Background and Motivations

In this chapter, we propose an in-network location management scheme to sup-port imprecision-tolerant queries for object tracking sensor networks. Two types of imprecision are considered. Spatial imprecision means that an object could be located near the location answered by the WSN rather than at the location answered by the WSN. Temporal imprecision means that the location answered by the WSN may be recorded near the current time rather than at the current time. For both spatial imprecision and temporal imprecision, we argue that an imprecision-tolerant location management solution should achieve two desirable goals. First, multiple precision levels should be provided. Second, the query cost should be proportional to the precision level. For example, for spatial impreci-sion, the answer provided by node C should be more accurate than that provided by node A, because node C is farther from the sink (Fig. 5.1(a)). Similarly, for temporal imprecision, the location answered by node C should be newer than that answered by node A (Fig. 5.1(b)).

We observe that the tree-based location management schemes proposed in Chapter 3 could achieve these two goals naturally. For example, Fig. 5.2(a) shows a tree used for location management. In the tree-based location management scheme, when an object moves from one sensor to another, the update message will be forwarded to the lowest common ancestor of those two sensors. Thus,

Sink A

Figure 5.1: Examples of spatial imprecision and temporal imprecision.

when an object originally located outside the spatial range of the subtree rooted at y moves into the range of A at time t₀, node x (the parent of y) will be updated.

Thus, x knows that the object is located at A at time t₀. When a user receives such an answer provided by x, the user can only drive that the object is located at some sensor belonging to a descendant of y. On the contrary in Fig. 5.2(b), if the query is forwarded to y, y can provide that the object is located at B at time t₁, and the user can derive that the object is located at some sensor belonging to a descendant of z. Therefore, we can see that a user can get more precise location information when the query is forwarded more deeply down the tree. Further, if the tree is a deviation-avoidance tree defined in Chapter 3, it ensure thats the hop count between y and the sink will be less than the hop count between z and the sink. It implies that the query cost will be proportional to the precision level.

(Note that when a query is not issued from the sink, it is possible that the querying node is close to the object rather than the sink, and it needs to forward the query to the sink first. This may violate this goal. In this case, the multi-sink system pro-posed in Chapter 4 can be used to solve this problem, because each query is sent to the nearest sink.) In addition, because of its hierarchical structure, a tree-based solution can provide multiple precision levels easily.

Therefore, we propose a tree-based location management model to support tolerant queries. To begin with, we define the format of imprecision-tolerant queries and describe how such queries are processed. The proposed query model can be applied to any tree structure. We then make some observations

x

(a) the query result replied by x Que

ry

(b) the query result replied by y Sink

Figure 5.2: A tree-based location management scheme.

regarding the relationship between query cost and tree structure, and propose a tree construction algorithm to facilitate the proposed imprecision-tolerant location management model by reducing the query cost while minimizing the increment of the update cost. Finally, performance studies are conducted via simulations.

5.1.2 Network Model

The network model used in this chapter is the same with that proposed in Chap-ter 3. We consider a WSN to be used for object tracking. We adopt a simple nearest-sensor tracking model, in which the sensor that receives the strongest sig-nal from an object is responsible for tracking the object (this can be achieved by [7] and we omit the details). Therefore, the sensing field can be modelled by a Voronoi graph [4], where each sensor’s responsible area is the polygon containing itself. Two sensors are called neighbors if their sensing ranges share a common boundary on the Voronoi graph. Multiple objects may be tracked concurrently by the network, and we assume that from mobility statistics, it is possible to collect the frequency that objects move between each pair of neighboring sensors, called the event rate.

5.2 Imprecision-tolerant Location Management Model

We will propose a tree-based, imprecision-tolerant location management model in this section. Below, we will first introduce the update and query mechanisms.

Since the update and query mechanisms can be applied to any tree structure, we will then discuss how to reduce the communication cost (i.e., update cost plus query cost) by adjusting the tree structure. We observe that uncorrelated sensors should not be put together under a subtree to reduce the query cost while corre-lated sensors should be put together to minimize the increment of update cost, where the formal definition of correlation of sensors will be introduced later. We will discuss how to collect query statistics to identify the correlation of sensors.

Finally, based on query statistics, we propose a tree construction algorithm IQT (Imprecision-tolerant Query Tree) to reduce the query cost while minimize the increment of the update cost.