Automated Behavior Analysis

To understand how automated behavior analysis would work, it is appropriate to first take a step back and review the basics of video surveillance. Although the application of video surveillance varies from system to system, Taha et al. (2014) cite six consecutive steps required for a general video surveillance system (p. 22). Figure 1 depicts the six steps, their relation to one another and what each step seeks to accomplish. When combined, these steps build upon the previous to construct a machine readable scene capable of performing basic surveillance tasks. These low-level processing techniques, generally concerned with “forming the basic building blocks of human behavior recognition” (Gowsikhaa, Abirami, & Baskaran, 2012, p. …show more content…
NA). Today, there are several commercially available products, known as Intelligent Video Surveillance (IVS) systems, on the market offering varying degrees of automated analysis. Some of these products include ActivEye, Perceptrak, VideoIQ, FenceWATCH and The Algorithm Based Object Recognition and Tracking (ABORAT) system (fig. 2). Unlike other commercially available products, which are generally software based, ABORAT “is a vision-based intelligent surveillance system” (Appiah & Hunter, 2009, p. 1) capable of “learning to characterize the actions taking place [in a video stream and] infers whether events present a threat that should be signaled to a human operator” (Appiah & Hunter, 2009, p. …show more content…
Trajectory classification serves as the basis for abnormal activity detection. By analyzing the position, speed and appearance of the tracked object’s trajectory and then comparing those trajectories to the “normal” standard used to train the system, ABORAT is “able to detect normal/abnormal activities” (Appiah & Hunter, 2009, p. 6). These activities are then used to determine if abnormal behavior is present (i.e. behavior detection) and, if required, issues one of three alerts, “object type, zone and tripwire violations and the behavioral alert” (Appiah & Hunter, 2009, p. 7). In short, “The system is fully autonomous…capable of using the trajectory of moving object[s] to classify the behavior” (Appiah & Hunter, 2009, p. 8) and alerts monitors to the presence of abnormal behavior. Initial tests yielded promising results with a 4.4% error rate for data collected on a normal day and 14.5% error rate for data collected on a rainy day (Appiah & Hunter, 2009, p.