This article considers a problem of periodically estimating energy consumption breakdowns for main appliances inside building using a single power meter and the knowledge of the ON/OFF states of individual appliances. In the first part of this article, we formulate the problem as a constrained convex optimization problem with tunable parameters. Then we propose an online algorithm that adaptively determines the optimization parameters to robustly estimate the breakdown information. The proposed solution is evaluated by experiment using a scaled-down proof-of-concept prototype with real measurements. In the second part, we provide detailed analysis to understand the performance of our proposed algorithm. We first develop a stochastic model to describe evolution of appliances’ ON/OFF states using continuous-time Markov chain. Then we derive analytical bounds of estimation error and the probability of a rank-deficient binary matrix. Those analytical bounds are verified by extensive simulations. Finally, we study the effect of collinearity of binary data matrix on estimation performance. Simulation results suggest that our algorithm is robust against the collinearity of binary dataset.

The thesis of this dissertation is that by analyzing the temporal properties of sensor measurements, patterns generated by macroscopic behaviors can be discovered and used to form virtual sensors that can convert low-level sensor events into actionable knowledge. Macroscopic behaviors are defined as human activities and routines that evolve over large spaces and extended periods of time and can thus be learned only in an unsupervised manner. Our work is based on two key observations: (1) most human behaviors are sequences of very primitive actions or events that can either be sensed directly by sensors or indirectly by pre-processing the sensor data; and (2) the same type of human activity will often trigger a similar pattern of sensor events in space and time. Based on these observations, this dissertation introduces three common types of macroscopic behaviors and proposes methods for their recognition. The first concerns the problem of identifying periodic human activities. The second focuses on discovering classes of frequent Spatio-Temporal Activities (STAs) from location traces, namely areas that people consistently spend time at approximately the same time intervals every day. The third problem deals with the detection of simple group behaviors specified in the form of sequences of interactions. The ability to define virtual sensors for these macroscopic behaviors is in the core of the BehaviorScope system — a human-centric sensing system aiming to offer real-world services as assisted living and power efficiency in large buildings. In the latter case we show how our system can potentially reduce the electricity consumption of an office building by up to 33.8%.

Given the ongoing widespread deployment of low frequency electricity sub-metering devices at residential and commercial buildings, fine-grained usage information of end-loads can bring a new powerful sensing modality in Cyber-Physical Systems (CPS). Motivated by the opportunity, this paper describes an algorithm of estimating the ON/OFF sequences for typical household end-loads in close-to-real-time using an off-the-shelf power meter. Unlike previous algorithms that lacks in scalability to support diverse applications in CPS our algorithm is designed to provide control knobs to support various trade-offs between accuracy and computation load or delay to satisfy the different application requirements. We experimentally verify the proposed algorithm using a collection of home appliances. Our experiment result shows that our algorithm is able to detect ON/OFF sequences of 7 appliances nearly without error and 3 appliances with moderate error rate less than 6% among 12 typical household appliances.

To help combat terrorist and insurgent threats, the DoD is deploying persistent surveillance systems to record the activities of people and vehicles in high-risk locations. Simple observation is insufficient for real-time monitoring of the vast amounts of data collected. Automated systems are needed to rapidly screen the collected data for timely interdiction of terrorist or insurgent activity. Effective analysis is hampered by the similarity of actions of individuals posing a threat to actions of individuals pursuing a benign activity. The analysis of the activity of groups of individuals, with requirements for team coordination, can potentially increase the ability to detect larger threats against the background of normal everyday activities. APL, in collaboration with Yale University, is developing sensor-independent approaches and tools to robustly and efficiently analyze complex group activities.

This poster paper describes a method for estimating the ON/OFF usage profile for appliances inside a home using an off-the-Unlike previous approaches that put their emphasis on the detection of individual events using high frequency sampling, our approach aims to reliably detect sequences of ON/OFF events from traces of less reliable ON/OFF events identified on data from low-frequency sampling. The proposed algorithm is evaluated experimentally using a collection of home appliances.

A large collection of mobile sensing applications depend on the knowledge of the user's whereabouts and are heavily based on GPS location measurements. Although knowledge of location is very desirable, in many mobile applications excessive GPS sampling is very energy taxing thus posing a barrier to application sustainability. To mitigate this problem, in this paper we examine how to reduce GPS sensing redundancies by extracting the state of a person and using it to drive GPS sampling on mobile phones. Using a GPS dataset we first describe how to extract the spatio-temporal states of the user. We then use the knowledge of the user's state to reduce GPS sampling rate, helping to make mobile applications more sustainable.

The proliferation of the smart grid creates new opportunities for large buildings to act as smart end-points that provide mutually beneficial services for building occupants and the grid. In this article we describe how Cyber-Physical systems that provide rapid access to information and decision-making can enable buildings to autonomously interact with the grid. By participating in new real-time electricity markets on the utility side of the meter and day-ahead markets, buildings have the potential of achieving local efficiency and reliability gains while also facilitating the effective utilization of renewables.

