N.S. Bellyustin, Yu.D. Kalafati, A.V. Koval’chuk, A.A. Tel’nykh, O.V. Shemagina, V.G. Yakhno
We describe neuron-like coding algorithms which make it possible to solve both the problem of video-stream detection of an arbitrary object against a complex background and the problem of its subsequent tracking.
The possibility of using a uniform two-dimensional neuron-like medium for location of objects against a complex background is shown. The principles of construction of a decision function which permits taking a decision on whether the desired object is present in the region are considered. An algorithm for the tracking system which allows tracking an object even with drastic variations in its characteristics (face tracking with arbitrary face position) is proposed. The possibility of increasing the information capacity of the video surveillance system due to merging of the video sequences is also demonstrated.
The solution of these problems makes it possible to create an advanced video surveillance system which is able not only to detect and track objects, but also analyze the obtained video information and make up a generalized informative report on observed events.