Object Tracking Using a Wireless Camera Network
Conventional cameras networks require an infrastructure of cables and computers to transmit, store and process the information coming from the cameras. In many applications, it is undesirable, or even impossible to have such infrastructure. As a consequence, there has been increasingly more focus on the substitution of conventional cameras by wireless smart cameras, i.e. wireless cameras endowed with processing power, capable of making local decisions before transmitting any data over the network. However, to make wireless smart cameras viable, it is necessary to extend the lifetime of the batteries as long as possible. Therefore, the hardware architecture of such devices is highly constrained and performing even very simple processing tasks efficiently is a challenge. In our wireless cameras project, we use a network of Cyclops cameras attached to MicaZ motes to track an object based on its color information. Figure 1 illustrates the Cyclops camera and the MicaZ mote
Figure 1: Cyclops camera attached to a MicaZ mote.
Currently, we have a network of 18 cameras deployed in our laboratory, running two different systems, as illustrated in Figure 2. The dark region in the right side of the image consists of a uniform tracking system, in which each camera sends the coordinates of the target to the base-station. The region in the left side of the image corresponds to a hierarchical tracking system, in which the cameras dynamically create clusters and only the cluster head communicates with the base-station. Figures 3 (a) and (b) show the actual camera networks, the 12 cameras composing the hierarchical network are shown in Figure 3(a) and the 6 cameras composing the uniform network are shown in Figure 3(b).
Figure 2: Camera networks for distributed object tracking.
(a) (b)
Figure 3: Actual networks of wireless cameras.
In the uniform tracking system, each camera continuously acquires images of the target, computes the coordinates of its center of mass, and sends them to a base-station computer. The position of the object is then displayed in the graphical user interface illustrated in Figure 4. The gray circles represent the center of each camera's field of view. The yellow circle corresponds to the target whose position with respect to the world reference frame is displayed in red in the bottom left corner of the image. The origin of the world coordinate frame is in the top left image corner. The grid represents 12x12 inches tiles in the floor of the room. The remaining gray, brown and black objects illustrate pieces of furniture present in the room.
Figure 4: Graphical user interface displaying the uniform tracking system.
To explore the advantages of creating clusters of cameras to process information locally before sending it to the base-station, we implemented a hierarchical tracking system in which the cameras dynamically create clusters as the object moves. When a camera detects the object, it creates a cluster and requests its four closest neighbors to become members of this cluster. The cluster members then send the coordinates of the target to the cluster head and only the cluster head communicates with the base-station. The position of the target, as well as the active clusters, are displayed in a graphical user interface as illustrated in Figure 5.
Figure 5: Graphical user interface displaying the hierarchical tracking system.
However, as illustrated in Figure 5, due to the unreliable communication between the motes, not all the cameras that are supposed to join a cluster actually do so. We are investigating the trade-off between the quality of the clusters formed (i.e., the number of cameras that should join the clusters and the number of cameras that actually do so) and the delay in the cluster formation process. We are also investigating more effective mechanisms to create clusters using the least amount possible of a priori information.
To be able to quantify the contributions from each camera, we included an uncertainty model in the system. Assuming the errors in the x and y directions of the coordinate frames of the cameras to be uncorrelated, we projected the error into the world reference. Figure 6 shows the uncertainty ellipses corresponding to the 95% confidence region of two observations from different cameras.
Figure 6: Graphical user interface displaying the uncertainty ellipses.
• Henry Medeiros
• Johnny Park
12 wireless cameras tracking an object.
