Combining intelligent video surveillance with advanced radiation detection
Hi everyone, I’m working in the Visualization and Computer Vision Lab on Intelligent Video applications. We’re developing systems that have the ability to automatically detect and track people in video. Given one or more cameras we can locate and track a person in 3D, and accurately determine the speed and the direction of travel. Unfortunately the resolution of typical security cameras is not very high and security operators can often not see a lot of details when reviewing video, which is especially important for forensic or biometrics applications. This is why many operators like to use pan tilt zoom (PTZ) cameras. These cameras can be rotated and zoomed remotely to focus on regions of interest. This is of course labor intensive and a single operator can only control one camera at a time. To help with this, we recently developed a new algorithm (see [1]) that can automatically control a network of PTZ cameras in a collaborative fashion. For every pairing of people to cameras, the system determines the expected quality of the capture which is based on factors such as localization accuracy, view angle at the face, and distance from the camera to the subject. From all possible pairings the algorithm then searches for the best capture strategy — essentially a list of tasks for every PTZ which, once completed, maximize the overall quality of all captures from all cameras. One can see in the video above that this automatic control of the four PTZ cameras (bottom) works very well.
This recent work shows our ability to use information from the video tracking system to guide other sensors. This is in spirit the same thing we will do for a program that we were just awarded by the Domestic Nuclear Detection Office (DNDO), which is part of the DHS. The goal of that program is to detect radiation sources from a distance using a small Compton imager developed here at GE Global Research. This imager has the ability to detect the presence and direction of radioactive material; however it needs to combine multiple measurements in order to perform the source location estimation. This is no problem if the source is stationary, but breaks down for moving sources. However, our video tracking system can detect and measure the location of targets of interest and we will use these location estimates to perform localization through motion compensation. This new system will be called the Target Linked Radiation Imager (TLRI). This is a great program and our technology will in the future be able to detect the presence and location of radioactive material in, for example, passing vehicles and hence solve an important homeland security challenge.
[1] Nils Krahnstoever, Ting Yu, Ser-Nam Lim, Kedar Patwardhan, Peter Tu, “Collaborative Real-Time Control of Active Cameras in Large-Scale Surveillance Systems”, Workshop on Multi-camera and Multi-modal Sensor Fusion Algorithms and Applications (M2SFA2), Marseille, France, October 18, 2008.
This project was supported by grant #2007-RG-CX-K015 awarded by the National Institute of Justice, Office of Justice Programs, US Department of Justice. The opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of the Department of Justice.
Congratulations for this breakthrough project!