The invention provides a video
stream-oriented
infrared and visible light image adaptive registration method, and belongs to the technical field of electric
digital data processing. The invention discloses a video
stream-oriented
infrared and visible light image adaptive registration method, which comprises the following steps of: firstly, constructing a scale adaptive
infrared and visible light
static image registration network, and extracting multi-scale image features for scale prediction and
transformation parameter estimation by using a pre-trained weight ResNet101; a two-stage training mode is designed, and registration is performed after scale, so that interference of registration training on
scale estimation is prevented. Aiming at the problem of lack of a multi-
modal registration
data set,
modal migration and synthetic transformation are performed on a visible light
data set, the visible light
data set is divided into a
training set and a
test set, and a
static image registration network is trained. And finally, designing a frame association registration strategy, eliminating inter-frame hopping, and outputting a
flicker-free continuous video. The method solves the problems that in the infrared and visible light
image registration process, it is difficult to deal with too large scale difference of different
modes, and video hopping is caused when a static registration method is directly applied to video streams, and has good universality and practicability.