Realistic Simulations for Evaluating Vehicular Sensing and Communication Networks
- Faculty lead: Prof. Bhaskar Krishnamachari
- Description: Develop data-driven inference algorithms and micro-simulation tools that use real-world traffic sensor and gps data traces to evaluate new city-scale vehicle-based sensing applications as well as vehicle-to-vehicle and vehicle-to-infrastructure communication protocols.
A Game Theoretic Approach to Maximize Information Diffusion on Social Networks
- Faculty lead: Prof. Milind Tambe
- Description: The goal of this project is to develop efficient game theoretic approaches to identify influencer(s) and maximize the information diffusion in social networks with focus on evacuation planning.
- Faculty lead: Prof. Gerard Medioni and Prof. Ram Nevatia
- Description: Persistent surveillance: We utilize crowdsourced mobile videos and CCTV videos to track targets of interests, such as people or vehicle, over large space and time. To achieve persistent surveillance, inferring the spatial relationships between the video’s Field-Of-Views, FOVs, is necessary. Thus, we developed a Geospatial Image Filtering Toolbox (GIFT) that acquire, store, index and access large amount of FOVs, as spatial objects, effectively and efficiently in support of persistent surveillance. Without GIFT, if one target is about to exit the FOV of one camera, the current computer vision technologies have to scan the re-appearance of this target in all videos, which is time-consuming and error-prone. GIFT makes use of geospatial database technologies to actively select a small number of videos which may cover this target. With GIFT, the overhead of image/video analytics can be reduced significantly.
Spatiotemporal Mobility Data Analytics
- Faculty lead: Prof. Antonio Ortega
- Description: We continue our research on wearable sensors analytics assuming sensors that provide spatial information over time such as position, acceleration, or gyroscopic data. The goal of this project is to develop efficient data analysis tools that generate higher-level measurements of mobility for medical applications. We will extend our previous research on de-noising and compression techniques to extract data measurements by exploiting data symmetry and characteristics of the movement sensed.
Large-scale Machine Learning Models for Spatio-temporal Data Mining
- Faculty lead: Prof. Yan Liu
- Description: Develop machine learning models for a large scale sptio-temporal big data such as people’s locations and social behaviors.
- Faculty lead: Prof. Hao Li
- Description: Create a communication system where the facial performance of one party is captured in real time using a depth sensor and the other party is wearing an HMD device. Capture brain activity so that even the one who is wearing an HMD device can have his face captured.