实时视场拼接系统的设计与实现

Design of real-time video mosaic system

当要求的场景尺寸超出一个光学传感器的范围时,同时取得完整的场景就成为一个难点。针对这一问题,采用多个光学传感器同时对场景进行采集,可以得到几幅互相有一定重叠的场景图像。应用改进的相位相关算法对重叠图像进行快速配准,应用渐入渐出融合算法消除拼缝,实现无缝大视场拼接。并将整套算法在以TMS320DM642为核心处理器的平台上实现,得以构成一个小型化视频拼接系统。实验结果表明,该系统可以自动地对存在一定重叠和旋转的两幅768×494分辨率、25帧/秒的视频图像进行拼接,获得无缝、清晰的大视场视频图像,满足系统实时性的要求。

When the size of a needing scene was beyond the scope of an optical sensor, it was difficult to acquire the whole scene at the same time. In order to solve this problem, several optical sensors were adopted to capture the scene simultaneously, and several overlapped images were acquired. The overlapped images were rapidly registered by using the improved phase correlation image alignment method, and the gradated in-and-out amalgamation algorithm was applied to eliminate the seam, and a seamless and wide field of view image was acquired. A minitype video mosaics system based on TMS320DM642 was developed and the whole algorithm was implemented on this system. The test results show that the system can automatically stitch the two overlapped images with the resolution of 768×494 and the speed of 25FPS, and the real-time property of system is satisfied.