Comparative analysis of monocular slam algorithms using tum and euroc benchmarks

Abstract

Stable and robust path planning and movement in ground mobile robots require a combination of accuracy and low latency in their state estimation. However, state estimation algorithms must provide these qualities under the computational and power constraints of embedded hardware. Simultaneous localization and mapping (SLAM) algorithms are the best choices for state estimation in these scenarios, in addition to their ability to operate without external localization from motion capture or global positioning systems. Moreover, a single-camera setup is the most common solution for robotic platforms, which reduces our domain of interest to the specific SLAM algorithms type—monocular SLAM. Yet, it is still not clear from the existing literature, which monocular SLAM algorithms perform well under the accuracy, latency, and computational constraints of a ground mobile robot with onboard state estimation. This paper evaluates an array of the most recent publicly available monocular SLAM methods: ORB-SLAM2, DSO, and LDSO. The evaluation considers the pose estimation accuracy (alignment error, absolute trajectory error, and relative pose error) while processing the TUM Mono and EuRoC datasets on the specific hardware platform with a balanced amount of computational resources and power consumption. We present our complete results as a benchmark for the research community.