A.G. Tashlinskii - Dr. Sci. (Eng.), professor, head of Radio engineering department, Ulyanovsk State Technical University. E-mail: tag@ulstu.ru
P.V. Smirnov - Post-graduate student, Ulyanovsk State Technical University. E-mail: rtcis@mail.ru

Images of moving object are often distorted due to its high speed. The most significant type of distortion is blur. Formulation of image restoration problem is the following. Video sequence from a camera contains images of a moving object which are noisy and blurred. The camera may also move. Using video sequence we need to find moving object, deblur its image, estimate its motion pattern, and filter additive noise. The paper describes main stages of the task: moving object region detection using frame sequence, blurring parameters estimation, object image restoration, estimation of location parameters of moving object (or its defined part), restored image registration to increase signal-to-noise ratio. New algorithm for motion region detection based on stochastic gradient procedure is proposed. The algorithm is resistant to noise and allows eliminating influence of global interframe geometric deformations and motion of small objects. The algorithm estimates blurring parameters using relation between them and cepstrum minimum location as well as property of cepstrum addition. We chose iterative algorithm for image restoration due to its ability to reduce noise and edge effects. Adaptive stochastic gradient procedures without identification are used to estimate object location parameters. The algorithm does not require a priori information of blurring parameters and can be fully automatized. Its main advantage is high processing speed due to joint use of detection, registration and restoration recurrent procedures.