MRSL: AUTONOMOUS NEURAL NETWORK-BASED SELF-STABILIZING SYSTEM

Abstract

Stabilizing and localizing the positioning systems autonomously in the areas without GPS accessibility is a difficult task. In this thesis we describe a methodology called Most Reliable Straight Line (MRSL) for stabilizing and positioning camera-based objects in 3-D space. The camera-captured images are used to identify easy-to-track points “interesting points” and track them on two consecutive images. The distance between each of interesting points on the two consecutive images are compared and one with the maximum length is assigned to MRSL, which is used to indicate the deviation from the original position. To correct this our trained algorithm is deployed to reduce the deviation by issuing relevant commands, this action is repeated until MRSL converges to zero. To test the accuracy and robustness, the algorithm was deployed to control positioning of a Quadcopter. It was demonstrated that the Quadcopter (a) was highly robust to any external forces, (b) can fly even if the Quadcopter experiences loss of engine, (c) can fly smoothly and positions itself on a desired location.