Design and Implementation of an Electronic Image Stabilization System Based on Multi-source Data Fusion

Abstract

With increasingly stringent real-time and stability requirements for high-definition video surveillance in industrial settings such as oil and gas fields, conventional electronic image stabilization (EIS) systems that rely solely on image processing lose accuracy under severe vibration because feature points disappear, and PC-based solutions suffer from high power consumption and latency. This paper introduces an FPGA-based gyro-vision fusion stabilization system. A three-axis gyroscope first supplies prior information on rapid camera attitude changes. On the vision side, FAST corner detection combined with Lucas–Kanade optical flow ensures reliable feature tracking even under motion blur. Inertial and visual displacements are then dynamically fused via a Kalman filter to build a highly robust global motion model. The design implements preprocessing, feature extraction, data fusion, and motion compensation as parallel pipeline modules on a Xilinx Artix-7 FPGA. Hardware-level parallelism and fixed-point optimization allow steady 30 fps real-time processing of 720p video, raising peak signal-to-noise ratio (PSNR) by an average of 15.83 %, while power consumption is only 12.9 % of a traditional CPU solution. Experiments confirm that the proposed system delivers high accuracy, low latency, and low power, offering a practical hardware-accelerated path for video stabilization in demanding industrial environments.