Task Allocation in Dynamic Environments
- Faculty lead: Prof. Ketan Savla
- Description: Design and development of algorithms for task allocation strategies for groups of autonomous or human supervised mobile agents. The approach will enable agents to complete tasks in uncertain and dynamically changing environments, where new task requests are generated in real-time over a geographically dispersed region. The applications of this research include crowd sourcing, pickup(/delivery) tasks in transportation networks, as well as surveillance and monitoring missions.
Facial Expression Recognition
- Faculty lead: Prof. Gerard Medioni
- Description: This project develops key algorithms for a realtime facial expression recognition of consumers as a part of the customer feedback analysis framework in B2C (Business to Consumer).
Study of Congested Corridors in Los Angeles
- Faculty lead: Prof. James Elliott Moore, II
- Description: Study the most congested corridors of Los Angeles County using historical traffic sensor data, and make before and after analysis of Carmegedons.
Analysis of Mobility Data Using Spatiotemporal Graph-based Techniques
- Faculty lead: Prof. Antonio Ortega (Electrical Engineering)
- We consider the scenario where multiple body-attached sensors are used and assume that the data provided by these sensors will be noisy. We will start by analyzing the data captured by real sensors in order to develop a better understanding of noise sources and characteristics. We will then study several de-noising approaches. The first one will be purely time based and will seek to develop techniques (e.g., based on Wavelets) to smooth the temporal trajectories of data at each sensor without removing information that will be important for evaluation and diagnostic. Then, we will also consider graph-based techniques to improve data quality by taking into consideration constraints on the sensor position due to the fact that they are attached to the body. As an example, two sensors attached to an individual's arm will be limited in the extent of their relative motions. We will develop methods were the system is initially calibrated under controlled circumstances, and then data acquired regarding relative sensor positions is used for de-noising. We further plan to explore the potential benefits of recently introduced graph wavelets which can capture all relevant information about body movement as either vertex data (sensor information) or edge data (distance between sensors). In addition to considering the de-noising problem, we will also study lossless and lossy techniques to compress the information generated by the sensors in order to make it easier to store complete records of sensor motion during extended periods of time. The compression techniques will be application-specific, with the goal of preserving key information for signal analysis.