As a result of the ever-increasing threat of terrorism both international and domestic, as well as a rise in street crimes in many municipalities, security directors nationwide are increasingly challenged to protect their facilities, employees and visitors. The security departments in many facilities have large numbers of CCTV cameras in their video surveillance systems, making the job of monitoring them a daunting task. In the past, cameras were displayed on the monitors only when motion was detected. This technique was used to minimize the amount of visual information that the security officer at the command and control center had to view at one time. However, there are many 24-hour facilities such as hospitals that are active during all hours of the day and night. This makes motion detection technology a less-than-optimal solution for these types of accounts.

However, recent advances in digital technology have made the job of monitoring these vast numbers of cameras a somewhat more manageable task. An innovative and developing technology with much promise for the future is behavior recognition software programs. These types of software programs utilize complex mathematical algorithms to track pedestrians and vehicles as they pass in the field of view of a camera, and then classify the motion and activities of each. This type of a program is designed to detect an array of aberrant behaviors such as someone lying on the floor, erratic pedestrian motions, a person or vehicle staying in one place for an extended period, a person or vehicle traveling against the flow of traffic, someone running, someone dropping a bag or other item, objects newly appearing on the scene, etc. A person who has fallen, for example, can be detected by the system program using a combination of velocity and geometric shape analysis. Using these programs, the security officer’s efficiency can be maximized by having a few key surveillance cameras displayed continuously, while the other cameras in the system can be called up only when there is some type of anomalous activity being detected.

Some behavior-recognition programs utilize recursive adaptive computer algorithms. These are dynamic programs where normal patterns of behavior are continually learned and updated to define acceptable behavior patterns. These types of programs are the most reliable, as they will alert all anomalies and not just those that have been previously identified. Behavior recognition and personnel tracking can also be accomplished by using other computational algorithms including Kalman filtering or proprietary programs. The details of the algorithms are unique, each having its own tradeoffs. The interface of the behavior recognition-software to the surveillance system will require some additional hardware beyond the base surveillance system components. This additional hardware may include:

• A video activity processor CPU with a frame grabber for a given group of cameras

• Node manager software

• Arbitrator software

• A port on the local area network (LAN) for each of the video activity processors and node manager processors

• A graphical user interface (GUI) to allow setting the rules for camera call-up

This intelligent video technology is currently being used at major airports such as Palm Beach International Airport, Dallas/Fort Worth Airport, Salt Lake City Airport and Lambert-STL Airport with good success. At Palm Beach International Airport, for example, the system detects “wrong-way” motion at the exit lane on the concourse level. This prevents anyone from attempting entry to the secure concourse level via the exit lane. At Dallas/Fort Worth, the system has been installed at three American Airlines checkpoints to also monitor “wrong-way” motion. This technology continues to be developed and improved, and will become an effective security management tool for those managers with corporate and industrial physical security responsibilities.

Facial-recognition software is another developing technology for the security industry. Facial recognition is a form of biometrics that analyzes facial features and landmarks called nodal points. There are approximately 80 nodal points on a human face. Some of the nodal points that are measured by the software include:

• Distance between the eyes

• Width of nose

• Depth of eye sockets

• Cheekbones

• Jaw line

• Chin

Figure 1 shows the distances between some typical facial nodal points. In theory, the software program can be set up so that only a given percentage of nodal points need to be matched by the computer to yield a positive identification. The software utilizes an algorithm called local feature analysis (LFA). Each faceprint is stored as an 84-byte file. Using this methodology, many faces can be stored in a given database using a minimal amount of digital memory.

The computer scans the face and then assigns a value using a scale of 1 to 10. If a score is above a predetermined threshold, the computer declares a match. The operator then reviews the face or group of faces that has been selected from the computer database to determine the correct match.

However, reports from the field have claimed that when used to monitor large public crowds to locate criminals at large, this technology has not provided optimal results. The problem here is that a person must position his face no more than approximately 35 degrees to the camera in order for a correct identification to be made. Furthermore, a criminal being sought by law enforcement authorities might easily disguise his face using eyeglasses, facial hair or a hat to easily defeat the system. At least one manufacturer of this type of software has claimed that a positive identification could be made if only 15 percent of an individual’s forehead was visible in the original image, which could be the case if the head of one person eclipsed another’s in a captured image from a crowd scene. This has yet to be empirically demonstrated in the field, however.

At present, facial-recognition technology holds much promise for ATM security, check-cashing identity verification and elimination of voter fraud, where local field conditions can be carefully controlled and monitored.