognition rate. Shadows or dark lightness could obscure markers that prevent them to be recognised. Figure 14: A nested marker 7) Infrared marker All type of fiducial markers are obtrusive due to their visibility. To solve this problem, Burnett and Coulton invented a infrared marker system [24]. This is an invisible marker because infrared light cannot be seen by human but can be detected by camera. Infrared marker has been practical applied in National History Museum Owing to those limitations of fiducial markers, several difficulties are imposed to the AR applications using them [5, 16, 23, 27]. They are interpreted as follows: Difficult to deal with viewpoint movement: viewpoint movement may causes the changes of lightness or occlusion which leads to fail to continuous registration. Disturbing user experience: fiducial markers need to be explicitly attached in physical environment with usually meaningless to users. Only suitable to prepared environment: fiducial markers need to be attached in real world in 6

7 advance. In addition, the information related markers need to be stored in relevant system before operation. This causes that fiducial markers just could be used in prepared context. All of these difficulties prevent those AR system using marker-based registration to be implemented in ubiquitous and complicated environment, which goes against the tendency for the development of AR systems. Therefore, a new registration methods without the need of fiducial markers becomes the new trend, named marker-less registration, V. MARKER-LESS REGISTRATION Marker-less registration is occurred and evolved with the development of computational vision algorithms. In general, marker-less method directly extracts some virtual indices, which could be points, edges, corners and segments, from images captured by camera that uses them as markers to register virtual objects [28]. An example of marker-less registration is shown in Figure 15. Figure 15: Example of marker-less registration An important and fundamental concept in markerless registration is natural feature tracking (NFT) which directly tracks features in the environment to instead of markers [16]. Pucihar and Coulton divided marker-less registration into offline system and online system that both need to create and maintain a 3D map of current environment to realise registration [5]. However, offline system in which maps need to be initialised in advance and stored in the system is understood as an analogy of marker-based registration, which just use natural features to instead of fiducial markers. Therefore, this paper keeps the concern of online system which has no requirement of prior information of environment. There are two main approaches to achieve online marker-less registration model-based tracking and move-matching tracking [5]. A. Model-based tracking and related research As Pucihar and Coulton introduced, model-based tracking method estimates camera pose with correspondences which are calculated by comparing features of the initial key frame and current frame [5]. The term of Model in this context could base on edgefeatures or line-features. The former one is more popular used depending on its high identifiability and adaptability of lighting changes [29]. The following steps consist of the complete procedure of model-based tracking [29]: 1) Extracting features from initial model image and current camera frames. 2) Comparing features in model image and camera frames then creating correspondence with the pair of features. 3) Using this correspondence to estimate camera poses. 4) Keeping to refine the result until the threshold of similarity measure has been satisfied. After Comport et al. presented the suspiciously first model-based tracking system [30], a great number of related research have been conducted. Wagner et al. proposed a model-based tracking methods based on textured planar targets [31]. Hagbi et al. invented an AR system based on model-based tracking that allows user to create new shapes with the rectification of system [32]. Furthermore, Metaio SDK is a commercial library to support model-based tracking method [33]. Model-based tracking is able to provide a robust and efficient tracking system. However, it could be extremely time-consuming with respect of the complexity of models and the range of environment. B. Move-matching tracking and related research Being distinctive with model-based tracking, movematching tracking extracts information related to correspondence and camera pose based on the frame-toframe movement of tracked features [5]. A technique called Simultaneous Localization and Mapping (SLAM), which is initially invented for autonomous robotics, is commonly used to achieve move-matching tracking [5, 29]. SLAM generates a 3D map of the environment where the user stand in and localises the user within this map [29]. The camera pose is meanwhile initialised and updated with this 3D map [5]. Two tasks localization and mapping in SLAM are calculated depending on results of each other [29]. The map needs to be built with allocating a 3D position to detected features in the image. Otherwise, the 3D position is calculated depending on camera pose which is estimated by localization. Moreover, the localization is based on the 3D position from the map to realise the correspondence search and following pose calculation [29]. Owing to those co-dependant tasks, initialising a map is a problem which could be solved by indicating an initially estimated pose [29]. Several methods have been researched to solve this problem. Klein and Murray integrated in stereo view in their solution, in which the initial key frame is selected by users then the second key frame is automatically selected to form a valid stereo pair to initialise the map [34]. In another related research, Ziegler used a marker to gain the initial pose [29]. The form of 3D map is another concerned research field. Some early-stage-researchers such as Klein and Murray created a global 3D map which covers the whole observed scene [34]. However, this global 3D map has some drawbacks. On one hand, a global map is always bigger than the current needed scene. For example, if some virtual objects need to be asynchronously superimposed in different floors in a building, a global map need to be created with all floors once. On the other hand, global maps only work for relatively small 7

8 workspace currently due to mapping a big and complex workspace is unachievable depending on current immaturity techniques [35]. Therefore, a technique of multiple sub-maps has been proposed by Guan and Wang, which only covers the current scene where virtual objects intend to be registered [35]. Using sub-maps could significantly reduce the computational complexity comparing with global maps. However, sub-map technique still remains some difficulties, including arbitrarily add new sub-maps and switch between different sub-maps. C. Advantages and disadvantages Comparing with marker-based registration, using marker-less methods has a remarkable advantages is that discarding the dependency of fiducial markers, which allows them to work in unknown and unprepared environment. Hence, those AR applications which is using marker-less methods are potentially able to realise the ubiquitous augmented suitable to work in anywhere with any device and any condition. However, marker-less registration still requires a long process of development to achieve its full potential. The current marker-less methods are focusing two obvious difficulties. The first one is that marker-less registration has a high dependency of computational vision techniques which are still not mature enough to support AR applications working in complex environment. Neither model-based nor move-matching tracking achieve a proper robustness and accuracy to maintain a correct registration. For example, in the research of Li et al. which aimed to recognise edges of building to register virtual objects, a registration drift occurred after the system continued to work for a certain time, which is shown in Figure 16 [12]. There are a great number of factors could affect the robustness of marker-less registration, including changes of camera intrinsic parameters such as focal length [27], the change of workspaces and the limitation of algorithms. To improve the robustness of marker-less registration, advanced camera self-calibration approaches need to be integrated in registration process [27]. Figure 16: Registration drift Another difficulty is that marker-less registration has a high requirement of computational resources which mainly indicates hardware performance [10, 29]. However, most of AR applications need to be deployed in moveable devices such as HMD and smartphone to realise their augmented features. These moveable devices are not able to provide high hardware performance comparing with workstations and PCs. Hence, even though marker-less registration has the potential to be arbitrarily extended into wide and complex environment, the hardware performance of current AR devices hardly support the necessary requirement of computing ability. VI. DISCUSSION AND CONCLUSION Three type of registrations in augmented reality have been introduced. The following Table 4 summarises main differences of three registrations to provide a brief comparison. Comparison Tracking methods Accuracy Robustness Stability Computational cost Application fields Table 4: Comparison of three registration methods Using sensor (non-vision) Low (lower than benchmark) Low (Sensor data not stable) Bad (Affected by natural conditions such us magnetic fields) Using camera with fiducial marker High (higher than benchmark) High (fiducial markers are easy to be recognized) Not good (Affected by optical condition such as visibility) Low Medium High Minimal environment Prepared environment Sensorbased Markerbased Markerless Using camera with natural feature tracking Depends on algorithms Depends on algorithms Not good (Affected by both correction algorithms and optical condition) Unknown and unprepared