Prototyping a mobile user interface for viewing footage from a panoramic camera recording system

• Main Author: Lim, Kian Hwee
• Other Authors: Yong Chuen-Tze, Mark
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/59900
• Institution: Nanyang Technological University

Increasing number of vehicles is being fitted with cameras to provide a video log that may be used in case of accidents on the road (and while parked and stationary). These cameras typically capture only one view with limited field of view. The project is to create a prototype system that allow multiple cameras to be connected to the system and combine their output into a single recording and display this recording on a mobile device for the ease of viewing. Researching on the previous work on panoramic images have shown two approach to constructing panoramic images, first is to use the curved mirror approach, another is to use the image stitching approach. To achieve automatic stitching we need to use features in overlapping frames to compute the relationship between them, but when the author experimented with series of camera nodes placed in different direction, introducing parallax and errors in the automatic stitching process. Therefore the author has to constraint the position of the camera nodes. The camera nodes have to be placed such that they share the same axis of projection because there is no parallax when camera is rotated about the projection center. The stitching process requires the stitcher to recognize the overlapping relationship and use that relationship to compose the panorama image and it usually takes longer than 10 microseconds to finish the construction of panorama those posing a problem with loss of frames when waiting for the stitcher. The author designs the architecture such that the calculation of overlapping relationship only happens once per video instead of once every sequence of frames. A buffering system was also introduced to ensure that no frames are lost when stitcher cannot perform faster that the frame rate of the camera nodes.