Revolutionary Automatic Target Recognition and Sensor Research (RASER)

Sponsor: Air Force Research Laboratory Sensors Directorate (AFRL/SNAT)

Research Team: John Kerekes, Michael Muldowney (M.S. student), Kristin Strackerjan (M.S. student), Lon Smith and Brian Leahy (B.S. student)

Project Scope: The goal of this effort is to investigate the feasibility of using high resolution hyperspectral imagery to aid detection and tracking of vehicles of interest in an urban environment. The project is investigating whether adequate spectral reflectance features of a vehicle can be measured by an airborne hyperspectral imager to uniquely associate measurements made at one time and location with those a short time later and distance away.

Project Status: In the summer of 2005, a vehicle tracking experiment was conducted at RIT's campus using volunteer drivers, field spectrometers and RIT's airborne MISI instrument. Locations of the experimental vehicles were noted before one data collection pass by MISI, the vehicles were then moved to new locations, which were noted, and then MISI performed a second collection pass. Figure 3.9-1 shows RGB composites of the MISI imagery from two sample passes.

Figure 3.9-1: MISI imagery of the RIT campus at 12:09 pm (left) and 12:24 pm (right)

Analysis of the data shown in Figure 3.9-1 confirmed that a vehicle spectrum taken in the initial image can be used in a target detection algorithm to successfully detect the same vehicle in the subsequent image when false alarm reduction (corresponding to fixed manmade structures) is applied. However this was only successful for vehicles with significant contrast (bright blue) compared to the background.

To explore the sensitivities of this detection/tracking problem, the Forecasting and Analysis of Spectroradiometric System Performance (FASSP) model was used to analytically predict detection/false alarm rates for various scenarios. A hypothetical scene was modeled and detection performance predicted as shown in Figure 3.9-2. The plot on the left shows an example sensitivity to paint type. The target spectra used here were measurements of two types of blue paint. As can be seen, the type 2 paint requires the hyperspectral image resolution be such that the vehicle completely fills a pixel in order to be even modestly detectable, while the type 1 paint allows the vehicle to be detected when the resolution is such that the vehicle may only occupy one-half of a pixel.

Additional vehicle tracking experiments were conducted both at RIT and in an alpine region of Montana . Future work will explore achievable performance in those data sets.

Figure 3.9-2: Model-predicted detection sensitivity to vehicle paint type (left) and spectral coverage/SNR (right) corresponding to typical performance of the MISI sensor and full or limited use of the spectral channels available in the HYDICE sensor