Many motion detection and tracking algorithms rely on the process of background subtraction, a technique which detects changes from a model of the background scene. We present a new algorithm for the purpose of background model initialization. The algorithm takes as input a video sequence in which moving objects are present, and outputs a statistical background model describing the static parts of the scene. Multiple hypotheses of the background value at each pixel are generated by locating periods of stable intensity in the sequence. The likelihood of each hypothesis is then evaluated using optical flow information from the neighborhood around the pixel, and the most likely hypothesis is chosen to represent the background. Our results are compared with those of several standard background modeling techniques using surveillance video of humans in indoor environments.

We present an image quality metric and prediction model for SAR imagery that addresses automated information extraction and exploitation by imagery analysts. This effort drarws on our team's direct experience with the development of the Radar National Imagery Interpretability Ratings Scale (Radar NIIRS), the General Image Quality Equations (GIQE) for other modalities, and extensive expertise in ATR characterization and performance modeling. In this study, we produced two separate GIQEs: one to predict Radar NIIRS and one to predict Automated Target Detection (ATD) performance. The Radar NIIRS GIQE is most significantly influenced by resolution, depression angle, and depression angle squared. The inclusion of several image metrics was shown to improve performance. Our development of an ATD GIQE showed that resolution and clutter characteristics (e.g., clear, forested, urban) are the dominant explanatory variables. As was the case with NIIRS GIQE, inclusion of image metrics again increased performance, but the improvement was significantly more pronounced. Analysis also showed that a strong relationship exists between ATD and Radar NIIRS, as indicated by a correlation coefficient of 0.69; however, this correlation is not strong enough that we would recommend a single GIQE be used for both ATD and NIIRS prediction.

Commercial security and surveillance systems offer advanced sensors, optics, and display capabilities but lack intelligent processing. This necessitates human operators who must closely monitor video for situational awareness and threat assessment. For instance, urban environments are typically in a state of constant activity, which generates numerous visual cues, each of which must be examined so that potential security breaches do not go unnoticed. We are building a prototype system called BALDUR (Behavior Adaptive Learning during Urban Reconnaissance) that learns probabilistic models of activity for a given site using online and unsupervised training techniques. Once a camera system is set up, no operator intervention is required for the system to begin learning patterns of activity. Anomalies corresponding to unusual or suspicious behavior are automatically detected in real time. All moving object tracks (pedestrians, vehicles, etc.) are efficiently stored in a relational database for use in training. The database is also well suited for answering human-initiated queries. An example of such a query is, "Display all pedestrians who approached the door of the building between the hours of 9:00pm and 11:00pm." This forensic analysis tool complements the system's real-time situational awareness capabilities. Several large datasets have been collected for the evaluation of the system, including one database containing an entire month of activity from a commercial parking lot.

In this paper, we focus on the problem of automated surveillance in a parking lot scenario. We call our research system VANESSA, for Video Analysis for Nighttime Surveillance and Situational Awareness. VANESSA is capable of: 1) detecting moving objects via background modeling and false motion suppression, 2) tracking and classifying pedestrians and vehicles, and 3) detecting events such as person entering or exiting a vehicle. Moving object detection utilizes a multi-stage cascading approach to identify pixels that belong to the true objects and reject any spurious motion, (e.g., due to vehicle headlights or moving foliage). Pedestrians and vehicles are tracked using a multiple hypothesis tracker coupled with a particle filter for state estimation and prediction. The space-time trajectory of each tracked object is stored in an SQL database along with sample imagery to support video forensics applications. The detection of pedestrians entering/exiting vehicles is accomplished by first estimating the three-dimensional pose and the corresponding entry and exit points of each tracked vehicle in the scene. A pedestrian activity model is then used to probabilistically assign pedestrian tracks that appear or disappear in the vicinity of these entry/exit points. We evaluate the performance of tracking and pedestrian-vehicle association on an extensive data set collected in a challenging real-world scenario.

No abstract