الفهرس | Only 14 pages are availabe for public view |
Abstract Occlusion awareness is one of the most challenging problems in several fields such as multimedia, remote sensing, computer vision, and computer graphics. Realistic interaction applications such as Mobile Augmented Reality are suffering from dealing with occlusion and collision problems in a dynamic environment and response in realtime. The problem appears when adding virtual content to a physical scene, it is mandatory to know what physical objects are in the scene and where exactly they are in the real world. This is needed to determine what needs to be occluded and render content accurately. Creating dense three-dimensional (3D) reconstruction methods is the best solution to solve this issue. However, these methods have poor performance in practical applications due to the absence of accurate depth, camera pose, and object motion. Moreover, creating a full 3D reconstruction is computationally intensive and can become a bottleneck for real-time Augmented Reality applications. In this thesis, a novel framework has been proposed to build a full 3D model reconstruction and overcome the occlusion problem in a complex dynamic scene without using sensors’ data and uses the most popular camera in mobile phone in real-time. The proposed framework can solve many problems and it is considered suitable for Realistic interaction applications such as Mobile Augmented Reality. The main objective of the proposed framework is to create a smooth and accurate 3D point-cloud for a dynamic environment using cumulative information of a sequence of RGB video frames. The framework is composed of two main phases. First, an unsupervised learning technique is used to predict; scene depth, camera pose, and objects’ motion from RGB monocular videos. Second, a frame-wise point cloud fusion is generated to reconstruct a 3D model based on a video frame sequence in real-time. A massive Graphical Processing Unit (GPU) is used to speed up the creation of a 3D point-cloud.everal evaluation metrics are measured to present the accuracy of the predicted pointcloud. Moreover, the proposed framework is evaluated with different widely used stateof-the-art evaluation methods. Experimental results show that the proposed framework surpassed the other methods and proved to be a powerful candidate in 3D model reconstruction in real time. |