MAVNet: An Effective Semantic Segmentation Micro-Network for MAV-based Tasks
Published in IEEE Robotics and Automation Letters (RA-L), 2019
This paper introduces MAVNet, a compact and efficient neural network for real-time semantic segmentation on resource-constrained micro aerial vehicles (MAVs). Designed with size, weight, and power (SWaP) constraints in mind, MAVNet uses a small number of parameters while maintaining performance comparable to heavier models in empirical tasks.
The system is optimized for on-board operation using platforms like the Jetson TX2, and is validated on inspection tasks such as infrastructure mapping in dam penstocks. The authors also release new datasets and benchmarks to support research in aerial semantic segmentation and perception. MAVNet supports real-time perception for navigation, target detection, and mission adaptation based on visual semantics.
Recommended citation: Nguyen, T., Shivakumar, S. S., Miller, I. D., Keller, J., Lee, E. S., Zhou, A., Özaslan, T., Loianno, G., Harwood, J. H., Wozencraft, J. M., Taylor, C. J., & Kumar, V. (2019). "MAVNet: An Effective Semantic Segmentation Micro-Network for MAV-based Tasks." IEEE Robotics and Automation Letters, 4(4), 3908–3915.
Download Paper
Related people: Tolga Ozaslan