environment In summary, these three registration methods not only provide different tracking and registering techniques, but also represent the evolution of augmented reality which aims to achieve higher augmented user experience. The sensor-based registration provides an easy implementation and low computational resource requirement with existing sensors in relevant devices. It is limited to use by characteristics of sensors which are instable and easy to be affected by environment. Following, marker-based registration based on camera has been proposed to realise indoor AR application with high accuracy and robustness. However, it just could be used in prepared environment owing to artificially distributed markers. Therefore, marker-less registration has been developed with the progress of computational vision techniques to extend AR application into unknown and unprepared environment. However, marker-less registration highly depends on computational visual algorithms which is still immature to achieve high reliable and accurate registration. 8

9 With respect of the developing requirement of augmented reality, which is performing augmented reality with better user experience in anywhere the user wants, it is possible to predict that marker-less registration will be the direction to further research and be the main registration technique in future advanced AR applications. To provide a higher robust and accurate marker-less registration technique, there are two possible directions which are worth to conduct more research, including integrating advanced algorithms into markerless registration and combining sensors to perform correction. The former one means that involving advanced computational graphic recognition algorithms into marker-less registration to enhance the performance, such as ORB (Oriented FAST and Rotated BRIEF) and KLT (Kanade Lucas Tomasi feature trakcer). Owing to the high dependency of algorithms in marker-less registration, deploying advanced could significantly reduce the computation cost and increase performance. A small number of researchers have started to concern this field [10, 27, 36]. The latter one indicated that using sensor data to correct the failures and mistakes generated during the performing of marker-less registration. As mentioned before, marker-less tracking is instable because algorithms are easily affected by changes of workspace and view of users. However, sensor data could not be affected by this type of changes, which has higher robustness in this particular case. Therefore, an available correction technique is that using sensor data to assist the calibration of camera pose and correspondences in marker-less registration. There are few researchers have begun to investigate this aspect [37, 38]. To conclude, registration methods of augmented reality have been researched with a certain extent. The marker-based registration has reached the practical level and applied in some AR applications. However, the marker-less registration which is the necessity of advanced AR application mainly stays in the experimental phase. Improving related graphic recognition algorithms and correction techniques are able to develop it to practical level and therefore promote AR applications from mainly experiential concepts to commercial products. REFERENCES [1] A. B. Craig, "Chapter 1 - What Is Augmented Reality?," in Understanding Augmented Reality, A. B. Craig, Ed., ed Boston: Morgan Kaufmann, 2013, pp [2] M. Kanbara, H. Iwasa, H. Takemura, and N. Yokoya, "A stereo vision-based augmented reality system with a wide range of registration," in Pattern Recognition, Proceedings. 15th International Conference on, 2000, pp vol.4. [3] G. Kipper, "Chapter 1 - What Is Augmented Reality?," in Augmented Reality, G. Kipper, Ed., ed Boston: Syngress, 2013, pp [4] R. Azuma, "A Survey of Augmented Reality," Presence, vol. 6, pp , / [5] K. Č. Pucihar and P. Coulton, "Exploring the Evolution of Mobile Augmented Reality for Future Entertainment Systems," Comput. Entertain., vol. 11, pp. 1-16, [6] A. Dünser, L. Walker, H. Horner, and D. Bentall, "Creating interactive physics education books with augmented reality," presented at the Proceedings of the 24th Australian Computer- Human Interaction Conference, Melbourne, Australia, [7] R. Azuma, Y. Baillot, R. Behringer, S. Feiner, S. Julier, and B. MacIntyre, "Recent advances in augmented reality," Computer Graphics and Applications, IEEE, vol. 21, pp , [8] W. A. Hoff, K. Nguyen, and T. Lyon, "Computer-vision-based registration techniques for augmented reality," 1996, pp [9] A. B. Craig, "Chapter 2 - Augmented Reality Concepts," in Understanding Augmented Reality, A. B. Craig, Ed., ed Boston: Morgan Kaufmann, 2013, pp [10] X. Ying and Z. Wenjing, "A tracking and registration method based on ORB and KLT for augmented reality system," in Wireless and Optical Communication Conference (WOCC), nd, 2013, pp [11] R. Sebastian. (2001, 4.Feb). Hitting the spot. Available: [12] Y.-b. Li, S.-p. Kang, Z.