AI Summary & Key Takeaways
AI InsightBesides the performance of these methods, energy consumption and resource usage are evaluated allowing the selection of the appropriate SLAM method.
📌 3 Key Takeaways
Abstract
It is important to solve the autonomous mapping problem with high accuracy using limited energy resources in an environment without prior knowledge and/or signal. Visual Simultaneous Localization and Mapping (SLAM) deals with the problem of determining the position and orientation of an autonomous vehicle or robot with various on-board sensors, and simultaneously creating a map of environment with low energy consumption. However visual SLAM methods require high processing performance for real-time operations. Also, processing capability of the hardware is limited by the power constraints. Therefore, it is necessary to compare the processing load and power consumption of visual SLAM methods for autonomous vehicles or robots. For visual SLAM methods, although there are different comparison studies, there is no comprehensive computational cost analysis covering different datasets and important parameters including absolute trajectory error, RAM Usage, CPU load, GPU load, with total power consumption. In this paper, ORB-SLAM2, Direct Sparse Odometry (DSO), and DSO with Loop Closure (LDSO), which are state of the art visual SLAM methods, are compared. Besides the performance of these methods, energy consumption and resource usage are evaluated allowing the selection of the appropriate SLAM method.