DESIGN AND REAL TIME IMPLEMENTATION OF STITCHING ALGORITHM FOR PANORAMIC SCANNING

Abstract

Panoramic Images are wide angle view and wide aspect ratio images. Wide aspect ratio images mean the images with width greater than that of their height. These Panoramic images are formed using a stitching process in which multiple photographic images with overlapping field of view are aligned and combine together in a seamless manner. These high resolution images are widely used in medical and biological fields for clinical studies, Because of their high resolution and wide angle view panoramic images are very suitable for surveillance and military application. For joining these images first we have to align them and for aligning images to same frame of reference different mathematical models are used to map pixel coordinates from one image to another, parameters for these transformations are estimated through different techniques generally categorized as direct pixel base methods and feature base methods. In direct pixel base methods pixel to pixel base dissimilarities are minimized using error matrices and these techniques include difference base methods and correlation methods while on the other hand in feature base methods prominent features of images are detected matched and optimization techniques are used to estimate the parameters of transformation. We have designed an algorithm that will used physical camera orientation in azimuth and elevation to estimate the parameters of translational transformation to align and stitch the images together. We have used encoders for position estimation in azimuth and elevation these encoders have 20000 encoder counts per 360 degree which are then change in to pixels using scales that are found after camera calibration, these camera positions are sampled every 20ms. This technique is implemented for Real time Panoramic scanning Surveillance application on FPGA Spartan 6. For achieving real time performance an advance memory management scheme is designed and implemented on in built 128 MB ddr2 memory. The camera used is a dual field of view long range thermal camera capturing 50 frames per second with a resolution of 320 x 256. There are two surveillance modes sector scan mode and 360 degree scanning mode and to display it on a HDMI LCD with limited resolution of 1024 x 768 a display scheme is designed.