Research Of Multiple View Panoramic Video Technology Based On Webcams

As a rising technology, Panorama can create wide field-of-view and provide more abundant information about the scene. This technology has various application prospects in the areas of scientific research, commerce, and art, such as the satellitic photography, video conference, monitoring, traffic guiding, virtual reality, photography, to name only a few.Currently, the static panoramic images are created by image stitching and blending algorithms, except for some special cases in which the super wide angle lens or fish-eye lens are used for capturing the panoramic images directly. As for the dynamic video, the special hardware systems are used. One solution is the fast ball system, in which the fast ball rotates in high speed in order to capture the surrounding scene. However, due to the limitation on the rotation speed, it is inevitable to lose some scenen in this kind of system. Another solution is to use special capturing system with panoramic lens. But this type of system is too expensive and complicated to be applied.This paper presents a technique to create panoramic video from common webcams. This solution is based on the analysis of both hardware and software panorama techniques, and it is feasibility and applicability has been proved by the experiments. First of all, we investigate into the existing panoramic video technology, and introduce the basic frame of our system. Secondly, we discuss the algorithms of main modules including the feature extracting module, the feature matching module, and the projection and blending module in detail. Thirdly, we make use of the GPGPU (General-Purpose computation on GPUs) technology to optimize our system. Lastly, we test the system and make an detailed analysis and discussion.We have implemented the entire panorama video system. The main contributions and features are listed as follows:1. Capturing the video by common webcams so the system does not need any special lens or hardware, which means that the system is inexpensive and convenient. Besides, the system is constructed by modules, so it is easily expandable.2. Designing a feature extraction and matching modules based on SIFT algorithm. These two modules enable the system to work under challenging conditions such as the translation, big rotation and so on. So the angles and directions of webcams can be adjusted to some extent.3. Desiging two types of blending. The first one is the classical bilinear blending algorithm, which work well to blend common scenes. The other one is the nonlinear blending algorithm proposed by us. This algorithm can reduce the ghosting artifcats when the parallax is strong.4. Researching the GPGPU technology, and transplant the most computational-expensive module, the feature extraction module, on to the GPU. This optimization speed up the whole system greatly.