Vision System For Mini Unmanned Aerial Vehicles

Mini UAV(Unmanned Aerial Vehicle) which features low cost and high flexibility,has a broad application in the civil and army field. As an indispensable part of MiniUAV, Vision System has been gaining more and more attention from colleges andresearch organizations around the world. A typical vision system is designed toimplement air-shooting, object recognition, vision navigation and many other tasks.This paper introduced a realization of a vision system for carrying out the landingtarget detection and pose estimation tasks. First, we made an overview of the visionsystem for Mini UAV. Then, we introduced several popular target detection and poseestimation algorithms. Among these, we choose the Hu-invariant moments algorithmas our implementation and described the target detection and pose estimation methodbased on it in detail. On the basis of the above theory, we designed a vision systemwhose main functions including: video capture, video recording, target detection andpose estimation. In the realization of the video capture module, in order to improvethe video quality, we choose the MPEG2 video capture card to replace the traditionalraw capture device and designed a special filter graph building algorithm for decodingthe MPEG2 video stream successfully. In the realization of the image processingmodule, we combined the Microsoft DirectShow with Intel OpenCV & IPL softwaredevelopment kit to implement the processing algorithms. Finally, we presented theprogram real-time running interface and verified the performance of the applicationand the detecting accuracy in outdoor experiments.The vision system described in this paper features DirectShow architecture andHu-invariant moments target detection algorithm which shows high performance inpractice. It provided a good alternative in the vision system design domain.