Real time mosaicing and change detection system

Abstract

A system capable of demonstrating surveillance of a vast area for target tracking in real time has been a challenging problem and in this paper we present architecture for a real time, mosaicing and change detection system which proves to be an efficient solution for the same. A method of pipelining SURF (Speeded Up Robust Features) has been proposed for performing real time registration to generate panoramic view of a very large scene at far away distances. The pipelined architecture of the proposed system performs the computationally intensive SURF registration over 6 times faster than conventional SURF without any significant compromise in accuracy. The atmospheric turbulence may restrict accurate registration of images therefore a solution has been proposed which performs atmospheric turbulence restoration. Spatial information based object detection and tracking of slow moving objects in the mosaics of large areas is performed which ensures robust and accurate detection of changes in the scene. The proposed system works for large range PTZ cameras with both TI and CCD sensors in real world conditions, with fast panning and vast area of monitoring giving high target detection accuracy. The experimentation has been performed on varied datasets of both CCD and TI sensors and in different environmental conditions for both day and night scenarios. It has been observed from the experimental result that the proposed system outperforms similar systems such as [1, 13, 10] in terms of robustness, ability to handle diverse environments and overall system performance.