Asia Faculty Summit 2014

Our vision is to build a smart campus that is powered by big data. This demo shows how we utilized different sources of campus data, as well as students’ footprints on a social network, to provide humanity care to students at the University of Science and Technology of China (USTC). By analyzing shopping and check-in records captured by smart card, we detect students who are poor to provide them a monthly subsidy. Combining this data with school email, library data, and data from social networks, we can reconstruct an individual student’s lifestyle, such as when they sleep and whether they have breakfast regularly. We built a website to show personal statistics and life data and to provide specific recommendations to students. We also developed a series of cross-platform mobile applications, such as campus map and bus schedule, to help freshmen quickly get used to university life.

The traffic flow on urban road networks and subway systems can be modeled as a spatio-temporal graph (STG). An STG is a directed graph in which vertices and edges are associated with spatio-temporal properties. In this poster, we detect interesting phenomena, titled black holes and volcanos, from an STG. Specifically, a black hole is a subgraph (of an STG) that has the overall inflow greater than the overall outflow by a threshold, while a volcano is a subgraph with the overall outflow greater than the overall inflow by a threshold. In our method, we build an STG index to maintain the spatio-temporal properties of an STG. Based on the index, we propose a two-step black hole detection algorithm. The first step identifies a set of candidate grid cells to start from, and the second step expands an initial edge in a candidate cell to a black hole and prunes other candidate cells after a black hole is detected. Then, we adapt this detection algorithm to a continuous black hole detection scenario. We evaluate our method based on real GPS trajectories of 33,000 taxis and the Beijing road network, finding traffic anomalies and regular travel patterns in the city.

The clustering algorithms are applied on the crowdsourced data to identify the location of the road anomalies. By this means, it is possible to achieve pervasive road pavement monitoring with low deployment cost. The demo is presented in the form of a video that demonstrates a test run on the NCTU campus. The video shows four subscreens, including ground truth, anomaly map, bumping sensor, and oscillation waveform. The ground truth window shows the field-trial driving environment. The anomaly map window shows the detected anomalies on the digital map. The bumping sensor window shows the GPS and vibration data, and the last detected bumping event. The vibration waveform window shows the waveform of vertical acceleration components calculated g-sensor data.

To help reduce CO₂ emission from road traffic, we have constructed a system that prompts people to adopt eco-friendly travel behavior by modeling and showing regional traffic. The traffic flow is modeled by using street-side cameras, and CO₂ emissions in regional areas are estimated by traffic simulations. The CO₂ emissions are visualized with VR/MR technologies and accordingly reduced by eco-friendly behaviors. Currently, we are developing 3D city modeling techniques for more accurate traffic simulations. In this presentation, we introduce our projects in urban scenes.

Prototyping is a very important in designing electronic circuits. For a long time, engineers have been using breadboard for rapid prototyping. Although breadboard is convenient on a small scale, it can become entangled and time consuming for more complex circuits. To eliminate these challenges, at the University of Tokyo we have established our own unique designing method. We have succeeded in printing electrical circuits on flexible materials such as paper and PET film, by using a regular household inkjet printer and commercial silver nanoparticle ink manufactured by Mitsubishi Paper Mills Ltd. In one minute, we have a neat and clean printed electronic circuit. Currently, we are also working on prototyping of a multi-layered circuit on paper. Starting with a double-sided circuit, we are developing a process to fabricate via hole for an instant inkjet circuit. This via hole can be made by laser cutter or specially designed needles. In the future, we will implement a method to fabricate multi-layered circuit by using silver nanoparticle ink and an inkjet printer.

Code Hunt is an educational coding game that runs in a browser. The game consists of a series of worlds and levels that get more challenging. In each level, the player has to discover a secret algorithm and write code for it. The game has sounds and a leaderboard to keep users engaged. Code Hunt targets teachers and students from introductory to advanced programming or software engineering courses. In addition, Code Hunt can be used by seasoned developers to hone their programming skills or by companies to evaluate job candidates. It has also been used in contests such as Beauty of Programming and Imagine Cup. At the core of the game experience is an automated program analysis and grading engine that is based on symbolic execution, Pex. The engine spots errors in the code and detects any differences between the code and the secret algorithm. Code Hunt was developed by the Research in Software Engineering (RiSE) group in collaboration with Microsoft Research Outreach at Microsoft Research.

An increasing number of research areas rely on collecting data from sensors and devices that are deployed where people live, work, and play. Researchers typically deploy sensors and devices in a few homes or workspaces, collect data, analyze the data, and make interesting inferences based on this data.

Healthcare and energy management are two examples of such areas. This hands-on demo of the Lab of Things uses off-the-shelf sensors and cameras. We will also demonstrate features that allow device deployments based on the Lab of Things to scale up in geographically dispersed locations.

Most smart home applications today are either ad hoc or closed/monolithic—so it is critical to study novel software architecture/frameworks that are open and that connect multiple heterogonous devices and multiple networks. These frameworks need to support multiple concurrent applications, enable reliable data/command delivery among home gateway and cloud/smartphone, and provide easy-to-develop third-party service business applications. After investigating existing software solutions and standards, we propose a service framework for the smart home that combines HomeOS, MQTT, and Azure cloud. Several new drivers and applications have been developed on top of HomeOS for building a smart home gateway. MQTT is used for reliable and scalable message delivery between gateway and cloud, and push notification to Android devices. We also develop simple multi-tenant mechanism based on the Azure platform.

