Context-Aware Search using Cooperative Agents in a Smart Environment Dan Xie Roderic A. Grupen Allen Hanson Department of Computer Science, University of Massachusetts Amherst Amherst, MA 01003, USA {dxie,grupen,hanson}@cs.umass.edu Abstract In this paper we present the design of a decentralized vision-based object search system that can be used for el- der care in a smart environment. In our approach, each autonomous search agent maintains separate estimates of the probability density function (PDF) of the object location and makes independent decisions about its search process. Asynchronous cooperative search is achieved by transmit- ting perceptual information among the agents. Our work also investigates how context such as the detection history and density of activity by people influence the estimation of the prior PDF of the target and the use of this information to improve the search efficiency. Our experimental results demonstrate that the proposed cooperative search strategy is efficient and the methods we use to incorporate contextual information into the target’s posterior PDF can improve the efficiency further. 1. Introduction Health care for the elderly poses a major challenge as the baby boomer generation ages. Part of the solution is to develop technology using sensor networks and service robotics to increase the length of time that an elder can re- main at home. In addition to monitoring for illnesses and potentially life-threatening situations, an equally important challenge in in-home elderly care is providing assistance in their day-to-day life. Since moderate immobility and memory impairment are common as people age, a major problem for the elderly is locating and retrieving frequently used “common” objects such as keys, cellphone, books, etc. Therefore, it is important to develop effective and efficient approaches for automated in-home object search. Heuristic strategies in target search. There has been considerable recent interest in addressing the problem of “target search”. Bourgault et al. [3] proposed a Bayesian approach to the problem of target search by a single au- tonomous sensor platform. Ye et al. [10] formulated target search as an optimization problem where the goal is to max- imize the probability of detecting the target within a given time constraint. Since planning such search activity is NP- complete, some heuristic strategies were proposed that of- ten lead to practical solutions. Wixon et al. [9] use the idea of indirect search, in which one first finds an object that typically has a spatial relationship to the target, and then restricts the search in the spatial area defined by that rela- tionship. Sujan [8] proposes an iterative planning approach driven by an evaluation function based on Shannon’s infor- mation theory. The camera parameter space is explored and each configuration is evaluated according to the evaluation function. The concept of a visibility map is introduced in [6, 7] to constrain the sensor parameter space according to the detection characteristics of the recognition algorithm. These techniques reduce the dimension of the sensor pa- rameter space. Cooperative search strategies. In search operations, a team of intelligent agents can provide a robust solution with greater efficiency than can be achieved by single agents, even with comparatively superior mobility and sensors. The key is to develop a cooperative decentralized control strat- egy that allows each agent to determine its actions indepen- dently while optimizing the team’s performance. A syn- chronized coordinated search strategy was developed in a Bayesian framework in [2]. DeLima et al. [4] proposed a rule-based search method with which multiple unmanned aerial vehicles can cooperatively search an area for mobile target detection. The approach proposed in this paper is related to both the aspects of cooperative strategy and heuristic solution. We use multiple Pan/Tilt/Zoom (PTZ) camera nodes as the search agent to explore a specified area in a living space and focus on developing an efficient cooperative search strat- egy. In the design of our smart environment, the camera nodes are supposed to perform multiple tasks such as ob- ject search, people tracking, and are controlled by a re- source management unit. In this framework, the search process of an agent can be interrupted by other tasks with higher priority. This problem, along with the possible node and network failures, recommends a decentralized search