Hardware implementation of the kalman filter for video signal processing

Abstract

Probabilistic methods for detecting a moving object are widely used in solving problems of digital processing of a video signal in computer vision systems. At the same time, the primary tasks remain to increase the quantitative and qualitative characteristics of the digital processing of video information. This article discusses a probabilistic method for detecting a moving object in a video data stream using the Kalman filter as an example. A scheme for detecting a moving object based on the Kalman filter is proposed. The architectures of calculators for predicting the system state and covariance errors are built. Hardware simulations have shown the ability to reduce the computation time of the system state and covariance error by 5.4% when using high-speed parallel Carry-Save Adders (CSA) and Kogge-Stone prefix adders (KSA) compared to an architecture based on built-in addition and multiplication operations. Software modeling made it possible to implement the proposed algorithm for detecting a moving object in a video data stream under affine transformations.