Complementing the theoretical concepts taught in the classroom with practice has been known to enhance students' contextual understanding of the subject matter. Exposing students to practical knowledge is crucial as employers are expressing discontent with the skills of newly hired graduates. In construction education, site visits have been identified as one of the most effective tools to support theory with practice. While site visits allow students to observe construction projects and engage with field personnel, numerous barriers limit its use as an effective educational tool. For instance, there are safety, cost, schedule, and weather constraints, in addition to the logistics of accommodating large class sizes. As a result, instructors employ videos of construction projects as an alternative to physical site visits. However, videos alone are insufficient to draw students' attention to essential practice concepts. Annotations can be used to attract students' attention to practical knowledge while reducing distractions and assumptions. Leveraging on the recent progress in computer vision techniques, this study presents an AI-annotated video learning tool that instructors can utilize to equip students with practice knowledge when there is limited access to physical construction sites. First, this study investigated the construction practice concepts that industry practitioners would want students to know when engaging them in site visits. Afterward, the design and development of the AI-annotated learning tool were guided by the identified practice concepts, cognitive theory of multimedia learning, and dual coding theory. To determine if the learning tool can call students' attention to annotated practice concepts in videos, a usability evaluation was conducted. Finally, this research investigated the influence of individual differences that could contribute to how learners notice practice concepts in videos. This study contributes to the body of knowledge by identifying what construction professionals notice about their work and what they would like students to notice about construction practice. This study reveals that annotations of learning contents in construction videos can direct students' focus to the annotated contents, thereby contributing to the cognitive theory of multimedia learning and dual coding theory. By leveraging machine learning classification algorithms, this research identified the extent to which individual differences such as gender, academic program, and cognitive load can be detected from the ways students notice information in construction videos. Results from this research provide opportunities for researchers to further advance the potential of annotated videos in the construction domain and other fields that employ video as a learning tool. / Doctor of Philosophy / Instructors often support classroom teaching with practical experiences to enhance students' understanding. This is especially important as employers are expressing discontent with the skills of fresh graduates. In construction engineering education, taking students to construction sites to observe the processes and operations is one of the common ways of providing students with these practical experiences. However, barriers such as safety concerns, cost, schedule, weather constraints, and the logistics of accommodating large class sizes make it challenging to engage students in construction site visits. Owing to these barriers, instructors utilize construction site videos instead of physical site visit experiences. Despite the benefits of using videos to teach, research has shown that presenting videos only to students might not be sufficient for learning as relevant and irrelevant information are usually present in videos. Therefore, calling out relevant information in videos would enable students to focus on them, enhancing their learning. To this end, this study presented a video-based learning tool that instructors can utilize to provide students with site visit experiences. In the environment, important information are called out using boundary boxes and texts. To achieve this, first, the study identified the practical knowledge that industry experts would want students to know about construction sites. Then, the identified information was annotated in construction videos via the guidance of learning theories such as the cognitive theory of multimedia learning and dual coding theory. A usability evaluation was conducted to test if students could notice the annotated contents in the video. Afterward, individual differences such as gender, academic program, and mental workload that could contribute to how students would notice annotated information in construction videos were investigated. The study contributes to the practical concepts learners need to acquire to prepare them for the workforce. Additionally, this study proved that annotating important information in videos can direct student attention to those contents. Furthermore, to make learning environments flexible for different learners, this study identified the extent to which individual differences such as gender, academic programs, and cognitive loads can be recognized from the way learners notice annotated contents of videos. Finally, the outcomes of this study would make it possible for other researchers to further advance the potentiality of teaching with annotated videos in the construction domain and other related fields.

