Vision-based autonomous landing of unmanned aerial vehicles

We present our design of a real-time vision-based landing pad detection and pose estimation for many Unmanned Aerial Vehicles(UAV) and implementation on Raspberry Pi. We describe the vision algorithm for precise landing pad detection and recognition and estimate the position and orientation of the UAV relative to the landing pad. The vision algorithm is robust, accurate, and computationally inexpensive. As a computer board which is easy to embed in the moving body, Raspberry Pi is applied to run the vision algorithm and then outputs the position and orientation information through the serial port. We present results from a simulated landing flight test on Raspberry Pi which shows the vision-based detection is accurate to 98.5% and the state estimates are fast at a frame rate of approximately 13Hz.