Research
Design of Health Systems Technology for Developing Countries
An aspect of this project involves the design and implementation of a web-based tool to monitor the collection, screening and distribution of blood in developing countries. Another portion of the project involves the design of a blood flow system to montior, predict and allocate blood units in developing countries. This research is conducted in collaboration with the Centers for Disease Control and Prevention(CDC) and Zambia National Blood Transfusion System( ZNBTS).
Activity Based Approach to Functional Object Recognition
This project explores an approach to object and activity recognition based on detecting and analyzing the sequence of objects that are manipulated by the user. This approach to activity recognition senses the nouns(objects) instead of the verbs(ongoing movements or actions). In order to scale to many activities and objects, it is necessary to minimize the amount of human-labeled data that is required for modeling; this is done by relying on RFID sensors. RFID sensors can be used to acquire a sufficient amount of labeled training data albeit noisy and sparse. Therefore the RFID sensors are used in conjunction with vision(image) information to create models for objects. This approach leverages readings from RFID tagged objects and a common-sense model about which objects are likely to be used during an activity to bootstrap the learning process. A Dynamic Bayesian Network model is used to combine RFID and image data to jointly infer the most likely activity and object labels. We showed that activity recognition based on object use is viable by testing on a real world data set consisting of 16 activities and 33 objects. We obtained 80.67% and 72.36% activity and object recognition rates respectively.
Human Identification by Gait
Human identification is important because of its impact in surveillance and other security features. Gait based identification is distinguishing people using their wlaking patterns. This research focuses on using dynamic silhouettes as the basis for identification. A database with videos of a number of people walking was created and from that, silhouettes were extracted for use in computation and analysis of gait characteristics.
Implementation and Evaluation of a Peripheral Display
This research focuses on the design, implementation and testing of a ticker display to determine its usefulness and evaluation ratings by potential users. The ticker displays current news headlines and is updated about every three minutes. A pilot study for the ticker display was conducted to generate feedback about the ticker performance and effectiveness of delivering information. The results of the study indicate users awareness of the information displayed on the ticker and provide user feedback about the ticker peripheral display. It also provided feedback on study design for future iterations. The study provides valuable information to display builders about how users perceive and interact with a ticker display, which can lead to better design decisions.