Drone Autonomous Navigation by Hardware Image Processing

Abstract

Our approach for autonomous navigation is to apply image processing for estimating the drone position. Two techniques are employed: visual odometry, and computer vision. Edge detection is one important step for computer vision, and it is performed by neural network implemented on FPGA. After image segmentation, a correlation between satellite image - or reference image - and image obtained by the drone is computed. Finally, the positioning by visual odometry and computer vision are combined using a new formulation of particle filter. Our results show better results in comparison with other edge identification procedures, implying in a more precise trajectory correction.