Thesis (MScEng)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: In percutaneous procedures, where the surgeon inserts a needle into a target in the patient, it is challenging to access the target at the rst attempt. Repeated needle insertion attempts are highly undesirable, as they lead to increased theatre time, patient haemorrhage and radiation exposure. The proposal of this thesis was to develop a low-cost system to aid surgeons with the process of positioning and orientating the needle prior to insertion. The proposed system uses a stereo pair of images produced by a standard C-arm uoroscope. By applying computer vision techniques, such as triangulation, the desired needle position and orientation are determined from the set of images. An articulated manipulator is used to position and orientate the needle. The surgeon makes selections on the images, via a graphical user interface (GUI), to indicate the desired position and orientation of the needle. Following these selections, an algorithm determines the necessary angles for the manipulator. The surgeon then positions the manipulator accordingly. Results from tests on a phantom showed the system to be repeatable and accurate to 2 mm. This is less accurate than similar, existing systems which, reported accuracies of 0:25 and 1:21 mm. However, 2 mm accuracy is considered adequate as it allows a range of percutaneous procedures to be performed such as needle biopsy, regional anaesthesia, brachytherapy and percutaneous nephrolithotomy (PCNL). Tests on both a phantom kidney and a porcine kidney indicated that the system can function in a real percutaneous procedure and with an anatomical target. The total cost to develop the prototype system was R15 000. The accuracy of the proposed system and the time to gain access to the target indicate that the system can be a bene cial aid to surgeons when performing percutaneous procedures. As minimal X-ray imaging is required, patients will also be spared excessive radiation exposure and theatre time. / AFRIKAANSE OPSOMMING: In perkutane prosedures, waar die chirurg 'n naald in 'n teiken op die pasiënt invoeg, is dit moeilik om met die eerste probeerslag toegang tot die teiken te bekom. Herhaalde pogings om die naald in te voeg is hoogs ongewens, aangesien dit tot verlengde teatertyd, bloeding by pasinte sowel as stralingsblootstelling lei. 'n Laekostestelsel is ontwikkel om chirurge by te staan met die proses om die naald voor invoeging te plaas en te oriënteer. Die stelsel gebruik 'n paar stereobeelde wat deur 'n standaard-C-arm- uoroskoop voortgebring word. Met behulp van rekenaarsigtegnieke, soos triangulasie, word die gewenste naaldposisie en -oriëntasie uit die stel beelde bepaal. 'n Geartikuleerde manipuleerder word gebruik om die naald te plaas en te oriënteer. Die chirurg voer die vereiste invoerdata deur middel van 'n gra ese gebruikerskoppelvlak (GGK) in, waarna 'n algoritme die vereiste hoeke vir die manipuleerder bepaal. Die chirurg plaas dan die manipuleerder dienooreenkomstig. Die resultate van toetse op 'n fantoom toon dat die stelsel herhaalbaar en tot op 2 mm akkuraat is. Dit is minder akkuraat as soortgelyke, bestaande stelsels, wat 'n akkuraatheid van 0:25 mm en 1:21 mm onderskeidelik aangemeld het. Tog kan akkuraatheid van 2 mm as voldoende beskou word, aangesien dit 'n verskeidenheid perkutane prosedures moontlik maak, soos naaldbiopsies, regionale anestesie, bragiterapie en perkutane nefrolitotomie (PCNL). Toetse op 'n fantoomnier dui daarop dat die stelsel wél in 'n werklike perkutane prosedure kan funksioneer. Toetse op 'n varknier toon dat die stelsel wél ook met 'n anatomiese teiken kan werk. Die totale koste om die prototipe te ontwikkel was R15 000. Die akkuraatheid van die voorgestelde stelsel en die tydsduur om toegang tot die teiken te bekom, dui daarop dat die stelsel 'n nuttige hulpmiddel kan wees vir chirurge wat perkutane prosedures uitvoer. Aangesien minimale Xstraalbeelding vereis word, sal dit pasiënte ook oormatige stralingsblootstelling en verlengde teatertyd spaar.

Thesis (MScEng)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: Augmented reality is a fairly young research field, still in an infancy stage at Stellenbosch University. Since this is one of the first augmented reality projects, one goal is to present a theoretical study of augmented reality. This study is given in the literature study, along with a review of the available development solutions. While there are various tools available with which one can create markerbased augmented reality applications, these tools are not meant for testing new techniques and algorithms in an augmented or mixed reality. The remaining goals of this project is to create a platform for the rapid design of augmented reality applications, and to expand the capabilities of this platform beyond marker-based augmented reality. In this project we present the design and implementation of a pragmatic mixed-reality framework capable of a wider variety of applications. A design is shown where marker tracking can be used alongside other computer vision techniques to design new applications. The framework utilises stereo cameras to find the position of real world objects, and a 3D display to make the mixed reality environment as immersive as possible. Proof of concept test applications built with the framework are presented. Colour based techniques are used to find a user’s hand and create a virtual representation of it. This allows the user to interact with a virtual object in an augmented reality scene by ‘touching’ it with her hand. / AFRIKAANSE OPSOMMING: Toegevoegde realiteit is ’n jong navorsingsveld by Universiteit Stellenbosch. Aangesien hierdie een van die eerste projekte is wat fokus op toegevoegde realiteit, is ’n teoretiese studie van toegevoegde realiteit as ’n doel gestel. Dit word verskaf in die literatuurstudie, tesame met ’n oorsig oor bestaande oplossings vir die ontwikkeling van toegevoegde realiteit sagteware. Bestaande oplossings is gefokus op die ontwikkeling van merker-gebaseerde toegevoegde realiteit, maar los min ruimte vir die toets van nuwe tegnieke toepasbaar op die veld. Dit lei tot die oorblywende doelwitte van die projek: om ’n platform te ontwerp vir die ontwikkeling van merker-gebaseerde toegevoegde realiteit programme, asook om die platform uit te brei. Ons lewer ’n pragmatiese ontwikkelingsraamwerk wat dit moontlik maak om ’n verskeidenheid nuwe toegevoegde realiteit programme te ontwikkel. Die raamwerk is ontwerp sodat die ontwikkelaar merkers saam met ander rekenaarvisie tegnieke kan gebruik om sagteware te skep. Stereo kameras word gebruik om die posisie van werklike voorwerpe te vind. Die raamwerk maak ook gebruik van ’n 3D skerm om virtuele objekte te vertoon. Toetsprogramme gebou as ’n bewys van die konsep, word vertoon en bespreek. ’n Kleur-gebaseerde tegniek word gebruik om ’n gebruiker se hand te vind, en ’n virtuele voorstelling van die hand word geskep. Die gebruiker kan virtuele voorwerpe laat reageer deur dit met haar hand aan te raak.

<div> <div> <div> <p>In this thesis work, we propose computer vision techniques for 3D reconstruction and object height measurement using a single camera and multi-laser emitters, which have an intersection on the projected image plane. Time-division and color division methods are first investigated for our structured light system. Although the color division method offers better accuracy for object height measurement, it requires the laser emitters equipped with different color lights. Furthermore, the color division method is sensitive to light exposure in the measurement environment. Next, a new multi-level random sample consensus (MLRANSAC) algorithm has been developed. The proposed MLRANSAC method not only offers high accuracy for object height measurement but also eliminates the requirement for the laser emitters with different colors. Our experiment results have validated the effectiveness of the MLRANSAC algorithm. </p> </div> </div> </div>