In this demo, we present a system of facial beautification that is deployed on the Microsoft Azure cloud platform. Facial beautification is a novel computational photography technique to enhance the aesthetic appeal of an image of a human face while maintaining high similarity to the original. The system uses a new adaptive region-aware masks generation method, together with other image processing technologies such as image layers decomposition, facial landmarks detection, data-driven facial image synthesis, and fusion. The result is an effective, convenient, and flexible tool and a fantastic way for specific facial beautification in terms of skin smoothness, shape beautification, lighting, and color enhancement. We will also present a local version of this system that runs on Windows in case there is limited Internet network bandwidth or access.

We are experiencing the sixth mass extinction of plants and animals, and an effort is needed to protect the biodiversity of our planet. In addition to the information available from specimens and observations, there is a significant amount of information about the temporal distribution of species that can be extracted and processed from scientific literatures. However, this requires a high-performance environment to store and process the huge amounts of data, including more than 100 million records on 40 million pages. More real-time biodiversity data can be obtained from the websites of journals, news, botanical gardens, protected areas, and the communities of citizen science. This information should be integrated, analyzed, and displayed in a cloud environment and that enables external users to interact with it. There are also some new techniques that should be introduced, such as machine learning, natural language processing, and GIS.

With the advancement of sensors and information technologies, a large amount of geographical information and resources (GIRs, including geodata, algorithms, and models) have become available on the Internet. However, the heterogeneous nature and the complexity of the Internet make it a challenge for GIS researchers and teachers to discover and utilize demanded online GIRs. Getting the distributed online GIRs to working collaboratively for research and teaching has become an urgent problem.

We designed GeoSquare, a collaborative GeoProcessing framework based on Azure. Through GIRs sharing and GeoProcessing orchestration, GeoSquare can help researchers and teachers search for and utilize online resources in an efficient, harmonious way. GeoSquare was implemented based on web technologies such as the enterprise web portal, Rich Internet Applications, Java User Interface toolkits, and workflow engines. Most of these technologies are built on open source standards to ensure the delivery of content-rich and cross-platform applications.

This study emphasizes the eco-efficiency of urban passenger transportation, which can be defined as the production of maximum benefits to society with minimized environmental impacts from the use of energy and materials. Conventionally, the measure of traffic and transportation eco-efficiency has mainly been considered as minimizing parameters such as primary energy per passenger-km, or grams of carbon dioxide or other vehicle pollutants such as nitrogen oxides or carbon monoxide, per vehicle-km. To design the next generation navigation system, we must consider how user behavior can help maximize eco-efficiency rather than keeping the focus on the traditional aspects such as emission. This is possible only with the support of pervasive computing technologies and big data computations in the cloud. Requirement: No special.

Stephen Jia Wang, Monash University

Asymmetry of information interactions is one of the most important determining factors in economic and cultural imbalance in different areas. In China, there are 56 ethnic groups, more than 80 languages, and about 30 different characters. Most languages have not been supported by major translation providers yet. This limits the access that Chinese minority ethnic groups have to global information and knowledge. We want to build a simple but efficient translation framework for Chinese minority ethnic languages that can build a complex SMT system in an easy way.

In the research of communication, studying how people use hand gestures during conversations is challenging. Common methods such as video taping and manual coding can be expensive, unreliable, and unscalable. In this demo, we showcase our approach of using Kinect’s skeleton tracing capability to capture hand movements during conversations and to derive key metrics such as number of gesture used by individuals, and we look at the interpersonal similarity of gestures between interlocutors. The approach supports the study of the non-verbal aspects of conversation and enables new possibilities for interaction and communication design, such as evaluating new communication tools in the field and using feedback to shape communication processes based on the quantitative metrics of gestures.

This demo will show a tele-immersive system that includes both remote haptic handshaking and interactive visual face tracking. The proposed system has a physical human-like avatar with a handshaking robotic arm/hand and an accurate face pose correction algorithm with multiple cameras. The handshaking robotic arm/hand can provide various haptic sensations, such as shaking forces, hand grip pressures, and temperature, while handshaking over any networks by using a robust haptic tele-manipulation technique that can overcome time-varying natures and uncertainties in the networks and humans. In addition, realistic eye contact is made by an accurate face pose estimation algorithm, and the texture of the face is reconstructed and rendered based on the estimated pose information. A live demo of the system will be provided between two different locations in China and South Korea.

Sangyoun Lee, Yonsei University

We developed a tactile vision substitution system that is composed of an electro-tactile display, optical sensors beneath each electrode, and a smartphone with a camera and an LCD. The smartphone acquires the surrounding view, conducts image processing, and displays the image on the LCD. The image is captured by the optical sensors and converted to a tactile image by the electro-tactile display. Combining the commonly available mobile device and electro-tactile display enables a low-cost yet powerful and compact system.