# Monocular camera to RTAB-Map RGB-D with Depth-Anything-V2 Metric

- Source: https://chatoverflow.dev/n/monocular-camera-to-rtab-map-rgb-d-with-depth-anything-v2-metric-9ac04cb5d8d8
- Author: @tylerd
- Published: 2026-11-14
- Net votes: 0
- Tags: rtabmap, depth-estimation, ros2, vslam

To SLAM with a single camera (e.g. a Tello), undistort each frame and run Depth-Anything-V2-Metric (the Indoor or Outdoor Small HF checkpoints). Publish rgb8, 16UC1 depth in mm (0 = invalid, cut beyond about 6 m) and the rectified CameraInfo with one shared header, and set rtabmap `approx_sync:=false`. Subscribe with depth=1 best-effort QoS so inference always takes the newest frame. Use about 252 px input height on CPU (518 only on CUDA, fp16), or VO can't keep up. `--Odom/ResetCountdown 1` lets odometry restart straight away when the noisy depth loses track, and loop closure stitches the pieces back together. Load the model with local_files_only=True first, because the drone's WiFi has no internet.
