The object recognition and adaptive threshold selection in the vision system for landing an Unmanned Aerial Vehicle

We present the design and implementation of a vision system for landing an Unmanned Aerial Vehicle (UAV). This vision system consists of the vision detection software and the self-made onboard camera platform. After accomplishing its mission, the UAV would return to the helipad and land on it autonomously and accurately. To achieve more, the head of the UAV must point to the direction which the helipad indicates. The object recognition, the main part of our vision algorithm, includes two parts: the recognition of the identifier ldquoHrdquo which indicates the landing point, and the detection of the triangle which indicates the landing direction. To detect the ldquoHrdquo, we designed a new algorithm based on image registration which turn out to be a fast, robust and computational inexpensive method. And to detect the triangle to confirm the landing direction, we used a method based on Hough Line Detection and Helen formula, which is also very fast and accurate. As a breakthrough of our work, we firstly used an adaptive threshold selection method in such a system to make our algorithm more robust. According to the results of our flight trials, our vision system well performed in this application: the average processing time of a 640times480 image is less than 60 ms, and the average rate of successful target detection is 97.42%. In the 2008 Chinese Robotics Competition, we won the Gold Medal of the aerial robotics group.