For tomorrow's smart and assistive environments to fully realize, it will be of fundamental importance to be able to obtain the location of people in an environment, as well as their evolution in time—sometimes even across large sensing gaps. And so, the ultimate goal is to obtain a cheap, scalable solution for person-detection and tracking for use in long-term real-world scenarios. In this work I present a system that takes a step in that direction. After a comprehensive review of existing human sensing approaches, I reason that the best multi-sensor configuration to solve this problem robustly and cost-effectively consists of cameras and inertial sensors. The proposed system localizes people using the existing infrastructure of CCTV cameras. People can, then, opt-in on being tracked and identified by carrying a mobile phone equipped with a custom software client. Using the inertial sensors on the phone, the client calculates and transmits a motion signature which is then compared with the motion hypotheses observed with the camera network. When a match is found, people are identified. The final solution is lightweight enough to potentially execute in real time on existing sensor nodes, thus providing a compact, cheap, and effective human-sensing solution. The system is evaluated through extensive simulations as well as a number of experiments.

This paper presents an automated methodology for extracting the spatiotemporal activity model of a person using a wireless sensor network deployed inside a home. The sensor network is modeled as a source of spatiotemporal symbols whose output is triggered by the monitored person’s motion over space and time. Using this stream of symbols, the problem of human activity modeling is formulated as a spatiotemporal pattern-matching problem on top of the sequence of symbolic information the sensor network produces, and is solved using an exhaustive search algorithm. The effectiveness of the proposed methodology is demonstrated on a real 30-day dataset extracted from an ongoing deployment of a sensor network inside a home monitoring an elder. The developed algorithm examines the person’s data over these 30 days and automatically extracts the person’s daily pattern.

The in-house monitoring of elders using intelligent sensors is a very desirable service that has the potential of increasing autonomy and independence while minimizing the risks of living alone. Because of this promise, the efforts of building such systems have been spanning for decades, but there is still a lot of room for improvement. Driven by the recent technology advances in many of the required components, in this article, we present a scalable framework for detailed behavior interpretation. Our framework supports in-house monitoring of elders using an intelligent gateway and a set of cheap commercially available sensors, in addition to more advanced camera-based human localization sensors and a client for GPS-enabled mobile phones that provides monitoring when outdoors. In this article, we report our experiences and present our current progress in three main components: sensors, middleware, and behavior interpretation mechanisms spanning from simple programmable rule-based alerts to algorithms for extracting the temporal routines of individuals.

This paper describes an algorithm for determining if an event occurs persistently within an interval where the interval is periodic but the event is not. The goal of the algorithm is to identify events with this property and also determine the minimum interval in which they occur. This solution is geared towards discovering human routines by considering the triggering of simple sensors over a diverse set of spatial and temporal scales. After describing the problem and the proposed solution, in this paper we demonstrate using testbed data and simulations that this approach uncovers components of routines by identifying which events are parts of the same routine through their temporal properties.

This poster abstract introduces the problem of macroscopic sensing composition, where a sensor capable to detect complex events is synthesized dynamically by a collection of simpler sensors using a data-driven approach. Our solution is geared towards discovering the structure of human activities by considering the triggering of simple sensors over a diverse set of spatial and temporal scales. The goal is to identify routines from their components by leveraging the fact that the components have the same temporal persistence as the routines themselves. To this end we have devised an algorithm for determining if an event occurs consistently within a time interval where the interval is periodic but the event is not. The goal of the algorithm is to identify events with this property and also determine the minimum interval in which they occur. Our first results using testbed data and simulations indicate that this approach can uncover components of routines by identifying which events are parts of the same routine through their temporal properties.

We present a method to identify and localize people by leveraging existing CCTV camera infrastructure along with inertial sensors (accelerometer and magnetometer) within each person's mobile phones. Since a person's motion path, as observed by the camera, must match the local motion measurements from their phone, we are able to uniquely identify people with the phones' IDs by detecting the statistical dependence between the phone and camera measurements. For this, we express the problem as consisting of a two-measurement HMM for each person, with one camera measurement and one phone measurement. Then we use a maximum a posteriori formulation to find the most likely ID assignments. Through sensor fusion, our method largely bypasses the motion correspondence problem from computer vision and is able to track people across large spatial or temporal gaps in sensing. We evaluate the system through simulations and experiments in a real camera network testbed.

In this paper we present a classification of human movement in physical space into spatio-temporal activities (STAs) and classes thereof. Drawing from our experiences with real human data from GPS traces we define a clustering approach for STA extraction based on the amount of motion of the user in space and time. Our solution captures these properties in a lightweight online algorithm that can run inside mobile devices. We then cluster the discovered STAs into classes based on a similarity metric that aims to identify which activities (STAs) are consistent in time. In contrast to previous approaches of discovering important places, this work also utilizes the temporal properties of the data to extract more realistic STAs and STA classes. Our work is evaluated through simulations and real GPS traces.

This paper considers the problem of estimating the power breakdowns for the main appliances inside a building using a small number of power meters and the knowledge of the ON/OFF states of individual appliances. First we solve the breakdown estimation problem within a tree configuration using a single power meter and the knowledge of ON/OFF states and use the solution to derive an estimation quality metric. Using this metric, we then propose an algorithm for optimally placing additional power meters to increase the estimation certainty for individual appliances to the required level. The proposed solution is evaluated using real measurements, numerical simulations and by constructing a scaled down proof-of-concept prototype using binary sensors.