Studies On Key Technologies For Landborne Panoramic Image Acquisition

This paper studies several key technologies involved in panoramic image acquisition on a vehicular platform. Panoramic images of streets have gradually being used in pedestrian navigation and vehicle navigation, e.g., Google map, Map bar, and City bar. Exactly, the panoramic images have been an important source data for the development of digital city. This paper presents researches and solutions to the high-precision temporal positioning, the luminance balancing of a sequence of panoramic images, the efficient stitching of panoramic images. This research would make a great sense to the development of the urban information platform, the high-quality acquisition and management of road data, and the humanity-nice LBS (Location-based Service).High-precision spatial-temporal positioning for panoramic images. The system hardware includes4high resolution color CCD cameras, an industrial lens, a GPS/DR, a synchronizing controller, and a CPCI (compact PCI) computer, etc. The system temporal benchmark is based on combination of the GPS and the High stability crystal oscillator. The system spatial benchmark is based on the combination of the POS (positioning and orientation system) and the DMI (distance measurement instrument). Through the matching from time serials to the spatial serials, a unified spatial-temporal benchmark can be constituted. Therefore, it is possible to retrieve the time stamp or the event position of the panoramic image when given one another.Luminance balancing of panoramic images. Based on the traditional dodging algorithms for a single image, and for multiple images, an illumination balancing algorithm is proposed to handle a sequence of panoramic images. This method takes into account the spatial and temporal continuity of panoramic images. In the first step, each image collected by the camera is processed to gain a uniform luminance (i.e., single image dodging). In the second step, all the image collected by the camera array in the same frame (each frame containing four captured image by the camera array) are processed to achieve a uniform luminance (i.e., multiple images dodging). At last, the luminance between different frames of panoramic images is processed (sequential panoramic image dodging). Based on this luminance balancing algorithm, both global and local luminance uniformity can be obtained. Efficient stitching of panoramic images. At first, a camera-model based stitching method is constructed. The parameters of the model can be found by a calibration process in the calibration field, with which a coarse stitching of the panoramic images can be achieved. Then, based on the coarse stitching results, overlap regions of each pair of adjacent images can be located. SIFT feature point matching and image stitching can be processed within the scope of the overlap regions, and thus it makes a new region-constrained SIFT matching based image stitching method. This method makes full use of the calibration results of the camera model, which can greatly narrow the searching area for the SIFT feature matching, thus improves the efficiency of panoramic image stitching.