Vinod Kulathumani

Kansei: a testbed for sensing at scale

The Kansei testbed at the Ohio State University is designed to facilitate research on networked sensing applications at scale. Kansei embodies a unique combination of characteristics as a result of its design focus on sensing and scaling: (i) Heterogeneous hardware infrastructure with dedicated node resources for local computation, storage, data exfiltration and back-channel communication, to support complex experimentation, (ii) Time accurate hybrid simulation engine for simulating substantially larger arrays using testbed hardware resources, (iii) High fidelity sensor data generation and real-time data and event injection, (iv) Software components and associated job control language to support complex multi-tier experiments utilizing real hardware resources and data generation and simulation engines. In this paper, we present the elements of Kansei testbed architecture, including its hardware and software platforms as well as its hybrid simulation and sensor data generation engines

Hibernets: Energy-Efficient Sensor Networks Using Analog Signal Processing

Preprocessing of data before transmission is recommended for many sensor network applications to reduce communication and improve energy efficiency. However, constraints on memory, speed, and energy currently limit the processing capabilities within a sensor network. In this paper, we describe how ultra-low-power analog circuitry can be integrated with sensor nodes to create energy-efficient sensor networks. To demonstrate this concept, we present a custom analog front-end which performs spectral analysis at a fraction of the power used by a digital counterpart. Furthermore, we show that the front-end can be combined with existing sensor nodes to 1) selectively wake up the mote based upon spectral content of the signal, thus increasing battery life without missing interesting events, and to 2) achieve low-power signal analysis using an analog spectral decomposition block, freeing up digital computation resources for higher-level analysis. Experiments in the context of vehicle classification show improved performance for our ASP-interfaced mote over an all-digital implementation.

Real-time multi-view human action recognition using a wireless camera network

In this paper, we describe how information obtained from multiple views using a network of cameras can be effectively combined to yield a reliable and fast human action recognition system. We describe a score-based fusion technique for combining information from multiple cameras that can handle arbitrary orientation of the subject with respect to the cameras. Our fusion technique does not rely on a symmetric deployment of the cameras and does not require that camera network deployment configuration be preserved between training and testing phases. To classify human actions, we use motion information characterized by the spatio-temporal shape of a human silhouette over time. By relying on feature vectors that are relatively easy to compute, our technique lends itself to an efficient distributed implementation while maintaining a high frame capture rate. We evaluate the performance of our system under different camera densities and view availabilities. Finally, we demonstrate the performance of our system in an online setting where the camera network is used to identify human actions as they are being performed.

Design and analysis of a fast local clustering service for wireless sensor networks

We present a fast local clustering service, FLOC, that partitions a multi-hop wireless network into nonoverlapping and approximately equal-sited clusters. Each cluster has a clusterhead such that all nodes within unit distance of the clusterhead belong to the cluster but no node beyond distance m from the clusterhead belongs to the cluster. By asserting m /spl ges/ 2, FLOC achieves locality: effects of cluster formation and faults/changes at any part of the network are contained within most m units. By taking unit distance to be the reliable communication radius and m to be the maximum communication radius, FLOC exploits the double-band nature of wireless radio-model and achieves clustering in constant time regardless of the network size. Through simulations and experiments with actual deployments, we analyze the tradeoffs between clustering time and the quality of clustering, and suggest suitable parameters for FLOC to achieve a fast completion time without compromising the quality of the resulting clustering.

Hibernets: energy-efficient sensor networks using analog signal processing

In-network processing is recommended for many sensor network applications to reduce communication and improve energy efficiency. However, constraints on memory, speed, and energy currently limit the processing capabilities within a sensor network. In this paper, we describe how ultra-low-power analog circuitry can be integrated with sensor nodes to create energy-efficient sensor networks. We present a custom analog front-end which performs spectral analysis at a fraction of the power used by a digital counterpart. We then show that the front-end can be combined with existing sensor nodes to (1) selectively wake up the mote based upon spectral content of the signal, thus increasing battery life without missing interesting events, and to (2) achieve low-power signal analysis using an analog spectral decomposition block, freeing up digital computation resources for higher-level analysis.

Sensor network based vehicle classification and license plate identification system

Typically, for energy efficiency and scalability purposes, sensor networks have been used in the context of environmental and traffic monitoring applications in which operations at the sensor level are not computationally intensive. But increasingly, sensor network applications require data and compute intensive sensors such video cameras and microphones. In this paper, we describe the design and implementation of two such systems: a vehicle classifier based on acoustic signals and a license plate identification system using a camera. The systems are implemented in an energy-efficient manner to the extent possible using commercially available hardware, the Mica motes and the Stargate platform. Our experience in designing these systems leads us to consider an alternate more flexible, modular, low-power mote architecture that uses a combination of FPGAs, specialized embedded processing units and sensor data acquisition systems.

Multiview fusion for activity recognition using deep neural networks

Abstract. Convolutional neural networks (ConvNets) coupled with long short term memory (LSTM) networks have been recently shown to be effective for video classification as they combine the automatic feature extraction capabilities of a neural network with additional memory in the temporal domain. This paper shows how multiview fusion can be applied to such a ConvNet LSTM architecture. Two different fusion techniques are presented. The system is first evaluated in the context of a driver activity recognition system using data collected in a multicamera driving simulator. These results show significant improvement in accuracy with multiview fusion and also show that deep learning performs better than a traditional approach using spatiotemporal features even without requiring any background subtraction. The system is also validated on another publicly available multiview action recognition dataset that has 12 action classes and 8 camera views.

Network-Aware Double-Layer Distance-Dependent Broadcast Protocol for VANETs

Dissemination of traffic information over multiple hops in vehicular networks requires scalability measures that prevent network congestion and avoid wasting network capacity with unnecessary information forwarding. This issue is even more important if such traffic information coexists with critical safety information. In particular, a viable solution for traffic information dissemination is piggybacking compressed information over periodic basic safety messages. Leveraging this technique, we propose a network-aware double-layer distance-dependent protocol for fast broadcasting of aggregated traffic information over multiple hops. The first distance-dependent layer uses a message size control scheme and a probabilistic approach to ensure that the forwarding within each hop is as fast as possible. The second layer progressively decreases the forwarding rate of subsequent hops as the distance from the source increases. Jointly, the two layers considerably reduce the overall latency of information while also improving the scalability and robustness of the system. We evaluate the performance of the algorithm and compare it with the existing methods, using ns-3 simulations. The results confirm that significant improvement in performance is possible without complicating the forwarding algorithm.

Project exscal

Project ExScal (for Extreme Scale) fielded a 1000+ node wireless sensor network and a 200+ node ad hoc network of 802.11 devices in a 1.3km by 300m remote area in Florida during December 2004. In several respects, these networks are likely the largest deployed networks of either type to date. We overview here the key requirements of the project, describe briefly how they were met and experimentally tested, and provide a pointer to our experimental results.

Collaborative Face Recognition Using a Network of Embedded Cameras

In this chapter, we describe the design and implementation of a distributed real-time face recognition system using a network of embedded cameras. We consider a scenario that simulates typical corridors and passages in airports and other indoor public spaces, where real-time human identification is of prime significance. We characterize system performance on an embedded camera network testbed which is assembled using commercial off-the-shelf components. We quantify the impact of multiple views on the accuracy of face recognition, and describe how distributed pre-processing and local filtering help in reducing both the network load, and the overall processing time.