Skip to content

Latest commit

 

History

History
49 lines (32 loc) · 2.3 KB

README.md

File metadata and controls

49 lines (32 loc) · 2.3 KB

Underwater-Robotics

Development of an underwater simulator which will be used for oysters detection for the project. Simulator can generate random underwater landscape, randomize oysters location and oysters count and much more. Using a range scanner installed on BlueROV for navigation.

Preview of underwater scene

Rover has 2 cameras. 1 facing front at an angle of 25 degrees from horizontal and 1 facing the seabed Sped 4x

front_bottom.mp4

VSLAM result - Sped 32X

vslam_result.mp4

Oyster map generation

heat_map.mp4

Tasks

  • 2D bounding Box of objects from Blender 2.93
  • Integrate IMU with blender
  • Integrate LiDAR/SONAR with blender
  • Train yolo on the generated data from blender
  • Train network for semantic segmentation task
  • Train network for depth estimation task
  • Train GAN to get realistic underwater images from renderd images
  • Train multi-task learning network to predict segmentation, 3D depth estimation, and realistic underwater images in a single forward pass

Google Colab Notebook

  • colab notebook used to train the yolov4-tiny, find it here
  • Modified the colab notebook provided here

Models

  • We trained a yoloV4-tiny on a dataset of around 5000 images
  • Download the model best weights file from here
  • Copy the model weights in here

Blender model

  • BlueROV model downlaod from here
  • Oysters model download from here