Vanguard Magazine

s securItY www.vanguardcanada.com DECEMBER 2016/JANUARY 2017 25 face recognition and identification. Most current face recognition systems need an ideal face image directly facing the camera with proper illumination, no aging, no occlusion and so on. In reality, images can be covered by ob- jects (see Figure 1), making recognition difficult. Individuals may also have signifi- cantly altered their appearance, whether by aging or by different hair styles or colors. These impediments to consistent automat- ed recognition are enhanced by a lack of straight angles to the camera. what are the main challenges? The first challenge is the design and devel- opment of efficient algorithms for face rec- ognition that are reasonably fast and yet ro- bust enough against so-called performance degrading factors such as variation in pose, illumination, blur occlusion, and aging. The second challenge is how to imple- ment and realistically prototype practical systems while working with such intelligent algorithms. More importantly, it is crucial to make sure that prototypes will eventually turn into commercial products and will not be stuck in libraries as research projects only. human motion and action analysis Although computerized human face recog- nition is a relatively known concept, a new subject of research has emerged over the past few years called human motion and ac- tion analysis. In its basic form, human action analy- sis can be defined as follows: given video sequences of human motion, e.g. videos captured by surveillance cameras in gov- ernment buildings, military zones, big factories, nuclear plants, shipyards, banks, airports, borders, or hospitals, we are inter- ested in finding the identity of individuals Figure 2 Figure 3 Sample video frames captured from different activities of an individual from the KTH database. From le to right, samples here present walking, jogging, running, clapping, boxing, and waving, respectively. More activities can be defined and processed. Human action recog- nition and motion analysis algorithms could provide human activity information in advance before an incident goes wrong. Schematic of the Red Carpet Kiosk idea, where individuals walk along a pathway for just a few meters before facing an officer. RCK records the individual's body motion and habits, face image in different positions, face image with straight angle to camera, iris images and, finally, fingerprints. based on their style of walking, running, mannerisms, etc. An instant application of a simpler ver- sion of the RCK is to equip border cross- ings, or other checkpoint-style gateways, with surveillance sensors which employ human motion and action recognition sys- tems. For example, if a number of employ- ees regularly pass through the gates of a restricted organization (such as a shipyard or nuclear plant) a database containing motion and face recognition information of those employees can be created. Using this database on a day-to-day ba- sis, the system can then identify individu- als passing through the gate much more efficiently. Additionally, it would be possible to automatically monitor and determine if crossing objects are humans, animals, ve- hicles, etc. This can also help to signifi- cantly reduce the number of suspicious and unexpected activities such as human trafficking, drugs, or illegal vehicles. As the technology continues to be developed, automated human action detection and analysis services are expected to become a capability of future security systems. Proposed complete multi-step system As mentioned above, a complete version of an RCK could be armed with a fingerprint scanner, human face recognition system, iris imaging and identification algorithms, human motion and action analyzer, and could be modified according to customer requirements. A self-operating RCK could provide a more accurate and efficient security pro- cedure, as well as reduce the number of employed officers, which in addition to making a significant improvement to secu- rity procedures, also has economic impli- cations.

• Cover