FPGA-based visual navigation sensor for spacecraft docking

With the onset of autonomous systems, the use of camera based visual navigation systems for relative position and attitude estimation is a growing trend. Scenarios involving high-speed decision making, like spacecraft docking, calls for the need of a pose estimation algorithm capable of running at high refresh rates.This paper showcases realization of such a high refresh rate visual navigation sensor on an FPGA based system on chip, tuned to detect 12-point LED tags. Target/Object detection is the bottleneck for vision-based techniques. The high parallel processing power of the FPGA is used to meet the high-speed requirement. The entire pose estimation chain was implemented in the SmartFusion2 system on chip. The setup consists of a camera with HDMI interface for sensing, detection and centroid calculation of objects using blob detection algorithms implemented in the FPGA.The sensor was able to process a 1280 by 720 pixel resolution, 30 FPS video stream. Following frame grabbing, object detection provided sub-pixel locations of the 12 points of the LED tag within the captured image. Relative pose estimation was then performed for the frame. Each frame was processed well within the a time of 16 milliseconds.