Research & Projects
Spatiotemporal Data Mining
In this research, several supervised learning models are applied to select “salient” objects as landmarks. Features are extracted from various sources, including traditional databases, social networks and LiDAR data. The features do not only represent static aspects of the object, but also the interactions between objects and human. [Slides]
“Trending” is a property to represent the popularity of venues in social networks (e.g. Foursquare). This research concentrated on exploring the patterns of trending venues. Methods like simple exponential smoothing model, halt-winters exponential smoothing model, ARIMA model, as well as some supervised models are explored in this work. [Slides, Report]
This research endeavors to explore the relations between the cerebral blood flow and brain activities, which is a sub-topic of brain computer interface (BCI). The goal is to use the cerebral blood flow as an indirect measure of the brain activities. Three states are considered in this work: geometric, verbal and rest tasks. Artificial neural network and support vector machines are used to classify these three states. [Slides, Report]
How to provide navigation services to visually impaired persons is always a challenge. Due to the low vision, visually impaired persons could not benefit from visual cues on the route. In this work, auditory, tactile, olfactory and thermal landmarks are explored to supplement this shortage. Further research on how to collect these “special” landmarks is still in process. [Slides]
Provided is a novel application of visual encoding and topology in the understanding of user-behavior in a simple, local task of network centrality. Coupled with eye-tracking, this work provides both a fundamental metric called scanpaths and a novel eye-tracking metric called weighted error measure to combine both the visual encoding and topology of the network to better the understanding of networks at the visceral level.
Personalized Accessible Maps is a project focusing on providing navigation services to a wide range of users in campus. The prototype of this project is implemented at the University of Pittsburgh. The service provides three kinds of route for different users, including shuttle path, shortest path and personalized path. For the personalized path, factors like slope, traffic, width of the edge and so on are considered in generating the route. In addition, both turn-by-turn and landmark-based directions are implemented in this service. Different POIs, service areas, accessible entrances and sidewalks could also be demonstrated on the maps.
This is the mobile version of the Personalized Accessibility Maps. All the functions are converted from the web vertion to this mobile version. It has been tested on various browsers including Firefox, Safari, Explore 6 and Chrome.
This project aims to build a navigation system for pedestrians. So the sidewalk, rather than road networks, is considered to generate the route. Due to the smaller granularity of pedestrian navigation compared to vehicle navigation, more challenges are imposed to the techniques of map matching. It requires the results of map matching to be highly accurate. To overcome this issue, a new algorithm, chain-code map matching, is implemented in this system. This technique largely improved the accuracy of the positioning of pedestrians.
Tools: Java, Android API