Enhancing Position Estimation in a Small Quadrotor via Onboard Real-Time Depth Feature Matching

This paper presents a position estimation improvement technique in a small quadrotor employing depth feature matching, focusing on real-time onboard processing implementation. Small quadrotors have been gaining significant popularity due to their agility, versatility, safety, and ability to operate in confined spaces. However, their limited payload capacity and computational resources present challenges that restrict their full potential. In the context of robotics, odometry is crucial to provide critical information about their position and movement. This is especially important in indoor environments where global navigation satellite system (GNSS) signals are unavailable. Unfortunately, most position estimation systems either have low accuracy or require high computational costs. Therefore, we propose an onboard, real-time depth feature matching method to improve odometry accuracy in a small quadrotor. It utilizes plane features extracted from topological maps with bundle adjustment (BA) as an optimization technique to refine those planes across frames. Experimental results demonstrate that odometry drift can be reduced effectively under the system constraints.