Organizations have made significant investments in their video surveillance infrastructure consisting of hundreds of cameras, recorders, storage devices, and video monitors. Yet, with all this state of the art infrastructure, analysis of real-time or recorded video is bounded to the limitations of humans who are often required to monitor more than one monitors to detect security threats (see Figure).

False-Negatives

No matter how highly trained or how dedicated a human observer, it is impossible to provide full attention to more than one or two things at a time; and even then, only for a few minutes at a time. A Harvard University study concluded that humans are surprisingly unaware of the details of their environment, and often do not detect large changes to objects or scenes (‘change blindness’). Furthermore, without attention, humans may not even perceive objects (‘inattentional blindness’). The Harvard experiment results showed that 50% of people counting the passes made between two basketball teams will not notice a gorilla walk into the middle of the viewing area, beat its chest, and walk out. In another study, military experiments demonstrated that after 12 minutes of continuous viewing of 2 or more sequencing monitors, an operator will miss up to 45% of all scene activity. After approximately 22 minutes, an operator will miss up to 95% of scene activity. The conclusion is clear - humans do not reliably detect security threats, whether watching live video or reviewing archived data, resulting in false conclusions that nothing occurred when, in fact, something did (referred to as ‘false negatives’).

False-Positives

Prior to the technological advancements that have made computer vision based solutions commercially viable, many manufacturers of cameras and digital video recorders introduced Video Motion Detection (VMD). VMD technology essentially looks for pixels that are different than the current background model in the same region of the scene. Unfortuntely, these systems end up causing high number of false alarms in environments where there is a lot of irrelevant motion – such as weather, clouds, shadows, changes in lighting, etc. In fact, this caused so many false alarms (referred to as ‘false positives’) that distracted the monitoring process, the end user simply turned off the VMD feature.

Retrieval of Relevant Video Content

Due to the high number of ‘false positives’ and ‘false-negatives’, organizations record video output in case something occurred that was either not detected in real-time or needs to be reviewed to confirm the actual occurrence. In the absence of “intelligent” analysis and classification, accessing the desired video is an exercise in trial and error that returns incomplete information and is time consuming. For instance, in a case where someone wanted to determine if a person passed through a restricted area over a weekend, over 48 hours of video would have to be culled through to find the relevant information needed. And even then, the view of the historical data may be limited to a remote view of the person making it virtually impossible to detect any specific feature on the person (such as gender, hair color, clothing, etc.).

Limited Scalability

As organizations expand the area in which they wish to monitor, the actual returns on their investment often do not meet expectations. The more video output created and recorded does not directly result in an increased level of security. Rather, it simply creates more unintelligent data. In order to actually obtain relevant information that could result in actionable response, an organization needs to add people to analyze the data. In essence, the more video you create - the more man-hours that are needed to monitor the video. The end result is growing security overhead costs.

Another limitation on the scalability of human base video monitoring involves areas that are either extremely hazardous or too costly to put humans. For instance, it is extremely expensive to pay for man based monitoring of the entire border between Mexico and the United States.

Delays in Time Critical Response

The level or responsiveness to security breaches directly impacts security performance. Under conventional video surveillance processes, when a security breach has occurred often a human must respond to the alarm. Though sometimes the response may be immediate, it is often the case that a delay may occur in the security guards response due to several factors, including unfamiliarity with alarm response procedures, not at his station at time of the alarm notice, distracted by other monitoring scenes, not reporting security breaches correctly, or has a conflict of interest (e.g. he knows the individual breaching security). The response time to a security threat is completely dependant on the ability and integrity of a monitoring agent.