Virtual Reality & Intelligent Hardware
Virtual Reality & Intelligent Hardware
News » More
Bags of tricks for learning depth and camera motion from monocular videos
Based on the seminal work proposed by Zhou et al., much of the recent progress made in learning monocular visual odometry, i.e., depth and camera motion from monocular videos can be attributed to the tricks in the training procedure, such as data augmentation and learning objectives. In this paper, we will categorize a collection of such tricks with theoretical examination and empirical evaluation of their impact on the final accuracy of the visual odometry. We find that, by combining these tricks together, we are able to improve a baseline model adapted from SfMLearner significantly without extra inference costs. We also analyze the principles behind these tricks and why they are successful, offering practical guidelines for future research.
Application and research trends of digital human models in manufacturing industry
Virtual reality (VR) has been widely used in manufacturing industry, and VR-based virtual manufacturing has received extensive attention and research in this new intelligent manufacturing era. Among them, digital human models (DHMs), being an important tool and method, is essential and necessary for virtual manufacturing. Moreover, the application and research of DHMs have become an important academic research field. This paper aims to identify the application and research trends of DHMs in manufacturing industry, and to provide certain reference for the development of virtual manufacturing and DHMs. First, a total of 49 related articles are filtered from a large number of publications published between 2014 and 2019. Then, the application of DHMs in manufacturing industry is analyzed from different aspects, and some relevant technical limitations are discussed. The results indicate that the application of DHMs has notable difference in different fields and types. Automotive industry is the main application field, while assembly/maintenance simulation and evaluation is the main application type. Meanwhile, there are still some shortcomings in the establishment of virtual environment, motion control and evaluation of DHMs, which should be further improved. Finally, research trends of the application of DHMs are illustrated and discussed, i.e., planning and assessment of human-robot collaboration system, combination with AR, improved motion planning of DHMs. To sum up, the application of DHMs improves the realistic and effectiveness of virtual manufacturing, and it will be more widely and deeply studied and applied in manufacturing industry.
A survey of 3D modeling using depth cameras
3D modeling is an important topic in computer graphics and computer vision. In recent years, the introduction of consumer-grade depth cameras has brought profound advances in 3D modeling. Starting with the basic data structure, this survey reviews the latest developments of 3D modeling based on depth cameras, including research works on camera tracking, 3D object and scene reconstruction and high-quality texture reconstruction. We also discuss the future works and possible solutions for 3D modeling based on depth camera.
Edge vector based large graph visualization and interactive exploration
The demand for graph analysis is increasing. High quality and high readability graph layout is important for graph analysis. In the past years, we investigate this topic and propose a unified framework for graph layout and exploration. This framework maintains the readability during layout and interaction process. It controls the edge lengths and directions instead of only lengths. We can model most existing layout constraints, as well as develop new ones. For interactive exploration on the detail of a graph, we extend our framework to a new focus + context fisheye view. Traditional fisheye views for exploring large graphs introduce substantial distortions that often lead to a decreased readability of paths and other interesting structures. We use edge directions as constraints for graph layout optimization allows us not only to reduce spatial and temporal distortions during fisheye zooms, but also to improve the readability of the graph structure. Furthermore, the framework enables us to optimize fisheye lenses towards specific tasks and design a family of new lenses. We implement our framework with GPU parallel computing, which allows us process large graphs with up to 10,000 nodes at interactive rates.