Analysis Of Images And Videos Based On PAL Camera

Panoramic Annular Lens (PAL) can capture its surroundings with a field of view which is 360-degrees horizontally and 10 to 60 degrees vertically without any mechanical scanning parts. Due to its feature of possessing large field of view, PAL camera plays an important role in robot navigation, medical imaging, virtual reality and missile guidance. However, images formed by PAL camera do not get agreed with the visual custom of human being. There is severe distortion in these images. This dissertation focuses on the applications and analysis of images and video based on PAL camera and is to propose several image processing algorithms, such as, the method of super resolution of PAL images and the method of its evaluation, algorithm of object tracking in PAL video sequences and algorithm of pedestrian detection in PAL images.Chapter 1 introduces the research background of the panoramic imaging, analyses the merit and fault of the fish lens and the conventional camera based on Central Convergence Perspective(CCP). Then the structure of panoramic annular lens based on Flat Cylinder Perspective(FCP) is analyzed. At last the context of the dissertation is introduced.Chapter 2 analyzes the geometric structure of PAL used in this thesis. The relationship between its object and image is represented in analytical method. According to the experiment, the images are represented. The result reveals that the images formed by PAL are destroyed by severe distortion.Chapter 3 discusses the linearization of PAL images firstly. During the procedure of unwrapping the PAL images, the problem of un-uniform resolution occurs. Based on this, the optimization cubic kernel interpolating, Bezier curve interpolating and Ferguson curve interpolating are cited to unwrap the PAL images. Then a method which uses MSE, SSIM and S-CIELAB indexes to evaluate the performances of the unwrapping algorithm is proposed.Chapter 4 solves the problems of object tracking and pedestrian detection based on PAL camera. In this chapter, the method of object tracking based on Mean Shift and the algorithm of pedestrian detection based on Histograms of Oriented Gradients (HOG) feature are used, the improvement is made for the characteristicsChapter 5 summarizes all the algorithms of the full thesis and proposes the future plan.