RTSDM: A Real-Time Semantic Dense Mapping System for UAVs

Intelligent drones or flying robots play a significant role in serving our society applications such as rescue, inspection, agriculture, etc. Understanding the scene of surroundings is an essential capability for further autonomous tasks. Intuitively, knowing self-location UAV and creating semantic 3D map fully However, integrating simultaneous localization, reconstruction, segmentation together huge challenge power-limited systems UAVs. To address this, we propose real-time mapping system that can help to understand its location surroundings. The proposed approach includes modified visual SLAM with direct method accelerate computationally intensive feature matching process module at back end. runs lightweight network, BiSeNetV2, performs only key frames from front-end task. Considering fast navigation on-board memory resources, provide dense-map-building generate OctoMap segmented map. verified experiments on platform Jetson TX2 computation unit. A frame rate around 12 Hz, accuracy 89% demonstrates efficient while providing sufficient information tasks