Font Size: a A A

Image-driven Label Placement For Augmented Reality

Posted on:2020-09-06Degree:MasterType:Thesis
Country:ChinaCandidate:J Q JiaFull Text:PDF
GTID:2428330575475538Subject:Probability theory and mathematical statistics
Abstract/Summary:PDF Full Text Request
Augmented Reality(AR)is a popular technology that integrates real objects in real scenes with computer-generated virtual annotations,which can significantly make users better understanding and feeling about their view.In order to make the fusion scene more real and understandable,determining the optimal location relationship between virtual annotation and real objects in Augmented Reality it is a challenging problem.We aim to obtain the best location of virtual annotations.In this paper,we introduce a label placement technique for Augmented Reality in street view.One of the common challenges in Augmented Reality applications is lacking in knowledge of the real environment,limiting efficient representation and optimal layout of the digital information augment onto user's view.In order to overcome this problem,we proposed a semantic-aware label placement method,which combines identifying potentially important image regions through semantic information and manual placement tendencies study.Our method brings the following contributions:1.Our method firstly integrates the semantic information into the label placement optimization.We introduce a new feature map called Guidance Map.In addition to using saliency information of the image,we add the semantic information and task-specific importance prior to make the importance map more reasonable.2.We manually collected a label placement dataset.Different from former task-unaware saliency detection,the dataset provides the task specific adjustment in our system.Another benefit of the dataset is serving as a quantitative evaluation benchmark.3.We propose a spatiotemporal coherence-based label placement method.Firstly,we formulated the requirements for Augmented Reality.Then we define an energy function to transform the label placement problem into an optimization problem.4.We apply four evaluation metrics to evaluate the performance of annotation results.And both qualitative and quantitative experiment results proved our method better than other label placement methods.Also,the label placement algorithm framework we proposed can be used to many applications in the domain of Augmented Reality,for example,the AR street view navigation.
Keywords/Search Tags:Augmented Reality, Image Processing, Label Placement, Saliency Detection, Semantic Segmentation
PDF Full Text Request
Related items