-h. Qiao, and Q. Zhu, "Development Actuality and Application of Registration Technology in Augmented Reality," in Computational Intelligence and Design, ISCID '08. International Symposium on, 2008, pp [13] S. Feiner, B. Macintyre, Dor, #233, and e. Seligmann, "Knowledge-based augmented reality," Commun. ACM, vol. 36, pp , [14] M. Tuceryan, D. S. Greer, R. T. Whitaker, D. E. Breen, C. Crampton, E. Rose, et al., "Calibration requirements and procedures for a monitor-based augmented reality system," Visualization and Computer Graphics, IEEE Transactions on, vol. 1, pp , [15] T. P. Caudell and D. W. Mizell, "Augmented reality: an application of heads-up display technology to manual manufacturing processes," in System Sciences, Proceedings of the Twenty-Fifth Hawaii International Conference on, 1992, pp vol.2. [16] M. L. Yuan, S. K. Ong, and A. Y. C. Nee, "Registration using natural features for augmented reality systems," Visualization and Computer Graphics, IEEE Transactions on, vol. 12, pp , [17] S. Han, E. J. Rhee, J. Choi, and J.-I. Park, "User-created marker based on character recognition for intuitive augmented reality interacion," presented at the Proceedings of the 10th International Conference on Virtual Reality Continuum and Its Applications in Industry, Hong Kong, China, [18] H. Kato and M. Billinghurst, "Marker tracking and HMD calibration for a video-based augmented reality conferencing system," in Augmented Reality, (IWAR '99) Proceedings. 2nd IEEE and ACM International Workshop on, 1999, pp [19] L. Naimark and E. Foxlin, "Circular data matrix fiducial system and robust image processing for a wearable vision-inertial selftracker," in Mixed and Augmented Reality, ISMAR Proceedings. International Symposium on, 2002, pp [20] F. Bergamasco, A. Albarelli, E. Rodola, and A. Torsello, "RUNE- Tag: A high accuracy fiducial marker with strong occlusion resilience," in Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on, 2011, pp [21] T.-W. Kan, C.-H. Teng, and W.-S. Chou, "Applying QR code in augmented reality applications," presented at the Proceedings of the 8th International Conference on Virtual Reality Continuum and its Applications in Industry, Yokohama, Japan, [22] R. Kong and J. Hong, "An Augmented Reality System Using Qr Code as Marker in Android Smartphone," in Engineering and Technology (S-CET), 2012 Spring Congress on, 2012, pp [23] K. Tateno, I. Kitahara, and Y. Ohta, "A Nested Marker for Augmented Reality," in Virtual Reality Conference, VR '07. IEEE, 2007, pp

10 [24] D. Burnett and P. Coulton, Eds., Using Infra-Red Beacons as Unobtrusive Markers for Mobile Augmented Reality (1st Workshop on Mobile Augmented Reality: Design Issues and Opportunities (Mobile HCI 2011) Stockholm, Sweden, 2011, p.^pp. Pages. [25] (08.Mar). Interactive film - Who do you think you really are? Available: [26] G. Kipper, "Chapter 2 - The Types of Augmented Reality," in Augmented Reality, G. Kipper, Ed., ed Boston: Syngress, 2013, pp [27] L. Li, I. S. Burnett, and E. Cheng, "A flexible markerless registration method for video augmented reality," in Multimedia Signal Processing (MMSP), 2011 IEEE 13th International Workshop on, 2011, pp [28] H. Belghit, N. Zenati-Henda, A. Bellabi, S. Benbelkacem, and M. Belhocine, "Tracking Color marker using projective transformation for augmented reality application," in Multimedia Computing and Systems (ICMCS), 2012 International Conference on, 2012, pp [29] E. Ziegler, "Real-time markerless tracking of objects on mobile devices," B. Sc. Thesis, [30] A. I. Comport, E. Marchand, and F. Chaumette, "A real-time tracker for markerless augmented reality," in Mixed and Augmented Reality, Proceedings. The Second IEEE and ACM International Symposium on, 2003, pp [31] D. Wagner, G. Reitmayr, A. Mulloni, T. Drummond, and D. Schmalstieg, "Pose tracking from natural features on mobile phones," in Mixed and Augmented Reality, ISMAR th IEEE/ACM International Symposium on, 2008, pp [32] N. Hagbi, O. Bergig, J. El-Sana, and M. Billinghurst, "Shape Recognition and Pose Estimation for Mobile Augmented Reality," Visualization and Computer Graphics, IEEE Transactions on, vol. 17, pp , [33] (10.Mar). Metaio SDK. Available: [34] G. Klein and D. Murray, "Parallel Tracking and Mapping on a camera phone," in Mixed and Augmented Reality, ISMAR th IEEE International Symposium on, 2009, pp [35] T. Guan and C. Wang, "Registration Based on Scene Recognition and Natural Features Tracking Techniques for Wide-Area Augmented Reality Systems," Multimedia, IEEE Transactions on, vol. 11, pp , [36] D. Yuanyuan, M. Zhenjiang, and C. Yigang, "Markless augmented reality registration algorithm based on ORB," in Signal Processing (ICSP), th International Conference on, 2014, pp [37] M. Bajura and U. Neumann, "Dynamic registration correction in augmented-reality systems," in Virtual Reality Annual International Symposium, Proceedings., 1995, pp [38] B. MacIntyre, E. M. Coelho, and S. J. Julier, "Estimating and adapting to registration errors in augmented reality systems," in Virtual Reality, Proceedings. IEEE, 2002, pp