Embedded video stabilization system on field programmable gate array for unmanned aerial vehicle
Unmanned Aerial Vehicles (UAVs) equipped with lightweight and low-cost cameras have grown in popularity and enable new applications of UAV technology. However, the video retrieved from small size UAVs is normally in low-quality due to high frequency jitter. This thesis presents the development of vi...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 语言: | English |
| 出版: |
2017
|
| 主题: | |
| 在线阅读: | http://eprints.utm.my/id/eprint/86019/1/MohdFarizalFirdausMFKE2017.pdf |
| 标签: |
添加标签
没有标签, 成为第一个标记此记录!
|
| 总结: | Unmanned Aerial Vehicles (UAVs) equipped with lightweight and low-cost cameras have grown in popularity and enable new applications of UAV technology. However, the video retrieved from small size UAVs is normally in low-quality due to high frequency jitter. This thesis presents the development of video stabilization algorithm implemented on Field Programmable Gate Array (FPGA). The video stabilization algorithm consists of three main processes, which are motion estimation, motion stabilization and motion compensation to minimize the jitter. Motion estimation involves block matching and Random Sample Consensus (RANSAC) to estimate the affine matrix that defines the motion perspective between two consecutive frames. Then, parameter extraction, motion smoothing and motion vector correction, which are parts of the motion stabilization, are tasked in removing unwanted camera movement. Finally, motion compensation stabilizes two consecutive frames based on filtered motion vectors. In order to facilitate the ground station mobility, this algorithm needs to be processed onboard the UAV in real-time. The nature of parallelization of video stabilization processing is suitable to be utilized by using FPGA in order to achieve real-time capability. The implementation of this system is on Altera DE2-115 FPGA board. Full hardware dedicated cores without Nios II processor are designed in stream-oriented architecture to accelerate the computation. Furthermore, a parallelized architecture consisting of block matching and highly parameterizable RANSAC processor modules show that the proposed system is able to achieve up to 30 frames per second processing and a good stabilization improvement up to 1.78 Interframe Transformation Fidelity value. Hence, it is concluded that the proposed system is suitable for real-time video stabilization for UAV application. |
|---|