Design Of FPGA-based Multi-channel Video Stitching Systems

Both with the rapid development of digital video technology and the enhancement of people’s security consciousness, the video processing technology has been widely used in communications, broadcasting, surveillance, consumer electronics and medical fields, and met more challenge meanwhile. For example, the monitoring pictures will never meet the demand which need any perspective of particular scene with the limited hardware devices. As a matter of fact, the solution which combine pictures from multiple cameras to a panoramic or wide viewing angle one is deeply investigated recently, and such a research become a hot spot in digital video technology.We proposed a solution which can achieve the multi-channel HD-SDI video capture and stitching based on the characteristics of high-definition video streams and associated image stitching algorithm. This solution is based on the platform environment of Xilinx’s ZYNQ series SOC processor, the ARM dual-core Cortex-A9 processing unit and the hardware FPGA logic resources. Works and achivememts are presented as follows:(1) The functions of video stitching are analyzed, the FPGA-based video processing technology is presented and the scheme of image stitching algorithm is illustrated.(2) This section designes a multi-channel video capture and stitching hardware prototype system, which has a multi-channel SDI capture device, high definition video input and output interfaces, Gigabit Ethernet port and a TFT screen. The multi-channel video capturing and stitching algorithms are verified on this prototype system.(3) Taking the needs of image stitching into account, this thesis implements multiple HD video streams capture, by using the FPGA hardware resources to achieve the image color adjustment, video scaling, video filtering and other functions, and then the video data is transmitted into the DDR through the VDMA for ARM core processing.(4) This section presents a stitching algorithm that based on Harris corner, firstly extracts the original image Harris corner feature value, then calculates homography matrix, and finally gets a complete picture of results through the projection transformation and fusion.(5) This section summarized the full content of the paper, meanwhile, made some prospect of further research.