Real-time video stabilization on an FPGA
Toru Yabuki, Yoshiki Yamaguchi · 2013
For video processing systems with real-time requirements, FPGAs are used as embedded accelerators. The Systems with FPGAs (i.e., FPGA-based System-on-Chips) are composed of many building blocks such as video compression, target tracking, and object recognition. Here, this paper proposes a video-stabilization implementation for these SoC systems. FPGAs will require video-stabilization functions to apply itself over large areas such as car and aerospace applications. Although the function is achieved by acceleration sensors, its performance relies on the accuracy of acceleration sensors. The increase of component is also not good for highly reliable Applications, which brings the possibility of the growth of the failure rate. Thus, this paper presents video-stabilization module without any sensors. The FPGA system captures 24-bit colour, 542-by-496-pixel-resolution images at 30 frames per second and removes screen movement for the video output within real time. It was successfully implemented and evaluated by the use of Xilinx XUPV5-LX110T Evaluation Platform. The system performance is limited by the specification of an applied image sensor but not limited by that of FPGAs.