Camera motion classification using a genetic functional-link neural network

Abstract

In this paper camera motion classification for compressed videos using a genetic functional-link network (GFLN) is proposed. GFLN is a feedforward functional-link network (FLN) and Gaussian functions are used in the functional nodes. The parameters in GFLN are adjusted using genetic evolutionary approach. GFLN provides feature selection capability by selecting the links between input layer and functional nodes dynamically. Genetic coding is used for combining evolution of weights and Gaussian parameters in a single chromosome. Seven categories of camera motion: static, pan-right, pan-left, tilt-up, tilt-down, zoom-in, and zoom-out decoded from the MPEG-I video stream are used for neural classification. Our aim is to rapidly extract and process motion vector information from MPEG video without full frame decompression. Video streams with aforementioned classes of camera motion have been successfully classified.