Image registration in low resolution visual sensor networks

Abstract

Image registration is the process of integrating data from different coordinate systems into one coordinate system. It is of great interest in video surveillance because of its capability to combine images and generate a larger view of the area under observation, while retaining all the information in the images. This thesis proposes a novel method for registering images obtained from low resolution visual sensor networks by using change detection as a tool for image registration. Two sensors with overlapping fields of view are used to capture images at regular intervals. The images differences are found and significant change is identified using a random threshold. This significant change forms the basis of identifying control points. Once the control points are identified then, the reference image is transformed with respect to the base image using affine transformation. The transformed image and base image are stitched together to obtain the registered image. Experiments using the proposed scheme have shown that precise registration can be achieved.