Researchers from the University of Bologna and ETH Zurich, including Jonas Kühne, Christian Vogt, Michele Magno, and Luca Benini, have developed a new system designed to enable accurate motion tracking on resource-constrained devices. Their work, titled “LEVIO: Lightweight Embedded Visual Inertial Odometry for Resource-Constrained Devices,” was published in the journal IEEE Transactions on Mobile Computing.
The team’s research focuses on creating a visual-inertial odometry (VIO) system that can operate efficiently on low-power hardware, such as micro-drones and smart glasses. VIO systems combine data from cameras and inertial measurement units (IMUs) to track the position and orientation of a device in real-time, without the need for external infrastructure. However, current state-of-the-art VIO systems often require significant computational resources, making them unsuitable for devices with limited processing power and energy capacity.
To address this challenge, the researchers developed LEVIO, a VIO pipeline optimized for ultra-low-power compute platforms. LEVIO incorporates established VIO components, such as ORB feature tracking and bundle adjustment, but emphasizes a computationally efficient architecture with parallelization and low memory usage. The team employed a hardware-software co-optimization approach to ensure that LEVIO could run in real-time on resource-constrained hardware.
The researchers validated LEVIO on a parallel-processing ultra-low-power RISC-V system-on-chip (SoC), achieving 20 frames per second (FPS) while consuming less than 100 milliwatts (mW) of power. They also benchmarked LEVIO against public VIO datasets, demonstrating that it offers a compelling balance between efficiency and accuracy. To encourage adoption and further research, the complete implementation of LEVIO has been released as open-source.
In the energy sector, LEVIO’s efficient motion tracking capabilities could be applied to various applications, such as monitoring and inspecting energy infrastructure, including power lines, wind turbines, and pipelines. Micro-drones equipped with LEVIO could perform these tasks more efficiently and with lower energy consumption, reducing operational costs and environmental impact. Additionally, LEVIO’s technology could enhance augmented reality (AR) applications in the energy sector, providing workers with real-time, hands-free access to critical information and improving safety and productivity.
The research was published in the journal IEEE Transactions on Mobile Computing, and the open-source implementation of LEVIO is available for further exploration and development.
This article is based on research available at arXiv.