Research And Implementation Of Automatic Panoramic Image Stitching Technology Based On ARM

Panoramic image stitching is a technology that allows multiple overlapping images with narrow field of vision stitched into an image with broader vision and high-resolution. This technology can solve the problem that single camera with narrow field of vision cannot get high resolution images. With the panoramic image stitching technology, people can get broader vision and higher resolution even with limited technical conditions.Currently, there are a number of special image processing equipment on the market can also achieve the function of image stitching, such as some mobile phones, digital cameras, etc. These devices usually use a dedicated image processor. Although, they have very powerful image processing capabilities, but they cannot be used widely and have very limited extensions. The integrated chips cannot meet the needs of diverse applications. With the rapid development of ARM technology, ARM has a more powerful function, some ARM has integrated GPU module which can be used to implement complex image processing functions. Some ARM processors can also meet image processing requirements; they can meet some requirements of special applications through their rich peripheral resources and mature manufacturing process.The main research and achievements of this thesis present as follows:(1) This thesis studies the concept of panoramic image stitching and provides an overview of the basic flow about panoramic image stitching. What’s more, panoramic image stitching process flow based on the ARM platform is introduced; it presents the detail of panoramic image from capture to output.(2) In order to obtain the feature extraction algorithm which is more suitable for ARM platform; this thesis analysis several efficient feature extraction algorithm, the analysis includes the speed of the algorithm, number of valid feature points and robustness of the algorithm. By comparison and analysis, we identified the feature extraction algorithm which is more suitable for the ARM platform.(3) For the problem that the stitched image may appear bent when multiple images stitched together. According to the relevant information in the process of shooting, this thesis proposes a method for solving the focal length based on the rotation of the image sequence. In this method the model of camera parameters is simplified, and then a mathematical model is established according to the condition that image rotation angle is already known in the process of shooting, so we can quickly estimate the focal length of the camera. Then use the bundle adjustment method to obtain more accurate results. With this method, the number of iteration is reduced; the real-time performance of the algorithm is improved. The method also lays the foundation for the back projection transformation.(4) The software design framework of image stitching system based on ARM platform is built and automatic stitching of panoramic images is realized. This work provides a reference for image stitching in the ARM platform.