Research Experience

Senior Engineer, Qualcomm Atheros Inc.

Positioning Technology Group.

Working on GNSS and other positioning technology development to bring robust, cost-effective commercial positioning solutions to market, esp. mobile and other embedded device markets.

Research Assistant, MURI project (ISIS, Vanderbilt University) Jan 2007-Jan 2010

Model-based Target Tracking using Video Features.

Model-based tracking allows the use of prior knowledge of the shape and appearance of specific targets in the inference of incoming image data. The target model consists of shape and appearance models for the target (e.g. a parametrized 3D shape model). Using the known camera and scene geometry, the data likelihood can be computed for the incoming image data. Standard Bayesian methods can be used for target state and parameter estimation. In feature-level fusion, instead of communicating the entire image data, several visual features are extracted from the images that are used to compute data likelihood.

Feature-based Collaborative Localization and Discrimination in WSNs.

Collaborative localization and discrimination of multiple acoustic sources is an important problem in Wireless Sensor Networks (WSNs). Approaches for source localization can be categorized as signal-based and feature-based methods. The signal-based methods are not suitable for collaborative localization in WSNs because they require transmission of raw acoustic data. In the feature-based methods, signal features are extracted at each sensor and the localization is done by multisensor fusion of the extracted features. Such methods are suitable due to their lower bandwidth requirements compared to the signalbased methods.

In this project, I developed a feature-based localization and discrimination approach for multiple harmonic acoustic sources in WSNs. The approach uses acoustic beamform and Power Spectral Density (PSD) from each sensor as the features. A probabilistic graphical model is used for multisensor fusion, localization and discrimination.

Fusion-Based Localization in HSNs.

Heterogeneous Sensor Networks (HSNs) are becoming more commonly used for purposes such as monitoring and surveillance, as they offer richer sources of data for situational awareness.

In this project, we developed a method for localizing a network of cameras equipped with wireless radios. The method fuses both the image data and radio interferometry data to estimate the position and the orientation of each camera. The method exploits both the image and radio data for a more computationally efficient process of localization. A linear and a nonlinear approach are developed for fusing the data which depend on different constraints on the network. We demonstrate the approach on a real network of camera and radio nodes.

Multimodal Target Tracking using HSNs.

In this project, we developed a system for multimodal target tracking in urban environments using an HSN of mote class devices equipped with acoustic sensor boards and embedded PCs equipped with webcameras. The targets to be tracked are moving vehicles emitting engine noise. The system uses acoustic beamforming and motion detection for audio and video sensors, respectively. Development of such systems require stringent real-time processing and communication bandwidth usage.

The system operates online in real-time at 4Hz, thus addressing the real-time processing requirement. The audio and video sensors compute local features and communicate them with the fusion node, thus addressing the limited communication bandwidth. Two different approaches are taken for multisensor information fusion based on Sequential Bayesian Estimation (SBE) and Monte Carlo Markov Chain Data Association (MCMCDA) algorithm. Our system has many components including audio processing, video processing, WSN middleware services, multimodal sensor fusion, and target tracking based on SBE and MCMCDA. The main challenge we addressed is system integration as well as making the system work on the actual platform in a realistic deployment scenario. Experimental results from a deployment in an uncontrolled urban environment are used to demonstrate our approach.

Summer Intern, VERDE project (Oak Ridge National Laboratory, TN) Summer 2007

Wide-Area Situational Awareness in Google Earth using an Amination Web-Service.

Google Earth has provided the users, experts and laymen alike, a fast and easy way to access and visualize geospatial data and analysis. Network-link feature in Google Earth has provided an easy way of integrating web-services developed by various GIS experts to visualize and present their data and analysis in spatial context in Google Earth.

In this project, I developed a web-service that generates snapshot animation for Google Earth. The basic idea is that the users, as well as the experts, can submit spatially- and temporally-tagged geographic data to the web-service, which then generates a Google Earth animation using the data. The animation of such geographic data from all over the region provides a better spatial and temporal awareness of the region.

Research Assistant, SOA group (ISIS, Vanderbilt University) Aug 2005-Dec 2006

Service-Oriented Middleware and Programming Framework for HSNs.

Heterogeneous sensor networks are comprised of ensembles of small, smart, and cheap sensing and computing devices that permeate the environment, as well as resource intensive sensors such as satellite imaging systems, meteorological stations, and security cameras. Emergency response, homeland security, and many other applications have a very real need to interconnect these diverse networks and access information in real-time. Web service technologies provide well-developed mechanisms for exchanging data between heterogeneous computing devices, but they cannot be used in resource-constrained wireless sensor networks.

In this project, we developed a lightweight service-oriented architecture for sensor networks, called OASiS, which provides dynamic service discovery and can be used to develop ambient-aware applications that adapt to changes in the network and the environment. An important advantage of OASiS is that it allows seamless integration with web-services. We have developed a middleware implementation that supports OASiS, and a simple tracking application to illustrate the approach. Our results demonstrate the feasibility of a service-oriented architecture for wireless sensor networks.

Research Assistant NEST Lab (ISIS, Vanderbilt University) Aug 2003-Aug 2005

Self-Localization Service for Sensor Nodes in WSNs.

In this project, I developed present a sensor node localization method using a mobile acoustic beacon. The technique is passive in that the sensor nodes themselves do not need to generate an acoustic signal for ranging. This saves cost, power and provides stealthy operation. Furthermore, the beacon can generate much more acoustic energy than a severely resource constrained sensor node, thereby significantly increasing the range. The acoustic ranging method uses a linear frequency modulated signal that can be accurately detected by matched filtering. This provides longer range and higher accuracy than the current state-of-the-art. The localization algorithm is especially designed to work in such acoustically reverberant environment, as urban terrain. The algorithm handles non-Gaussian ranging errors caused by echoes. Node locations are computed centrally by solving a global non-linear optimization problem in an iterative and incremental fashion.

Project Associate, TRIPP (IIT Delhi, India) Jul 2002-May 2003

The TRIPP center at IIT Delhi is an interdisciplinary effort doing research in safer transportaion systems, vehicle design, vehicle safety equipments and transportation infrastructure for the future.

In this project, I developed and tested a validation scheme for airbag systems. The challenges in the project are to develop methodology to generate a folded finite element mesh for an airbag and then validating it with the available experimental data. The mesh-folding algorithms are implemented in MATLAB, and the simulations are run in PAM-CRASH.

Undergraduate Research Project, Mechatronics Lab (IIT Delhi, India) Aug 2001-May 2002

In this undergraduate research project under the guidance of Prof. Sudipto Mukherjee, I studied the Massively Parallel Binary Systems (MPBS). MPBS is a concept of deploying a high number of small actuators in parallel to achieve desired output motion. The use of binary actuator ensure that the control remains computationally tractable. I analyzed, simulated and developed a prototype implementation of one-dimensional MPBS using solenoids.