OVERVIEW

The main objective of multiscale sensing is the accurate construction of the distributed environmental phenomena, such as the temperature over an area, solar illumination near the ground, or the water vapor density in the air. Due to the fast change of the environment both in space and time, to achieve the objective by one scale of sensing if possible, such as deploying a large amount of sensors in the field, can be prohibitively expensive. In such situations, multiscale sensing will help in reducing the required resources while meets or even exceeds the performance of one scale sensing. For instance, in two-scale sensing, one scale provides global information of the phenomena and one scale provides local details. This global information can indicate where more local details are needed. By allocating more resources to the most needed place and fewer resources elsewhere, we reduce the reconstruction error comparing to allocating resources uniformly over the environment.

In many cases, different type of sensors can provide different sensing resolutions. By combining the sensing measurements from different type of sensors, we can reach the effect of multiscale sensing. Therefore, multiscale sensing is often inseparable from multimode sensing. We will refer both as multiscale sensing.

The primary objective is to show that multiscale sensor information can be fused to provide more information about the environment than the sum of each individual sensor component. A testbed system will be created for measuring, testing, and verifying these hypotheses. Once a methodology and process is created to fuse multiscale sensor data, the algorithm will be autonomously benchmarked against collected groundtruth to measure the performance, namely uncertainty, of the system and algorithm.

SYSTEMS / EXPERIMENTS

The multiscale experiment is being performed on the NIMS-LS testbed system. Lights adopted from theater stage lighting is used to project light through an obstacle plane to create a variable light field in the NIMS-LS transect plane. The in-situ PAR sensors are located on the mobile NIMS-LS node and also embedded as static sensors on a back plane. A remote camera is used to image the scene. Figure 39 shows an image from the camera showing a projected light pattern. Figure 40 is a processed image.

Image of a projected light pattern onto the NIMS-LS transect plane.

Partitioned image of the above light pattern

To validate the approximation from geometry in the field reconstruction, the penumbra is measured in very fine detain through the mobile PAR sensor. Approximation is then made using the parameters estimated from the measurements. Both the measurement curve and approximation curve is shown in figure 3.

Measured light intensity in the penumbra and the approximation

PEOPLE

Faculty:

Prof. William Kaiser
Prof. Greg Pottie

Staff:

Heather Levin

Graduate Students:

Xiangming (Cathy) Kong
Richard Pon

Undergraduate Students:

Michael Stealy
Yeung Lam
Willie Chen
Eric Yuen