Robotic Cameraman for Augmented Reality based Broadcast and Demonstration

In recent years, a number of large enterprises have gradually begun to use vari-ous Augmented Reality technologies to prominently improve the audiences’ view oftheir products. Among them, the creation of an immersive virtual interactive scenethrough the projection has received extensive attention, and this technique refers toprojection SAR, which is short for projection spatial augmented reality. However,as the existing projection-SAR systems have immobility and limited working range,they have a huge difficulty to be accepted and used in human daily life. Therefore,this thesis research has proposed a technically feasible optimization scheme so thatit can be practically applied to AR broadcasting and demonstrations. Based on three main techniques required by state-of-art projection SAR applica-tions, this thesis has created a novel mobile projection SAR cameraman for ARbroadcasting and demonstration. Firstly, by combining the CNN scene parsingmodel and multiple contour extractors, the proposed contour extraction pipelinecan always detect the optimal contour information in non-HD or blurred images.This algorithm reduces the dependency on high quality visual sensors and solves theproblems of low contour extraction accuracy in motion blurred images. Secondly, aplane-based visual mapping algorithm is introduced to solve the difficulties of visualmapping in these low-texture scenarios. Finally, a complete process of designing theprojection SAR cameraman robot is introduced. This part has solved three mainproblems in mobile projection-SAR applications: (i) a new method for marking con-tour on projection model is proposed to replace the model rendering process. Bycombining contour features and geometric features, users can identify objects oncolourless model easily. (ii) a camera initial pose estimation method is developedbased on visual tracking algorithms, which can register the start pose of robot to thewhole scene in Unity3D. (iii) a novel data transmission approach is introduced to establishes a link between external robot and the robot in Unity3D simulation work-space. This makes the robotic cameraman can simulate its trajectory in Unity3D simulation work-space and project correct virtual content. Our proposed mobile projection SAR system has made outstanding contributionsto the academic value and practicality of the existing projection SAR technique. Itfirstly solves the problem of limited working range. When the system is running ina large indoor scene, it can follow the user and project dynamic interactive virtualcontent automatically instead of increasing the number of visual sensors. Then,it creates a more immersive experience for audience since it supports the user hasmore body gestures and richer virtual-real interactive plays. Lastly, a mobile systemdoes not require up-front frameworks and cheaper and has provided the public aninnovative choice for indoor broadcasting and exhibitions.